From cc53e255ffcc057008d1ec72fda7d710d5457593 Mon Sep 17 00:00:00 2001 From: Daniel Date: Fri, 3 Nov 2017 22:28:00 +0000 Subject: [PATCH] Merge PR #681 to update Manubot This build is based on https://github.com/greenelab/deep-review/commit/8eb858a277c7e31b6d0db5cfb10ebf7ebab59fe1. This commit was created by the following Travis CI build and job: https://travis-ci.org/greenelab/deep-review/builds/297004423 https://travis-ci.org/greenelab/deep-review/jobs/297004424 [ci skip] The full commit message that triggered this build is copied below: Merge PR #681 to update Manubot Update the Deep Review with latest Manubot --- README.md | 19 +- bibliography.bib | 3010 - citations.json | 46092 --------------- processed-citations.tsv => citations.tsv | 500 +- manuscript.html | 2510 + all-sections.md => manuscript.md | 538 +- manuscript.pdf | Bin 0 -> 706804 bytes bibliography.json => references.json | 62254 +++++++++++---------- stats.json | 248 - variables.json | 220 + 10 files changed, 34653 insertions(+), 80738 deletions(-) delete mode 100644 bibliography.bib delete mode 100644 citations.json rename processed-citations.tsv => citations.tsv (98%) create mode 100644 manuscript.html rename all-sections.md => manuscript.md (91%) create mode 100644 manuscript.pdf rename bibliography.json => references.json (83%) delete mode 100644 stats.json create mode 100644 variables.json diff --git a/README.md b/README.md index 1dfbb743..813943e0 100644 --- a/README.md +++ b/README.md @@ -1,17 +1,16 @@ -# Generated citation / reference files +# Output directory containing the formatted manuscript -The [`references`](https://github.com/greenelab/deep-review/tree/references) branch contains files automatically generated by the manuscript build process. -It consists of the contents of the [`references/generated`](https://github.com/greenelab/deep-review/tree/master/references/generated) directory of the `master` branch. -These files are not tracked in `master`, but instead written to the `references` branch by continuous Travis CI builds. +The `output` branch contains files automatically generated by the manuscript build process. +It consists of the contents of the `output` directory of the `master` branch. +These files are not tracked in `master`, but instead written to the `output` branch by continuous Travis CI builds. ## Files This directory contains the following files: -+ [`processed-citations.tsv`](processed-citations.tsv) is a table of references extracted from the manuscript sections and their mapping to standardized citation_ids. -+ [`all-sections.md`](all-sections.md) is a markdown document of all manuscript sections, with references replaced by citation_ids. -+ [`citations.json`](citations.json) is a cache of citation metadata that can include both bibtex and CSL citation records. -+ [`bibliography.bib`](bibliography.bib) is a bibtex bibliography file of a subset of citations. -+ [`bibliography.json`](bibliography.json) is CSL-JSON file of bibliographic item metadata ([see specification](https://github.com/citation-style-language/schema/blob/master/csl-data.json)) for all references. ++ [`citations.tsv`](citations.tsv) is a table of citations extracted from the manuscript and the corresponding standard citations and citation IDs. ++ [`manuscript.md`](manuscript.md) is a markdown document of all manuscript sections, with citation strings replaced by citation IDs. ++ [`references.json`](references.json) is CSL-JSON file of bibliographic item metadata ([see specification](https://github.com/citation-style-language/schema/blob/master/csl-data.json)) for all references. ++ [`variables.json`](variables.json) contains variables that were passed to the jinja2 templater. These variables contain those automatically generated by the manubot as well as those provided by the user via the `--template-variables-path` option. -Pandoc consumes `all-sections.md` and `bibliography.json` to create the formatted manuscript. +Pandoc consumes `manuscript.md` and `references.json` to create the formatted manuscript, which is exported to `manuscript.html`, `manuscript.pdf`, and optionally `manuscript.docx`. diff --git a/bibliography.bib b/bibliography.bib deleted file mode 100644 index 7c11e64b..00000000 --- a/bibliography.bib +++ /dev/null @@ -1,3010 +0,0 @@ -@article{3qm8sXnB, - abstract = {Stochastic Gradient Descent (SGD) is a popular algorithm that can achieve -state-of-the-art performance on a variety of machine learning tasks. Several -researchers have recently proposed schemes to parallelize SGD, but all require -performance-destroying memory locking and synchronization. This work aims to -show using novel theoretical analysis, algorithms, and implementation that SGD -can be implemented without any locking. We present an update scheme called -HOGWILD! which allows processors access to shared memory with the possibility -of overwriting each other's work. We show that when the associated optimization -problem is sparse, meaning most gradient updates only modify small parts of the -decision variable, then HOGWILD! achieves a nearly optimal rate of convergence. -We demonstrate experimentally that HOGWILD! outperforms alternative schemes -that use locking by an order of magnitude.}, - archiveprefix = {arXiv}, - author = {Feng Niu and Benjamin Recht and Christopher Re and Stephen J. Wright}, - eprint = {1106.5730v2}, - file = {1106.5730v2.pdf}, - month = {Jun}, - primaryclass = {math.OC}, - title = {HOGWILD!: A Lock-Free Approach to Parallelizing Stochastic Gradient -Descent}, - url = {https://arxiv.org/abs/1106.5730v2}, - year = {2011} -} - - -@article{8RAYEOPl, - abstract = {We develop stochastic variational inference, a scalable algorithm for -approximating posterior distributions. We develop this technique for a large -class of probabilistic models and we demonstrate it with two probabilistic -topic models, latent Dirichlet allocation and the hierarchical Dirichlet -process topic model. Using stochastic variational inference, we analyze several -large collections of documents: 300K articles from Nature, 1.8M articles from -The New York Times, and 3.8M articles from Wikipedia. Stochastic inference can -easily handle data sets of this size and outperforms traditional variational -inference, which can only handle a smaller subset. (We also show that the -Bayesian nonparametric topic model outperforms its parametric counterpart.) -Stochastic variational inference lets us apply complex Bayesian models to -massive data sets.}, - archiveprefix = {arXiv}, - author = {Matt Hoffman and David M. Blei and Chong Wang and John Paisley}, - eprint = {1206.7051v3}, - file = {1206.7051v3.pdf}, - month = {Jun}, - primaryclass = {stat.ML}, - title = {Stochastic Variational Inference}, - url = {https://arxiv.org/abs/1206.7051v3}, - year = {2012} -} - - -@article{g2vvbB91, - abstract = {After a more than decade-long period of relatively little research activity -in the area of recurrent neural networks, several new developments will be -reviewed here that have allowed substantial progress both in understanding and -in technical solutions towards more efficient training of recurrent networks. -These advances have been motivated by and related to the optimization issues -surrounding deep learning. Although recurrent networks are extremely powerful -in what they can in principle represent in terms of modelling sequences,their -training is plagued by two aspects of the same issue regarding the learning of -long-term dependencies. Experiments reported here evaluate the use of clipping -gradients, spanning longer time ranges with leaky integration, advanced -momentum techniques, using more powerful output probability models, and -encouraging sparser gradients to help symmetry breaking and credit assignment. -The experiments are performed on text and music data and show off the combined -effects of these techniques in generally improving both training and test -error.}, - archiveprefix = {arXiv}, - author = {Yoshua Bengio and Nicolas Boulanger-Lewandowski and Razvan Pascanu}, - eprint = {1212.0901v2}, - file = {1212.0901v2.pdf}, - month = {12}, - primaryclass = {cs.LG}, - title = {Advances in Optimizing Recurrent Networks}, - url = {https://arxiv.org/abs/1212.0901v2}, - year = {2012} -} - - -@article{15y7iq6HF, - abstract = {This paper shows how Long Short-term Memory recurrent neural networks can be -used to generate complex sequences with long-range structure, simply by -predicting one data point at a time. The approach is demonstrated for text -(where the data are discrete) and online handwriting (where the data are -real-valued). It is then extended to handwriting synthesis by allowing the -network to condition its predictions on a text sequence. The resulting system -is able to generate highly realistic cursive handwriting in a wide variety of -styles.}, - archiveprefix = {arXiv}, - author = {Alex Graves}, - eprint = {1308.0850v5}, - file = {1308.0850v5.pdf}, - month = {Aug}, - primaryclass = {cs.NE}, - title = {Generating Sequences With Recurrent Neural Networks}, - url = {https://arxiv.org/abs/1308.0850v5}, - year = {2013} -} - - -@article{1Fel6Bdb8, - abstract = {Deep neural networks are highly expressive models that have recently achieved -state of the art performance on speech and visual recognition tasks. While -their expressiveness is the reason they succeed, it also causes them to learn -uninterpretable solutions that could have counter-intuitive properties. In this -paper we report two such properties. -First, we find that there is no distinction between individual high level -units and random linear combinations of high level units, according to various -methods of unit analysis. It suggests that it is the space, rather than the -individual units, that contains of the semantic information in the high layers -of neural networks. -Second, we find that deep neural networks learn input-output mappings that -are fairly discontinuous to a significant extend. We can cause the network to -misclassify an image by applying a certain imperceptible perturbation, which is -found by maximizing the network's prediction error. In addition, the specific -nature of these perturbations is not a random artifact of learning: the same -perturbation can cause a different network, that was trained on a different -subset of the dataset, to misclassify the same input.}, - archiveprefix = {arXiv}, - author = {Christian Szegedy and Wojciech Zaremba and Ilya Sutskever and Joan Bruna and Dumitru Erhan and Ian Goodfellow and Rob Fergus}, - eprint = {1312.6199v4}, - file = {1312.6199v4.pdf}, - month = {12}, - primaryclass = {cs.CV}, - title = {Intriguing properties of neural networks}, - url = {https://arxiv.org/abs/1312.6199v4}, - year = {2013} -} - - -@article{8t43CQ9m, - abstract = {Predicting protein secondary structure is a fundamental problem in protein -structure prediction. Here we present a new supervised generative stochastic -network (GSN) based method to predict local secondary structure with deep -hierarchical representations. GSN is a recently proposed deep learning -technique (Bengio & Thibodeau-Laufer, 2013) to globally train deep generative -model. We present the supervised extension of GSN, which learns a Markov chain -to sample from a conditional distribution, and applied it to protein structure -prediction. To scale the model to full-sized, high-dimensional data, like -protein sequences with hundreds of amino acids, we introduce a convolutional -architecture, which allows efficient learning across multiple layers of -hierarchical representations. Our architecture uniquely focuses on predicting -structured low-level labels informed with both low and high-level -representations learned by the model. In our application this corresponds to -labeling the secondary structure state of each amino-acid residue. We trained -and tested the model on separate sets of non-homologous proteins sharing less -than 30% sequence identity. Our model achieves 66.4% Q8 accuracy on the CB513 -dataset, better than the previously reported best performance 64.9% (Wang et -al., 2011) for this challenging secondary structure prediction problem.}, - archiveprefix = {arXiv}, - author = {Jian Zhou and Olga G. Troyanskaya}, - eprint = {1403.1347v1}, - file = {1403.1347v1.pdf}, - month = {Mar}, - primaryclass = {q-bio.QM}, - title = {Deep Supervised and Convolutional Generative Stochastic Network for -Protein Secondary Structure Prediction}, - url = {https://arxiv.org/abs/1403.1347v1}, - year = {2014} -} - - -@article{pxdeuhMS, - abstract = {We describe \textit{deep exponential families} (DEFs), a class of latent -variable models that are inspired by the hidden structures used in deep neural -networks. DEFs capture a hierarchy of dependencies between latent variables, and are easily generalized to many settings through exponential families. We -perform inference using recent "black box" variational inference techniques. We -then evaluate various DEFs on text and combine multiple DEFs into a model for -pairwise recommendation data. In an extensive study, we show that going beyond -one layer improves predictions for DEFs. We demonstrate that DEFs find -interesting exploratory structure in large data sets, and give better -predictive performance than state-of-the-art models.}, - archiveprefix = {arXiv}, - author = {Rajesh Ranganath and Linpeng Tang and Laurent Charlin and David M. Blei}, - eprint = {1411.2581v1}, - file = {1411.2581v1.pdf}, - month = {Dec}, - primaryclass = {stat.ML}, - title = {Deep Exponential Families}, - url = {https://arxiv.org/abs/1411.2581v1}, - year = {2014} -} - - -@article{UtcyntjF, - abstract = {Several machine learning models, including neural networks, consistently -misclassify adversarial examples---inputs formed by applying small but -intentionally worst-case perturbations to examples from the dataset, such that -the perturbed input results in the model outputting an incorrect answer with -high confidence. Early attempts at explaining this phenomenon focused on -nonlinearity and overfitting. We argue instead that the primary cause of neural -networks' vulnerability to adversarial perturbation is their linear nature. -This explanation is supported by new quantitative results while giving the -first explanation of the most intriguing fact about them: their generalization -across architectures and training sets. Moreover, this view yields a simple and -fast method of generating adversarial examples. Using this approach to provide -examples for adversarial training, we reduce the test set error of a maxout -network on the MNIST dataset.}, - archiveprefix = {arXiv}, - author = {Ian J. Goodfellow and Jonathon Shlens and Christian Szegedy}, - eprint = {1412.6572v3}, - file = {1412.6572v3.pdf}, - month = {12}, - primaryclass = {stat.ML}, - title = {Explaining and Harnessing Adversarial Examples}, - url = {https://arxiv.org/abs/1412.6572v3}, - year = {2014} -} - - -@article{Z7fd0BYf, - abstract = {Deep convolutional neural networks comprise a subclass of deep neural -networks (DNN) with a constrained architecture that leverages the spatial and -temporal structure of the domain they model. Convolutional networks achieve the -best predictive performance in areas such as speech and image recognition by -hierarchically composing simple local features into complex models. Although -DNNs have been used in drug discovery for QSAR and ligand-based bioactivity -predictions, none of these models have benefited from this powerful -convolutional architecture. This paper introduces AtomNet, the first -structure-based, deep convolutional neural network designed to predict the -bioactivity of small molecules for drug discovery applications. We demonstrate -how to apply the convolutional concepts of feature locality and hierarchical -composition to the modeling of bioactivity and chemical interactions. In -further contrast to existing DNN techniques, we show that AtomNet's application -of local convolutional filters to structural target information successfully -predicts new active molecules for targets with no previously known modulators. -Finally, we show that AtomNet outperforms previous docking approaches on a -diverse set of benchmarks by a large margin, achieving an AUC greater than 0.9 -on 57.8% of the targets in the DUDE benchmark.}, - archiveprefix = {arXiv}, - author = {Izhar Wallach and Michael Dzamba and Abraham Heifets}, - eprint = {1510.02855v1}, - file = {1510.02855v1.pdf}, - month = {Nov}, - primaryclass = {cs.LG}, - title = {AtomNet: A Deep Convolutional Neural Network for Bioactivity Prediction -in Structure-based Drug Discovery}, - url = {https://arxiv.org/abs/1510.02855v1}, - year = {2015} -} - - -@article{15lbUf0as, - abstract = {Black box variational inference allows researchers to easily prototype and -evaluate an array of models. Recent advances allow such algorithms to scale to -high dimensions. However, a central question remains: How to specify an -expressive variational distribution that maintains efficient computation? To -address this, we develop hierarchical variational models (HVMs). HVMs augment a -variational approximation with a prior on its parameters, which allows it to -capture complex structure for both discrete and continuous latent variables. -The algorithm we develop is black box, can be used for any HVM, and has the -same computational efficiency as the original approximation. We study HVMs on a -variety of deep discrete latent variable models. HVMs generalize other -expressive variational distributions and maintains higher fidelity to the -posterior.}, - archiveprefix = {arXiv}, - author = {Rajesh Ranganath and Dustin Tran and David M. Blei}, - eprint = {1511.02386v2}, - file = {1511.02386v2.pdf}, - month = {Dec}, - primaryclass = {stat.ML}, - title = {Hierarchical Variational Models}, - url = {https://arxiv.org/abs/1511.02386v2}, - year = {2015} -} - - -@article{HRXii6Ni, - abstract = {Personalized predictive medicine necessitates the modeling of patient illness -and care processes, which inherently have long-term temporal dependencies. -Healthcare observations, recorded in electronic medical records, are episodic -and irregular in time. We introduce DeepCare, an end-to-end deep dynamic neural -network that reads medical records, stores previous illness history, infers -current illness states and predicts future medical outcomes. At the data level, DeepCare represents care episodes as vectors in space, models patient health -state trajectories through explicit memory of historical records. Built on Long -Short-Term Memory (LSTM), DeepCare introduces time parameterizations to handle -irregular timed events by moderating the forgetting and consolidation of memory -cells. DeepCare also incorporates medical interventions that change the course -of illness and shape future medical risk. Moving up to the health state level, historical and present health states are then aggregated through multiscale -temporal pooling, before passing through a neural network that estimates future -outcomes. We demonstrate the efficacy of DeepCare for disease progression -modeling, intervention recommendation, and future risk prediction. On two -important cohorts with heavy social and economic burden -- diabetes and mental -health -- the results show improved modeling and risk prediction accuracy.}, - archiveprefix = {arXiv}, - author = {Trang Pham and Truyen Tran and Dinh Phung and Svetha Venkatesh}, - eprint = {1602.00357v2}, - file = {1602.00357v2.pdf}, - month = {Feb}, - primaryclass = {stat.ML}, - title = {DeepCare: A Deep Dynamic Memory Model for Predictive Medicine}, - url = {https://arxiv.org/abs/1602.00357v2}, - year = {2016} -} - - -@article{173ftiSzF, - abstract = {Observational studies are rising in importance due to the widespread -accumulation of data in fields such as healthcare, education, employment and -ecology. We consider the task of answering counterfactual questions such as, "Would this patient have lower blood sugar had she received a different -medication?". We propose a new algorithmic framework for counterfactual -inference which brings together ideas from domain adaptation and representation -learning. In addition to a theoretical justification, we perform an empirical -comparison with previous approaches to causal inference from observational -data. Our deep learning algorithm significantly outperforms the previous -state-of-the-art.}, - archiveprefix = {arXiv}, - author = {Fredrik D. Johansson and Uri Shalit and David Sontag}, - eprint = {1605.03661v2}, - file = {1605.03661v2.pdf}, - month = {May}, - primaryclass = {stat.ML}, - title = {Learning Representations for Counterfactual Inference}, - url = {https://arxiv.org/abs/1605.03661v2}, - year = {2016} -} - - -@article{5Il3kN32, - abstract = {Large labeled training sets are the critical building blocks of supervised -learning methods and are key enablers of deep learning techniques. For some -applications, creating labeled training sets is the most time-consuming and -expensive part of applying machine learning. We therefore propose a paradigm -for the programmatic creation of training sets called data programming in which -users express weak supervision strategies or domain heuristics as labeling -functions, which are programs that label subsets of the data, but that are -noisy and may conflict. We show that by explicitly representing this training -set labeling process as a generative model, we can "denoise" the generated -training set, and establish theoretically that we can recover the parameters of -these generative models in a handful of settings. We then show how to modify a -discriminative loss function to make it noise-aware, and demonstrate our method -over a range of discriminative models including logistic regression and LSTMs. -Experimentally, on the 2014 TAC-KBP Slot Filling challenge, we show that data -programming would have led to a new winning score, and also show that applying -data programming to an LSTM model leads to a TAC-KBP score almost 6 F1 points -over a state-of-the-art LSTM baseline (and into second place in the -competition). Additionally, in initial user studies we observed that data -programming may be an easier way for non-experts to create machine learning -models when training data is limited or unavailable.}, - archiveprefix = {arXiv}, - author = {Alexander Ratner and Christopher De Sa and Sen Wu and Daniel Selsam and Christopher Ré}, - eprint = {1605.07723v3}, - file = {1605.07723v3.pdf}, - month = {May}, - primaryclass = {stat.ML}, - title = {Data Programming: Creating Large Training Sets, Quickly}, - url = {https://arxiv.org/abs/1605.07723v3}, - year = {2016} -} - - -@article{1FE0F2pQ, - abstract = {Medical practitioners use survival models to explore and understand the -relationships between patients' covariates (e.g. clinical and genetic features) -and the effectiveness of various treatment options. Standard survival models -like the linear Cox proportional hazards model require extensive feature -engineering or prior medical knowledge to model treatment interaction at an -individual level. While nonlinear survival methods, such as neural networks and -survival forests, can inherently model these high-level interaction terms, they -have yet to be shown as effective treatment recommender systems. We introduce -DeepSurv, a Cox proportional hazards deep neural network and state-of-the-art -survival method for modeling interactions between a patient's covariates and -treatment effectiveness in order to provide personalized treatment -recommendations. We perform a number of experiments training DeepSurv on -simulated and real survival data. We demonstrate that DeepSurv performs as well -as or better than other state-of-the-art survival models and validate that -DeepSurv successfully models increasingly complex relationships between a -patient's covariates and their risk of failure. We then show how DeepSurv -models the relationship between a patient's features and effectiveness of -different treatment options to show how DeepSurv can be used to provide -individual treatment recommendations. Finally, we train DeepSurv on real -clinical studies to demonstrate how it's personalized treatment recommendations -would increase the survival time of a set of patients. The predictive and -modeling capabilities of DeepSurv will enable medical researchers to use deep -neural networks as a tool in their exploration, understanding, and prediction -of the effects of a patient's characteristics on their risk of failure.}, - archiveprefix = {arXiv}, - author = {Jared Katzman and Uri Shaham and Jonathan Bates and Alexander Cloninger and Tingting Jiang and Yuval Kluger}, - eprint = {1606.00931v3}, - file = {1606.00931v3.pdf}, - month = {Jun}, - primaryclass = {stat.ML}, - title = {DeepSurv: Personalized Treatment Recommender System Using A Cox -Proportional Hazards Deep Neural Network}, - url = {https://arxiv.org/abs/1606.00931v3}, - year = {2016} -} - - -@article{mbEp6jNr, - abstract = {The International Symposium on Biomedical Imaging (ISBI) held a grand -challenge to evaluate computational systems for the automated detection of -metastatic breast cancer in whole slide images of sentinel lymph node biopsies. -Our team won both competitions in the grand challenge, obtaining an area under -the receiver operating curve (AUC) of 0.925 for the task of whole slide image -classification and a score of 0.7051 for the tumor localization task. A -pathologist independently reviewed the same images, obtaining a whole slide -image classification AUC of 0.966 and a tumor localization score of 0.733. -Combining our deep learning system's predictions with the human pathologist's -diagnoses increased the pathologist's AUC to 0.995, representing an -approximately 85 percent reduction in human error rate. These results -demonstrate the power of using deep learning to produce significant -improvements in the accuracy of pathological diagnoses.}, - archiveprefix = {arXiv}, - author = {Dayong Wang and Aditya Khosla and Rishab Gargeya and Humayun Irshad and Andrew H. Beck}, - eprint = {1606.05718v1}, - file = {1606.05718v1.pdf}, - month = {Jun}, - primaryclass = {q-bio.QM}, - title = {Deep Learning for Identifying Metastatic Breast Cancer}, - url = {https://arxiv.org/abs/1606.05718v1}, - year = {2016} -} - - -@article{7yE9K08a, - abstract = {We summarize the potential impact that the European Union's new General Data -Protection Regulation will have on the routine use of machine learning -algorithms. Slated to take effect as law across the EU in 2018, it will -restrict automated individual decision-making (that is, algorithms that make -decisions based on user-level predictors) which "significantly affect" users. -The law will also effectively create a "right to explanation," whereby a user -can ask for an explanation of an algorithmic decision that was made about them. -We argue that while this law will pose large challenges for industry, it -highlights opportunities for computer scientists to take the lead in designing -algorithms and evaluation frameworks which avoid discrimination and enable -explanation.}, - archiveprefix = {arXiv}, - author = {Bryce Goodman and Seth Flaxman}, - eprint = {1606.08813v3}, - file = {1606.08813v3.pdf}, - month = {Jun}, - primaryclass = {stat.ML}, - title = {European Union regulations on algorithmic decision-making and a "right -to explanation"}, - url = {https://arxiv.org/abs/1606.08813v3}, - year = {2016} -} - - -@article{ucHUOABT, - abstract = {Machine learning techniques based on neural networks are achieving remarkable -results in a wide variety of domains. Often, the training of models requires -large, representative datasets, which may be crowdsourced and contain sensitive -information. The models should not expose private information in these -datasets. Addressing this goal, we develop new algorithmic techniques for -learning and a refined analysis of privacy costs within the framework of -differential privacy. Our implementation and experiments demonstrate that we -can train deep neural networks with non-convex objectives, under a modest -privacy budget, and at a manageable cost in software complexity, training -efficiency, and model quality.}, - archiveprefix = {arXiv}, - author = {Martín Abadi and Andy Chu and Ian Goodfellow and H. Brendan McMahan and Ilya Mironov and Kunal Talwar and Li Zhang}, - doi = {10.1145/2976749.2978318}, - eprint = {1607.00133v2}, - file = {1607.00133v2.pdf}, - month = {Jul}, - primaryclass = {stat.ML}, - title = {Deep Learning with Differential Privacy}, - url = {https://arxiv.org/abs/1607.00133v2}, - year = {2016} -} - - -@article{Ohd1Q9Xw, - abstract = {Feature engineering remains a major bottleneck when creating predictive -systems from electronic medical records. At present, an important missing -element is detecting predictive regular clinical motifs from irregular episodic -records. We present Deepr (short for Deep record), a new end-to-end deep -learning system that learns to extract features from medical records and -predicts future risk automatically. Deepr transforms a record into a sequence -of discrete elements separated by coded time gaps and hospital transfers. On -top of the sequence is a convolutional neural net that detects and combines -predictive local clinical motifs to stratify the risk. Deepr permits -transparent inspection and visualization of its inner working. We validate -Deepr on hospital data to predict unplanned readmission after discharge. Deepr -achieves superior accuracy compared to traditional techniques, detects -meaningful clinical motifs, and uncovers the underlying structure of the -disease and intervention space.}, - archiveprefix = {arXiv}, - author = {Phuoc Nguyen and Truyen Tran and Nilmini Wickramasinghe and Svetha Venkatesh}, - eprint = {1607.07519v1}, - file = {1607.07519v1.pdf}, - month = {Jul}, - primaryclass = {stat.ML}, - title = {Deepr: A Convolutional Net for Medical Records}, - url = {https://arxiv.org/abs/1607.07519v1}, - year = {2016} -} - - -@article{c6MfDdWP, - abstract = {Disparate areas of machine learning have benefited from models that can take -raw data with little preprocessing as input and learn rich representations of -that raw data in order to perform well on a given prediction task. We evaluate -this approach in healthcare by using longitudinal measurements of lab tests, one of the more raw signals of a patient's health state widely available in -clinical data, to predict disease onsets. In particular, we train a Long -Short-Term Memory (LSTM) recurrent neural network and two novel convolutional -neural networks for multi-task prediction of disease onset for 133 conditions -based on 18 common lab tests measured over time in a cohort of 298K patients -derived from 8 years of administrative claims data. We compare the neural -networks to a logistic regression with several hand-engineered, clinically -relevant features. We find that the representation-based learning approaches -significantly outperform this baseline. We believe that our work suggests a new -avenue for patient risk stratification based solely on lab results.}, - archiveprefix = {arXiv}, - author = {Narges Razavian and Jake Marcus and David Sontag}, - eprint = {1608.00647v3}, - file = {1608.00647v3.pdf}, - month = {Aug}, - primaryclass = {cs.LG}, - title = {Multi-task Prediction of Disease Onsets from Longitudinal Lab Tests}, - url = {https://arxiv.org/abs/1608.00647v3}, - year = {2016} -} - - -@article{qXdO2aMm, - abstract = {The electronic health record (EHR) provides an unprecedented opportunity to -build actionable tools to support physicians at the point of care. In this -paper, we investigate survival analysis in the context of EHR data. We -introduce deep survival analysis, a hierarchical generative approach to -survival analysis. It departs from previous approaches in two primary ways: (1) -all observations, including covariates, are modeled jointly conditioned on a -rich latent structure; and (2) the observations are aligned by their failure -time, rather than by an arbitrary time zero as in traditional survival -analysis. Further, it (3) scalably handles heterogeneous (continuous and -discrete) data types that occur in the EHR. We validate deep survival analysis -model by stratifying patients according to risk of developing coronary heart -disease (CHD). Specifically, we study a dataset of 313,000 patients -corresponding to 5.5 million months of observations. When compared to the -clinically validated Framingham CHD risk score, deep survival analysis is -significantly superior in stratifying patients according to their risk.}, - archiveprefix = {arXiv}, - author = {Rajesh Ranganath and Adler Perotte and Noémie Elhadad and David Blei}, - eprint = {1608.02158v2}, - file = {1608.02158v2.pdf}, - month = {Aug}, - primaryclass = {stat.ML}, - title = {Deep Survival Analysis}, - url = {https://arxiv.org/abs/1608.02158v2}, - year = {2016} -} - - -@article{ULSPV0rh, - abstract = {Machine learning (ML) models may be deemed confidential due to their -sensitive training data, commercial value, or use in security applications. -Increasingly often, confidential ML models are being deployed with publicly -accessible query interfaces. ML-as-a-service ("predictive analytics") systems -are an example: Some allow users to train models on potentially sensitive data -and charge others for access on a pay-per-query basis. -The tension between model confidentiality and public access motivates our -investigation of model extraction attacks. In such attacks, an adversary with -black-box access, but no prior knowledge of an ML model's parameters or -training data, aims to duplicate the functionality of (i.e., "steal") the -model. Unlike in classical learning theory settings, ML-as-a-service offerings -may accept partial feature vectors as inputs and include confidence values with -predictions. Given these practices, we show simple, efficient attacks that -extract target ML models with near-perfect fidelity for popular model classes -including logistic regression, neural networks, and decision trees. We -demonstrate these attacks against the online services of BigML and Amazon -Machine Learning. We further show that the natural countermeasure of omitting -confidence values from model outputs still admits potentially harmful model -extraction attacks. Our results highlight the need for careful ML model -deployment and new model extraction countermeasures.}, - archiveprefix = {arXiv}, - author = {Florian Tramèr and Fan Zhang and Ari Juels and Michael K. Reiter and Thomas Ristenpart}, - eprint = {1609.02943v2}, - file = {1609.02943v2.pdf}, - month = {Sep}, - primaryclass = {cs.CR}, - title = {Stealing Machine Learning Models via Prediction APIs}, - url = {https://arxiv.org/abs/1609.02943v2}, - year = {2016} -} - - -@article{4TK06zOf, - abstract = {Neural Machine Translation (NMT) is an end-to-end learning approach for -automated translation, with the potential to overcome many of the weaknesses of -conventional phrase-based translation systems. Unfortunately, NMT systems are -known to be computationally expensive both in training and in translation -inference. Also, most NMT systems have difficulty with rare words. These issues -have hindered NMT's use in practical deployments and services, where both -accuracy and speed are essential. In this work, we present GNMT, Google's -Neural Machine Translation system, which attempts to address many of these -issues. Our model consists of a deep LSTM network with 8 encoder and 8 decoder -layers using attention and residual connections. To improve parallelism and -therefore decrease training time, our attention mechanism connects the bottom -layer of the decoder to the top layer of the encoder. To accelerate the final -translation speed, we employ low-precision arithmetic during inference -computations. To improve handling of rare words, we divide words into a limited -set of common sub-word units ("wordpieces") for both input and output. This -method provides a good balance between the flexibility of "character"-delimited -models and the efficiency of "word"-delimited models, naturally handles -translation of rare words, and ultimately improves the overall accuracy of the -system. Our beam search technique employs a length-normalization procedure and -uses a coverage penalty, which encourages generation of an output sentence that -is most likely to cover all the words in the source sentence. On the WMT'14 -English-to-French and English-to-German benchmarks, GNMT achieves competitive -results to state-of-the-art. Using a human side-by-side evaluation on a set of -isolated simple sentences, it reduces translation errors by an average of 60% -compared to Google's phrase-based production system.}, - archiveprefix = {arXiv}, - author = {Yonghui Wu and Mike Schuster and Zhifeng Chen and Quoc V. Le and Mohammad Norouzi and Wolfgang Macherey and Maxim Krikun and Yuan Cao and Qin Gao and Klaus Macherey and Jeff Klingner and Apurva Shah and Melvin Johnson and Xiaobing Liu and Łukasz Kaiser and Stephan Gouws and Yoshikiyo Kato and Taku Kudo and Hideto Kazawa and Keith Stevens and George Kurian and Nishant Patil and Wei Wang and Cliff Young and Jason Smith and Jason Riesa and Alex Rudnick and Oriol Vinyals and Greg Corrado and Macduff Hughes and Jeffrey Dean}, - eprint = {1609.08144v2}, - file = {1609.08144v2.pdf}, - month = {Sep}, - primaryclass = {cs.CL}, - title = {Google's Neural Machine Translation System: Bridging the Gap between -Human and Machine Translation}, - url = {https://arxiv.org/abs/1609.08144v2}, - year = {2016} -} - - -@article{1ENxzq6pT, - abstract = {We propose a criterion for discrimination against a specified sensitive -attribute in supervised learning, where the goal is to predict some target -based on available features. Assuming data about the predictor, target, and -membership in the protected group are available, we show how to optimally -adjust any learned predictor so as to remove discrimination according to our -definition. Our framework also improves incentives by shifting the cost of poor -classification from disadvantaged groups to the decision maker, who can respond -by improving the classification accuracy. -In line with other studies, our notion is oblivious: it depends only on the -joint statistics of the predictor, the target and the protected attribute, but -not on interpretation of individualfeatures. We study the inherent limits of -defining and identifying biases based on such oblivious measures, outlining -what can and cannot be inferred from different oblivious tests. -We illustrate our notion using a case study of FICO credit scores.}, - archiveprefix = {arXiv}, - author = {Moritz Hardt and Eric Price and Nathan Srebro}, - eprint = {1610.02413v1}, - file = {1610.02413v1.pdf}, - month = {Nov}, - primaryclass = {cs.LG}, - title = {Equality of Opportunity in Supervised Learning}, - url = {https://arxiv.org/abs/1610.02413v1}, - year = {2016} -} - - -@article{M2OLWojE, - abstract = {Conversational speech recognition has served as a flagship speech recognition -task since the release of the Switchboard corpus in the 1990s. In this paper, we measure the human error rate on the widely used NIST 2000 test set, and find -that our latest automated system has reached human parity. The error rate of -professional transcribers is 5.9% for the Switchboard portion of the data, in -which newly acquainted pairs of people discuss an assigned topic, and 11.3% for -the CallHome portion where friends and family members have open-ended -conversations. In both cases, our automated system establishes a new state of -the art, and edges past the human benchmark, achieving error rates of 5.8% and -11.0%, respectively. The key to our system's performance is the use of various -convolutional and LSTM acoustic model architectures, combined with a novel -spatial smoothing method and lattice-free MMI acoustic training, multiple -recurrent neural network language modeling approaches, and a systematic use of -system combination.}, - archiveprefix = {arXiv}, - author = {W. Xiong and J. Droppo and X. Huang and F. Seide and M. Seltzer and A. Stolcke and D. Yu and G. Zweig}, - eprint = {1610.05256v2}, - file = {1610.05256v2.pdf}, - month = {Nov}, - primaryclass = {cs.CL}, - title = {Achieving Human Parity in Conversational Speech Recognition}, - url = {https://arxiv.org/abs/1610.05256v2}, - year = {2016} -} - - -@article{1HbRTExaU, - abstract = {We quantitatively investigate how machine learning models leak information -about the individual data records on which they were trained. We focus on the -basic membership inference attack: given a data record and black-box access to -a model, determine if the record was in the model's training dataset. To -perform membership inference against a target model, we make adversarial use of -machine learning and train our own inference model to recognize differences in -the target model's predictions on the inputs that it trained on versus the -inputs that it did not train on. -We empirically evaluate our inference techniques on classification models -trained by commercial "machine learning as a service" providers such as Google -and Amazon. Using realistic datasets and classification tasks, including a -hospital discharge dataset whose membership is sensitive from the privacy -perspective, we show that these models can be vulnerable to membership -inference attacks. We then investigate the factors that influence this leakage -and evaluate mitigation strategies.}, - archiveprefix = {arXiv}, - author = {Reza Shokri and Marco Stronati and Congzheng Song and Vitaly Shmatikov}, - eprint = {1610.05820v2}, - file = {1610.05820v2.pdf}, - month = {Nov}, - primaryclass = {cs.CR}, - title = {Membership Inference Attacks against Machine Learning Models}, - url = {https://arxiv.org/abs/1610.05820v2}, - year = {2016} -} - - -@article{11aqfNfQx, - abstract = {We study fairness in linear bandit problems. Starting from the notion of -meritocratic fairness introduced in Joseph et al. [2016], we carry out a more -refined analysis of a more general problem, achieving better performance -guarantees with fewer modelling assumptions on the number and structure of -available choices as well as the number selected. We also analyze the -previously-unstudied question of fairness in infinite linear bandit problems, obtaining instance-dependent regret upper bounds as well as lower bounds -demonstrating that this instance-dependence is necessary. The result is a -framework for meritocratic fairness in an online linear setting that is -substantially more powerful, general, and realistic than the current state of -the art.}, - archiveprefix = {arXiv}, - author = {Matthew Joseph and Michael Kearns and Jamie Morgenstern and Seth Neel and Aaron Roth}, - eprint = {1610.09559v4}, - file = {1610.09559v4.pdf}, - month = {Nov}, - primaryclass = {cs.LG}, - title = {Fair Algorithms for Infinite and Contextual Bandits}, - url = {https://arxiv.org/abs/1610.09559v4}, - year = {2016} -} - - -@article{lERqKdZJ, - abstract = {This paper proposes a general method for improving the structure and quality -of sequences generated by a recurrent neural network (RNN), while maintaining -information originally learned from data, as well as sample diversity. An RNN -is first pre-trained on data using maximum likelihood estimation (MLE), and the -probability distribution over the next token in the sequence learned by this -model is treated as a prior policy. Another RNN is then trained using -reinforcement learning (RL) to generate higher-quality outputs that account for -domain-specific incentives while retaining proximity to the prior policy of the -MLE RNN. To formalize this objective, we derive novel off-policy RL methods for -RNNs from KL-control. The effectiveness of the approach is demonstrated on two -applications; 1) generating novel musical melodies, and 2) computational -molecular generation. For both problems, we show that the proposed method -improves the desired properties and structure of the generated sequences, while -maintaining information learned from data.}, - archiveprefix = {arXiv}, - author = {Natasha Jaques and Shixiang Gu and Dzmitry Bahdanau and José Miguel Hernández-Lobato and Richard E. Turner and Douglas Eck}, - eprint = {1611.02796v8}, - file = {1611.02796v8.pdf}, - month = {Dec}, - primaryclass = {cs.LG}, - title = {Sequence Tutor: Conservative Fine-Tuning of Sequence Generation Models -with KL-control}, - url = {https://arxiv.org/abs/1611.02796v8}, - year = {2016} -} - - -@article{AsLAb71x, - abstract = {Advances in machine learning (ML) in recent years have enabled a dizzying -array of applications such as data analytics, autonomous systems, and security -diagnostics. ML is now pervasive---new systems and models are being deployed in -every domain imaginable, leading to rapid and widespread deployment of software -based inference and decision making. There is growing recognition that ML -exposes new vulnerabilities in software systems, yet the technical community's -understanding of the nature and extent of these vulnerabilities remains -limited. We systematize recent findings on ML security and privacy, focusing on -attacks identified on these systems and defenses crafted to date. We articulate -a comprehensive threat model for ML, and categorize attacks and defenses within -an adversarial framework. Key insights resulting from works both in the ML and -security communities are identified and the effectiveness of approaches are -related to structural elements of ML algorithms and the data used to train -them. We conclude by formally exploring the opposing relationship between model -accuracy and resilience to adversarial manipulation. Through these -explorations, we show that there are (possibly unavoidable) tensions between -model complexity, accuracy, and resilience that must be calibrated for the -environments in which they will be used.}, - archiveprefix = {arXiv}, - author = {Nicolas Papernot and Patrick McDaniel and Arunesh Sinha and Michael Wellman}, - eprint = {1611.03814v1}, - file = {1611.03814v1.pdf}, - month = {Dec}, - primaryclass = {cs.CR}, - title = {Towards the Science of Security and Privacy in Machine Learning}, - url = {https://arxiv.org/abs/1611.03814v1}, - year = {2016} -} - - -@article{dO844vZn, - abstract = {Automated extraction of concepts from patient clinical records is an -essential facilitator of clinical research. For this reason, the 2010 i2b2/VA -Natural Language Processing Challenges for Clinical Records introduced a -concept extraction task aimed at identifying and classifying concepts into -predefined categories (i.e., treatments, tests and problems). State-of-the-art -concept extraction approaches heavily rely on handcrafted features and -domain-specific resources which are hard to collect and define. For this -reason, this paper proposes an alternative, streamlined approach: a recurrent -neural network (the bidirectional LSTM with CRF decoding) initialized with -general-purpose, off-the-shelf word embeddings. The experimental results -achieved on the 2010 i2b2/VA reference corpora using the proposed framework -outperform all recent methods and ranks closely to the best submission from the -original 2010 i2b2/VA challenge.}, - archiveprefix = {arXiv}, - author = {Raghavendra Chalapathy and Ehsan Zare Borzeshi and Massimo Piccardi}, - eprint = {1611.08373v1}, - file = {1611.08373v1.pdf}, - month = {Dec}, - primaryclass = {stat.ML}, - title = {Bidirectional LSTM-CRF for Clinical Concept Extraction}, - url = {https://arxiv.org/abs/1611.08373v1}, - year = {2016} -} - - -@article{apBChoyF, - abstract = {The recent rapid and tremendous success of deep convolutional neural networks -(CNN) on many challenging computer vision tasks largely derives from the -accessibility of the well-annotated ImageNet and PASCAL VOC datasets. -Nevertheless, unsupervised image categorization (i.e., without the ground-truth -labeling) is much less investigated, yet critically important and difficult -when annotations are extremely hard to obtain in the conventional way of -"Google Search" and crowd sourcing. We address this problem by presenting a -looped deep pseudo-task optimization (LDPO) framework for joint mining of deep -CNN features and image labels. Our method is conceptually simple and rests upon -the hypothesized "convergence" of better labels leading to better trained CNN -models which in turn feed more discriminative image representations to -facilitate more meaningful clusters/labels. Our proposed method is validated in -tackling two important applications: 1) Large-scale medical image annotation -has always been a prohibitively expensive and easily-biased task even for -well-trained radiologists. Significantly better image categorization results -are achieved via our proposed approach compared to the previous -state-of-the-art method. 2) Unsupervised scene recognition on representative -and publicly available datasets with our proposed technique is examined. The -LDPO achieves excellent quantitative scene classification results. On the MIT -indoor scene dataset, it attains a clustering accuracy of 75.3%, compared to -the state-of-the-art supervised classification accuracy of 81.0% (when both are -based on the VGG-VD model).}, - archiveprefix = {arXiv}, - author = {Xiaosong Wang and Le Lu and Hoo-chang Shin and Lauren Kim and Mohammadhadi Bagheri and Isabella Nogues and Jianhua Yao and Ronald M. Summers}, - eprint = {1701.06599v1}, - file = {1701.06599v1.pdf}, - month = {Jan}, - primaryclass = {cs.CV}, - title = {Unsupervised Joint Mining of Deep Features and Image Labels for -Large-scale Radiology Image Categorization and Scene Recognition}, - url = {https://arxiv.org/abs/1701.06599v1}, - year = {2017} -} - - -@article{AQ3N6Ayw, - abstract = {Deep generative models have been wildly successful at learning coherent -latent representations for continuous data such as video and audio. However, generative modeling of discrete data such as arithmetic expressions and -molecular structures still poses significant challenges. Crucially, state-of-the-art methods often produce outputs that are not valid. We make the -key observation that frequently, discrete data can be represented as a parse -tree from a context-free grammar. We propose a variational autoencoder which -encodes and decodes directly to and from these parse trees, ensuring the -generated outputs are always valid. Surprisingly, we show that not only does -our model more often generate valid outputs, it also learns a more coherent -latent space in which nearby points decode to similar discrete outputs. We -demonstrate the effectiveness of our learned models by showing their improved -performance in Bayesian optimization for symbolic regression and molecular -synthesis.}, - archiveprefix = {arXiv}, - author = {Matt J. Kusner and Brooks Paige and José Miguel Hernández-Lobato}, - eprint = {1703.01925v1}, - file = {1703.01925v1.pdf}, - month = {Mar}, - primaryclass = {stat.ML}, - title = {Grammar Variational Autoencoder}, - url = {https://arxiv.org/abs/1703.01925v1}, - year = {2017} -} - - -@article{wKioubsT, - abstract = {One of the most difficult speech recognition tasks is accurate recognition of -human to human communication. Advances in deep learning over the last few years -have produced major speech recognition improvements on the representative -Switchboard conversational corpus. Word error rates that just a few years ago -were 14% have dropped to 8.0%, then 6.6% and most recently 5.8%, and are now -believed to be within striking range of human performance. This then raises two -issues - what IS human performance, and how far down can we still drive speech -recognition error rates? A recent paper by Microsoft suggests that we have -already achieved human performance. In trying to verify this statement, we -performed an independent set of human performance measurements on two -conversational tasks and found that human performance may be considerably -better than what was earlier reported, giving the community a significantly -harder goal to achieve. We also report on our own efforts in this area, presenting a set of acoustic and language modeling techniques that lowered the -word error rate of our own English conversational telephone LVCSR system to the -level of 5.5%/10.3% on the Switchboard/CallHome subsets of the Hub5 2000 -evaluation, which - at least at the writing of this paper - is a new -performance milestone (albeit not at what we measure to be human performance!). -On the acoustic side, we use a score fusion of three models: one LSTM with -multiple feature inputs, a second LSTM trained with speaker-adversarial -multi-task learning and a third residual net (ResNet) with 25 convolutional -layers and time-dilated convolutions. On the language modeling side, we use -word and character LSTMs and convolutional WaveNet-style language models.}, - archiveprefix = {arXiv}, - author = {George Saon and Gakuto Kurata and Tom Sercu and Kartik Audhkhasi and Samuel Thomas and Dimitrios Dimitriadis and Xiaodong Cui and Bhuvana Ramabhadran and Michael Picheny and Lynn-Li Lim and Bergul Roomi and Phil Hall}, - eprint = {1703.02136v1}, - file = {1703.02136v1.pdf}, - month = {Mar}, - primaryclass = {cs.CL}, - title = {English Conversational Telephone Speech Recognition by Humans and -Machines}, - url = {https://arxiv.org/abs/1703.02136v1}, - year = {2017} -} - - -@article{xl1ijigK, - abstract = {Access to electronic health records (EHR) data has motivated computational -advances in medical research. However, various concerns, particularly over -privacy, can limit access to and collaborative use of EHR data. Sharing -synthetic EHR data could mitigate risk. In this paper, we propose a new -approach, medical Generative Adversarial Network (medGAN), to generate -realistic synthetic EHRs. Based on an input EHR dataset, medGAN can generate -high-dimensional discrete variables (e.g., binary and count features) via a -combination of an autoencoder and generative adversarial networks. We also -propose minibatch averaging to efficiently avoid mode collapse, and increase -the learning efficiency with batch normalization and shortcut connections. To -demonstrate feasibility, we showed that medGAN generates synthetic EHR datasets -that achieve comparable performance to real data on many experiments including -distribution statistics, predictive modeling tasks and medical expert review.}, - archiveprefix = {arXiv}, - author = {Edward Choi and Siddharth Biswal and Bradley Malin and Jon Duke and Walter F. Stewart and Jimeng Sun}, - eprint = {1703.06490v1}, - file = {1703.06490v1.pdf}, - month = {Mar}, - primaryclass = {cs.LG}, - title = {Generating Multi-label Discrete Electronic Health Records using -Generative Adversarial Networks}, - url = {https://arxiv.org/abs/1703.06490v1}, - year = {2017} -} - - -@article{17YaKNLKk, - abstract = {Empirical scoring functions based on either molecular force fields or -cheminformatics descriptors are widely used, in conjunction with molecular -docking, during the early stages of drug discovery to predict potency and -binding affinity of a drug-like molecule to a given target. These models -require expert-level knowledge of physical chemistry and biology to be encoded -as hand-tuned parameters or features rather than allowing the underlying model -to select features in a data-driven procedure. Here, we develop a general -3-dimensional spatial convolution operation for learning atomic-level chemical -interactions directly from atomic coordinates and demonstrate its application -to structure-based bioactivity prediction. The atomic convolutional neural -network is trained to predict the experimentally determined binding affinity of -a protein-ligand complex by direct calculation of the energy associated with -the complex, protein, and ligand given the crystal structure of the binding -pose. Non-covalent interactions present in the complex that are absent in the -protein-ligand sub-structures are identified and the model learns the -interaction strength associated with these features. We test our model by -predicting the binding free energy of a subset of protein-ligand complexes -found in the PDBBind dataset and compare with state-of-the-art cheminformatics -and machine learning-based approaches. We find that all methods achieve -experimental accuracy and that atomic convolutional networks either outperform -or perform competitively with the cheminformatics based methods. Unlike all -previous protein-ligand prediction systems, atomic convolutional networks are -end-to-end and fully-differentiable. They represent a new data-driven, physics-based deep learning model paradigm that offers a strong foundation for -future improvements in structure-based bioactivity prediction.}, - archiveprefix = {arXiv}, - author = {Joseph Gomes and Bharath Ramsundar and Evan N. Feinberg and Vijay S. Pande}, - eprint = {1703.10603v1}, - file = {1703.10603v1.pdf}, - month = {Mar}, - primaryclass = {cs.LG}, - title = {Atomic Convolutional Networks for Predicting Protein-Ligand Binding -Affinity}, - url = {https://arxiv.org/abs/1703.10603v1}, - year = {2017} -} - - -@article{18lZK7fxH, - abstract = {Although deep neural networks (DNNs) have achieved great success in many -computer vision tasks, recent studies have shown they are vulnerable to -adversarial examples. Such examples, typically generated by adding small but -purposeful distortions, can frequently fool DNN models. Previous studies to -defend against adversarial examples mostly focused on refining the DNN models. -They have either shown limited success or suffer from the expensive -computation. We propose a new strategy, \emph{feature squeezing}, that can be -used to harden DNN models by detecting adversarial examples. Feature squeezing -reduces the search space available to an adversary by coalescing samples that -correspond to many different feature vectors in the original space into a -single sample. By comparing a DNN model's prediction on the original input with -that on the squeezed input, feature squeezing detects adversarial examples with -high accuracy and few false positives. This paper explores two instances of -feature squeezing: reducing the color bit depth of each pixel and smoothing -using a spatial filter. These strategies are straightforward, inexpensive, and -complementary to defensive methods that operate on the underlying model, such -as adversarial training.}, - archiveprefix = {arXiv}, - author = {Weilin Xu and David Evans and Yanjun Qi}, - eprint = {1704.01155v1}, - file = {1704.01155v1.pdf}, - month = {Apr}, - primaryclass = {cs.CV}, - title = {Feature Squeezing: Detecting Adversarial Examples in Deep Neural -Networks}, - url = {https://arxiv.org/abs/1704.01155v1}, - year = {2017} -} - - -@article{ULagTifF, - abstract = {Many architects believe that major improvements in cost-energy-performance -must now come from domain-specific hardware. This paper evaluates a custom -ASIC---called a Tensor Processing Unit (TPU)---deployed in datacenters since -2015 that accelerates the inference phase of neural networks (NN). The heart of -the TPU is a 65,536 8-bit MAC matrix multiply unit that offers a peak -throughput of 92 TeraOps/second (TOPS) and a large (28 MiB) software-managed -on-chip memory. The TPU's deterministic execution model is a better match to -the 99th-percentile response-time requirement of our NN applications than are -the time-varying optimizations of CPUs and GPUs (caches, out-of-order -execution, multithreading, multiprocessing, prefetching, ...) that help average -throughput more than guaranteed latency. The lack of such features helps -explain why, despite having myriad MACs and a big memory, the TPU is relatively -small and low power. We compare the TPU to a server-class Intel Haswell CPU and -an Nvidia K80 GPU, which are contemporaries deployed in the same datacenters. -Our workload, written in the high-level TensorFlow framework, uses production -NN applications (MLPs, CNNs, and LSTMs) that represent 95% of our datacenters' -NN inference demand. Despite low utilization for some applications, the TPU is -on average about 15X - 30X faster than its contemporary GPU or CPU, with -TOPS/Watt about 30X - 80X higher. Moreover, using the GPU's GDDR5 memory in the -TPU would triple achieved TOPS and raise TOPS/Watt to nearly 70X the GPU and -200X the CPU.}, - archiveprefix = {arXiv}, - author = {Norman P. Jouppi and Cliff Young and Nishant Patil and David Patterson and Gaurav Agrawal and Raminder Bajwa and Sarah Bates and Suresh Bhatia and Nan Boden and Al Borchers and Rick Boyle and Pierre-luc Cantin and Clifford Chao and Chris Clark and Jeremy Coriell and Mike Daley and Matt Dau and Jeffrey Dean and Ben Gelb and Tara Vazir Ghaemmaghami and Rajendra Gottipati and William Gulland and Robert Hagmann and C. Richard Ho and Doug Hogberg and John Hu and Robert Hundt and Dan Hurt and Julian Ibarz and Aaron Jaffey and Alek Jaworski and Alexander Kaplan and Harshit Khaitan and Andy Koch and Naveen Kumar and Steve Lacy and James Laudon and James Law and Diemthu Le and Chris Leary and Zhuyuan Liu and Kyle Lucke and Alan Lundin and Gordon MacKean and Adriana Maggiore and Maire Mahony and Kieran Miller and Rahul Nagarajan and Ravi Narayanaswami and Ray Ni and Kathy Nix and Thomas Norrie and Mark Omernick and Narayana Penukonda and Andy Phelps and Jonathan Ross and Matt Ross and Amir Salek and Emad Samadiani and Chris Severn and Gregory Sizikov and Matthew Snelham and Jed Souter and Dan Steinberg and Andy Swing and Mercedes Tan and Gregory Thorson and Bo Tian and Horia Toma and Erick Tuttle and Vijay Vasudevan and Richard Walter and Walter Wang and Eric Wilcox and Doe Hyun Yoon}, - eprint = {1704.04760v1}, - file = {1704.04760v1.pdf}, - month = {Apr}, - primaryclass = {cs.AR}, - title = {In-Datacenter Performance Analysis of a Tensor Processing Unit}, - url = {https://arxiv.org/abs/1704.04760v1}, - year = {2017} -} - - -@article{39RPiE10, - abstract = {Computational prediction of membrane protein (MP) structures is very -challenging partially due to lack of sufficient solved structures for homology -modeling. Recently direct evolutionary coupling analysis (DCA) sheds some light -on protein contact prediction and accordingly, contact-assisted folding, but -DCA is effective only on some very large-sized families since it uses -information only in a single protein family. This paper presents a deep -transfer learning method that can significantly improve MP contact prediction -by learning contact patterns and complex sequence-contact relationship from -thousands of non-membrane proteins (non-MPs). Tested on 510 non-redundant MPs, our deep model (learned from only non-MPs) has top L/10 long-range contact -prediction accuracy 0.69, better than our deep model trained by only MPs (0.63) -and much better than a representative DCA method CCMpred (0.47) and the CASP11 -winner MetaPSICOV (0.55). The accuracy of our deep model can be further -improved to 0.72 when trained by a mix of non-MPs and MPs. When only contacts -in transmembrane regions are evaluated, our method has top L/10 long-range -accuracy 0.62, 0.57, and 0.53 when trained by a mix of non-MPs and MPs, by -non-MPs only, and by MPs only, respectively, still much better than MetaPSICOV -(0.45) and CCMpred (0.40). All these results suggest that sequence-structure -relationship learned by our deep model from non-MPs generalizes well to MP -contact prediction. Improved contact prediction also leads to better -contact-assisted folding. Using only top predicted contacts as restraints, our -deep learning method can fold 160 and 200 of 510 MPs with TMscore>0.6 when -trained by non-MPs only and by a mix of non-MPs and MPs, respectively, while -CCMpred and MetaPSICOV can do so for only 56 and 77 MPs, respectively. Our -contact-assisted folding also greatly outperforms homology modeling.}, - archiveprefix = {arXiv}, - author = {Zhen Li and Sheng Wang and Yizhou Yu and Jinbo Xu}, - eprint = {1704.07207v1}, - file = {1704.07207v1.pdf}, - month = {Apr}, - primaryclass = {q-bio.BM}, - title = {Predicting membrane protein contacts from non-membrane proteins by deep -transfer learning}, - url = {https://arxiv.org/abs/1704.07207v1}, - year = {2017} -} - - -@article{PGi9g7yV, - abstract = {The chest X-ray is one of the most commonly accessible radiological -examinations for screening and diagnosis of many lung diseases. A tremendous -number of X-ray imaging studies accompanied by radiological reports are -accumulated and stored in many modern hospitals' Picture Archiving and -Communication Systems (PACS). On the other side, it is still an open question -how this type of hospital-size knowledge database containing invaluable imaging -informatics (i.e., loosely labeled) can be used to facilitate the data-hungry -deep learning paradigms in building truly large-scale high precision -computer-aided diagnosis (CAD) systems. -In this paper, we present a new chest X-ray database, namely "ChestX-ray8", which comprises 108,948 frontal-view X-ray images of 32,717 unique patients -with the text-mined eight disease image labels (where each image can have -multi-labels), from the associated radiological reports using natural language -processing. Importantly, we demonstrate that these commonly occurring thoracic -diseases can be detected and even spatially-located via a unified -weakly-supervised multi-label image classification and disease localization -framework, which is validated using our proposed dataset. Although the initial -quantitative results are promising as reported, deep convolutional neural -network based "reading chest X-rays" (i.e., recognizing and locating the common -disease patterns trained with only image-level labels) remains a strenuous task -for fully-automated high precision CAD systems. Data download link: -https://nihcc.app.box.com/v/ChestXray-NIHCC}, - archiveprefix = {arXiv}, - author = {Xiaosong Wang and Yifan Peng and Le Lu and Zhiyong Lu and Mohammadhadi Bagheri and Ronald M. Summers}, - eprint = {1705.02315v4}, - file = {1705.02315v4.pdf}, - month = {May}, - primaryclass = {cs.CV}, - title = {ChestX-ray8: Hospital-scale Chest X-ray Database and Benchmarks on -Weakly-Supervised Classification and Localization of Common Thorax Diseases}, - url = {https://arxiv.org/abs/1705.02315v4}, - year = {2017} -} - - -@article{haHzVaaz, - abstract = {Neural machine translation is a recently proposed approach to machine -translation. Unlike the traditional statistical machine translation, the neural -machine translation aims at building a single neural network that can be -jointly tuned to maximize the translation performance. The models proposed -recently for neural machine translation often belong to a family of -encoder-decoders and consists of an encoder that encodes a source sentence into -a fixed-length vector from which a decoder generates a translation. In this -paper, we conjecture that the use of a fixed-length vector is a bottleneck in -improving the performance of this basic encoder-decoder architecture, and -propose to extend this by allowing a model to automatically (soft-)search for -parts of a source sentence that are relevant to predicting a target word, without having to form these parts as a hard segment explicitly. With this new -approach, we achieve a translation performance comparable to the existing -state-of-the-art phrase-based system on the task of English-to-French -translation. Furthermore, qualitative analysis reveals that the -(soft-)alignments found by the model agree well with our intuition.}, - archiveprefix = {arXiv}, - author = {Dzmitry Bahdanau and Kyunghyun Cho and Yoshua Bengio}, - eprint = {1409.0473v7}, - file = {1409.0473v7.pdf}, - month = {Sep}, - primaryclass = {cs.CL}, - title = {Neural Machine Translation by Jointly Learning to Align and Translate}, - url = {https://arxiv.org/abs/1409.0473v7}, - year = {2014} -} - - -@article{1G3owNNps, - abstract = {Multipliers are the most space and power-hungry arithmetic operators of the -digital implementation of deep neural networks. We train a set of -state-of-the-art neural networks (Maxout networks) on three benchmark datasets: -MNIST, CIFAR-10 and SVHN. They are trained with three distinct formats: -floating point, fixed point and dynamic fixed point. For each of those datasets -and for each of those formats, we assess the impact of the precision of the -multiplications on the final error after training. We find that very low -precision is sufficient not just for running trained networks but also for -training them. For example, it is possible to train Maxout networks with 10 -bits multiplications.}, - archiveprefix = {arXiv}, - author = {Matthieu Courbariaux and Yoshua Bengio and Jean-Pierre David}, - eprint = {1412.7024v5}, - file = {1412.7024v5.pdf}, - month = {12}, - primaryclass = {cs.LG}, - title = {Training deep neural networks with low precision multiplications}, - url = {https://arxiv.org/abs/1412.7024v5}, - year = {2014} -} - - -@article{1BTJ1KqRa, - abstract = {Motivation: The MinION device by Oxford Nanopore is the first portable -sequencing device. MinION is able to produce very long reads (reads over -100~kBp were reported), however it suffers from high sequencing error rate. In -this paper, we show that the error rate can be reduced by improving the base -calling process. -Results: We present the first open-source DNA base caller for the MinION -sequencing platform by Oxford Nanopore. By employing carefully crafted -recurrent neural networks, our tool improves the base calling accuracy compared -to the default base caller supplied by the manufacturer. This advance may -further enhance applicability of MinION for genome sequencing and various -clinical applications. -Availability: DeepNano can be downloaded at -http://compbio.fmph.uniba.sk/deepnano/. -Contact: boza@fmph.uniba.sk}, - archiveprefix = {arXiv}, - author = {Vladimír Boža and Broňa Brejová and Tomáš Vinař}, - doi = {10.1371/journal.pone.0178751}, - eprint = {1603.09195v1}, - file = {1603.09195v1.pdf}, - month = {Mar}, - note = {PLoS ONE 12(6): e0178751}, - primaryclass = {q-bio.GN}, - title = {DeepNano: Deep Recurrent Neural Networks for Base Calling in MinION -Nanopore Reads}, - url = {https://arxiv.org/abs/1603.09195v1}, - year = {2016} -} - - -@article{1AhGoHZP9, - abstract = {Currently, deep neural networks are the state of the art on problems such as -speech recognition and computer vision. In this extended abstract, we show that -shallow feed-forward networks can learn the complex functions previously -learned by deep nets and achieve accuracies previously only achievable with -deep models. Moreover, in some cases the shallow neural nets can learn these -deep functions using a total number of parameters similar to the original deep -model. We evaluate our method on the TIMIT phoneme recognition task and are -able to train shallow fully-connected nets that perform similarly to complex, well-engineered, deep convolutional architectures. Our success in training -shallow neural nets to mimic deeper models suggests that there probably exist -better algorithms for training shallow feed-forward nets than those currently -available.}, - archiveprefix = {arXiv}, - author = {Lei Jimmy Ba and Rich Caruana}, - eprint = {1312.6184v7}, - file = {1312.6184v7.pdf}, - month = {12}, - primaryclass = {cs.LG}, - title = {Do Deep Nets Really Need to be Deep?}, - url = {https://arxiv.org/abs/1312.6184v7}, - year = {2013} -} - - -@article{14DAmZTDg, - abstract = {Exponential growth in Electronic Healthcare Records (EHR) has resulted in new -opportunities and urgent needs for discovery of meaningful data-driven -representations and patterns of diseases in Computational Phenotyping research. -Deep Learning models have shown superior performance for robust prediction in -computational phenotyping tasks, but suffer from the issue of model -interpretability which is crucial for clinicians involved in decision-making. -In this paper, we introduce a novel knowledge-distillation approach called -Interpretable Mimic Learning, to learn interpretable phenotype features for -making robust prediction while mimicking the performance of deep learning -models. Our framework uses Gradient Boosting Trees to learn interpretable -features from deep learning models such as Stacked Denoising Autoencoder and -Long Short-Term Memory. Exhaustive experiments on a real-world clinical -time-series dataset show that our method obtains similar or better performance -than the deep learning models, and it provides interpretable phenotypes for -clinical decision making.}, - archiveprefix = {arXiv}, - author = {Zhengping Che and Sanjay Purushotham and Robinder Khemani and Yan Liu}, - eprint = {1512.03542v1}, - file = {1512.03542v1.pdf}, - month = {12}, - primaryclass = {stat.ML}, - title = {Distilling Knowledge from Deep Networks with Applications to Healthcare -Domain}, - url = {https://arxiv.org/abs/1512.03542v1}, - year = {2015} -} - - -@article{O7Vbecm2, - abstract = {Multivariate time series data in practical applications, such as health care, geoscience, and biology, are characterized by a variety of missing values. In -time series prediction and other related tasks, it has been noted that missing -values and their missing patterns are often correlated with the target labels, a.k.a., informative missingness. There is very limited work on exploiting the -missing patterns for effective imputation and improving prediction performance. -In this paper, we develop novel deep learning models, namely GRU-D, as one of -the early attempts. GRU-D is based on Gated Recurrent Unit (GRU), a -state-of-the-art recurrent neural network. It takes two representations of -missing patterns, i.e., masking and time interval, and effectively incorporates -them into a deep model architecture so that it not only captures the long-term -temporal dependencies in time series, but also utilizes the missing patterns to -achieve better prediction results. Experiments of time series classification -tasks on real-world clinical datasets (MIMIC-III, PhysioNet) and synthetic -datasets demonstrate that our models achieve state-of-the-art performance and -provides useful insights for better understanding and utilization of missing -values in time series analysis.}, - archiveprefix = {arXiv}, - author = {Zhengping Che and Sanjay Purushotham and Kyunghyun Cho and David Sontag and Yan Liu}, - eprint = {1606.01865v2}, - file = {1606.01865v2.pdf}, - month = {Jun}, - primaryclass = {cs.LG}, - title = {Recurrent Neural Networks for Multivariate Time Series with Missing -Values}, - url = {https://arxiv.org/abs/1606.01865v2}, - year = {2016} -} - - -@article{15lYGmZpY, - abstract = {As deep nets are increasingly used in applications suited for mobile devices, a fundamental dilemma becomes apparent: the trend in deep learning is to grow -models to absorb ever-increasing data set sizes; however mobile devices are -designed with very little memory and cannot store such large models. We present -a novel network architecture, HashedNets, that exploits inherent redundancy in -neural networks to achieve drastic reductions in model sizes. HashedNets uses a -low-cost hash function to randomly group connection weights into hash buckets, and all connections within the same hash bucket share a single parameter value. -These parameters are tuned to adjust to the HashedNets weight sharing -architecture with standard backprop during training. Our hashing procedure -introduces no additional memory overhead, and we demonstrate on several -benchmark data sets that HashedNets shrink the storage requirements of neural -networks substantially while mostly preserving generalization performance.}, - archiveprefix = {arXiv}, - author = {Wenlin Chen and James T. Wilson and Stephen Tyree and Kilian Q. Weinberger and Yixin Chen}, - eprint = {1504.04788v1}, - file = {1504.04788v1.pdf}, - month = {Apr}, - primaryclass = {cs.LG}, - title = {Compressing Neural Networks with the Hashing Trick}, - url = {https://arxiv.org/abs/1504.04788v1}, - year = {2015} -} - - -@article{10nDTiETi, - abstract = {Deep learning methods exhibit promising performance for predictive modeling -in healthcare, but two important challenges remain: -Data insufficiency:Often -in healthcare predictive modeling, the sample size is insufficient for deep -learning methods to achieve satisfactory results. -Interpretation:The -representations learned by deep learning methods should align with medical -knowledge. To address these challenges, we propose a GRaph-based Attention -Model, GRAM that supplements electronic health records (EHR) with hierarchical -information inherent to medical ontologies. Based on the data volume and the -ontology structure, GRAM represents a medical concept as a combination of its -ancestors in the ontology via an attention mechanism. We compared predictive -performance (i.e. accuracy, data needs, interpretability) of GRAM to various -methods including the recurrent neural network (RNN) in two sequential -diagnoses prediction tasks and one heart failure prediction task. Compared to -the basic RNN, GRAM achieved 10% higher accuracy for predicting diseases rarely -observed in the training data and 3% improved area under the ROC curve for -predicting heart failure using an order of magnitude less training data. -Additionally, unlike other methods, the medical concept representations learned -by GRAM are well aligned with the medical ontology. Finally, GRAM exhibits -intuitive attention behaviors by adaptively generalizing to higher level -concepts when facing data insufficiency at the lower level concepts.}, - archiveprefix = {arXiv}, - author = {Edward Choi and Mohammad Taha Bahadori and Le Song and Walter F. Stewart and Jimeng Sun}, - eprint = {1611.07012v3}, - file = {1611.07012v3.pdf}, - month = {Dec}, - primaryclass = {cs.LG}, - title = {GRAM: Graph-based Attention Model for Healthcare Representation Learning}, - url = {https://arxiv.org/abs/1611.07012v3}, - year = {2016} -} - - -@article{UcRbawKo, - abstract = {Accuracy and interpretability are two dominant features of successful -predictive models. Typically, a choice must be made in favor of complex black -box models such as recurrent neural networks (RNN) for accuracy versus less -accurate but more interpretable traditional models such as logistic regression. -This tradeoff poses challenges in medicine where both accuracy and -interpretability are important. We addressed this challenge by developing the -REverse Time AttentIoN model (RETAIN) for application to Electronic Health -Records (EHR) data. RETAIN achieves high accuracy while remaining clinically -interpretable and is based on a two-level neural attention model that detects -influential past visits and significant clinical variables within those visits -(e.g. key diagnoses). RETAIN mimics physician practice by attending the EHR -data in a reverse time order so that recent clinical visits are likely to -receive higher attention. RETAIN was tested on a large health system EHR -dataset with 14 million visits completed by 263K patients over an 8 year period -and demonstrated predictive accuracy and computational scalability comparable -to state-of-the-art methods such as RNN, and ease of interpretability -comparable to traditional models.}, - archiveprefix = {arXiv}, - author = {Edward Choi and Mohammad Taha Bahadori and Joshua A. Kulas and Andy Schuetz and Walter F. Stewart and Jimeng Sun}, - eprint = {1608.05745v4}, - file = {1608.05745v4.pdf}, - month = {Aug}, - primaryclass = {cs.LG}, - title = {RETAIN: An Interpretable Predictive Model for Healthcare using Reverse -Time Attention Mechanism}, - url = {https://arxiv.org/abs/1608.05745v4}, - year = {2016} -} - - -@article{VMkPJjVk, - abstract = {We present an interpretation of Inception modules in convolutional neural -networks as being an intermediate step in-between regular convolution and the -depthwise separable convolution operation (a depthwise convolution followed by -a pointwise convolution). In this light, a depthwise separable convolution can -be understood as an Inception module with a maximally large number of towers. -This observation leads us to propose a novel deep convolutional neural network -architecture inspired by Inception, where Inception modules have been replaced -with depthwise separable convolutions. We show that this architecture, dubbed -Xception, slightly outperforms Inception V3 on the ImageNet dataset (which -Inception V3 was designed for), and significantly outperforms Inception V3 on a -larger image classification dataset comprising 350 million images and 17,000 -classes. Since the Xception architecture has the same number of parameters as -Inception V3, the performance gains are not due to increased capacity but -rather to a more efficient use of model parameters.}, - archiveprefix = {arXiv}, - author = {François Chollet}, - eprint = {1610.02357v3}, - file = {1610.02357v3.pdf}, - month = {Nov}, - primaryclass = {cs.CV}, - title = {Xception: Deep Learning with Depthwise Separable Convolutions}, - url = {https://arxiv.org/abs/1610.02357v3}, - year = {2016} -} - - -@article{sLPsrfbl, - abstract = {Melanoma is the deadliest form of skin cancer. While curable with early -detection, only highly trained specialists are capable of accurately -recognizing the disease. As expertise is in limited supply, automated systems -capable of identifying disease could save lives, reduce unnecessary biopsies, and reduce costs. Toward this goal, we propose a system that combines recent -developments in deep learning with established machine learning approaches, creating ensembles of methods that are capable of segmenting skin lesions, as -well as analyzing the detected area and surrounding tissue for melanoma -detection. The system is evaluated using the largest publicly available -benchmark dataset of dermoscopic images, containing 900 training and 379 -testing images. New state-of-the-art performance levels are demonstrated, leading to an improvement in the area under receiver operating characteristic -curve of 7.5% (0.843 vs. 0.783), in average precision of 4% (0.649 vs. 0.624), and in specificity measured at the clinically relevant 95% sensitivity -operating point 2.9 times higher than the previous state-of-the-art (36.8% -specificity compared to 12.5%). Compared to the average of 8 expert -dermatologists on a subset of 100 test images, the proposed system produces a -higher accuracy (76% vs. 70.5%), and specificity (62% vs. 59%) evaluated at an -equivalent sensitivity (82%).}, - archiveprefix = {arXiv}, - author = {Noel Codella and Quoc-Bao Nguyen and Sharath Pankanti and David Gutman and Brian Helba and Allan Halpern and John R. Smith}, - eprint = {1610.04662v2}, - file = {1610.04662v2.pdf}, - month = {Nov}, - note = {IBM Journal of Research and Development, vol. 61, no. 4/5, 2017}, - primaryclass = {cs.CV}, - title = {Deep Learning Ensembles for Melanoma Recognition in Dermoscopy Images}, - url = {https://arxiv.org/abs/1610.04662v2}, - year = {2016} -} - - -@article{YwdqeYZi, - abstract = {We present a library of efficient implementations of deep learning -primitives. Deep learning workloads are computationally intensive, and -optimizing their kernels is difficult and time-consuming. As parallel -architectures evolve, kernels must be reoptimized, which makes maintaining -codebases difficult over time. Similar issues have long been addressed in the -HPC community by libraries such as the Basic Linear Algebra Subroutines (BLAS). -However, there is no analogous library for deep learning. Without such a -library, researchers implementing deep learning workloads on parallel -processors must create and optimize their own implementations of the main -computational kernels, and this work must be repeated as new parallel -processors emerge. To address this problem, we have created a library similar -in intent to BLAS, with optimized routines for deep learning workloads. Our -implementation contains routines for GPUs, although similarly to the BLAS -library, these routines could be implemented for other platforms. The library -is easy to integrate into existing frameworks, and provides optimized -performance and memory usage. For example, integrating cuDNN into Caffe, a -popular framework for convolutional networks, improves performance by 36% on a -standard model while also reducing memory consumption.}, - archiveprefix = {arXiv}, - author = {Sharan Chetlur and Cliff Woolley and Philippe Vandermersch and Jonathan Cohen and John Tran and Bryan Catanzaro and Evan Shelhamer}, - eprint = {1410.0759v3}, - file = {1410.0759v3.pdf}, - month = {Nov}, - primaryclass = {cs.NE}, - title = {cuDNN: Efficient Primitives for Deep Learning}, - url = {https://arxiv.org/abs/1410.0759v3}, - year = {2014} -} - - -@article{1Dzz0P0qr, - abstract = {Although artificial neural networks have occasionally been used for -Quantitative Structure-Activity/Property Relationship (QSAR/QSPR) studies in -the past, the literature has of late been dominated by other machine learning -techniques such as random forests. However, a variety of new neural net -techniques along with successful applications in other domains have renewed -interest in network approaches. In this work, inspired by the winning team's -use of neural networks in a recent QSAR competition, we used an artificial -neural network to learn a function that predicts activities of compounds for -multiple assays at the same time. We conducted experiments leveraging recent -methods for dealing with overfitting in neural networks as well as other tricks -from the neural networks literature. We compared our methods to alternative -methods reported to perform well on these tasks and found that our neural net -methods provided superior performance.}, - archiveprefix = {arXiv}, - author = {George E. Dahl and Navdeep Jaitly and Ruslan Salakhutdinov}, - eprint = {1406.1231v1}, - file = {1406.1231v1.pdf}, - month = {Jun}, - primaryclass = {stat.ML}, - title = {Multi-task Neural Networks for QSAR Predictions}, - url = {https://arxiv.org/abs/1406.1231v1}, - year = {2014} -} - - -@article{SAvEOARL, - abstract = {Each human genome is a 3 billion base pair set of encoding instructions. -Decoding the genome using deep learning fundamentally differs from most tasks, as we do not know the full structure of the data and therefore cannot design -architectures to suit it. As such, architectures that fit the structure of -genomics should be learned not prescribed. Here, we develop a novel search -algorithm, applicable across domains, that discovers an optimal architecture -which simultaneously learns general genomic patterns and identifies the most -important sequence motifs in predicting functional genomic outcomes. The -architectures we find using this algorithm succeed at using only RNA expression -data to predict gene regulatory structure, learn human-interpretable -visualizations of key sequence motifs, and surpass state-of-the-art results on -benchmark genomics challenges.}, - archiveprefix = {arXiv}, - author = {Laura Deming and Sasha Targ and Nate Sauder and Diogo Almeida and Chun Jimmie Ye}, - eprint = {1605.07156v1}, - file = {1605.07156v1.pdf}, - month = {May}, - primaryclass = {cs.LG}, - title = {Genetic Architect: Discovering Genomic Structure with Learned Neural -Architectures}, - url = {https://arxiv.org/abs/1605.07156v1}, - year = {2016} -} - - -@article{y4t9EzPn, - abstract = {As machine learning algorithms are increasingly applied to high impact yet -high risk tasks, e.g. problems in health, it is critical that researchers can -explain how such algorithms arrived at their predictions. In recent years, a -number of image saliency methods have been developed to summarize where highly -complex neural networks "look" in an image for evidence for their predictions. -However, these techniques are limited by their heuristic nature and -architectural constraints. -In this paper, we make two main contributions: First, we propose a general -framework for learning different kinds of explanations for any black box -algorithm. Second, we introduce a paradigm that learns the minimally salient -part of an image by directly editing it and learning from the corresponding -changes to its output. Unlike previous works, our method is model-agnostic and -testable because it is grounded in replicable image perturbations.}, - archiveprefix = {arXiv}, - author = {Ruth Fong and Andrea Vedaldi}, - eprint = {1704.03296v1}, - file = {1704.03296v1.pdf}, - month = {Apr}, - primaryclass = {cs.CV}, - title = {Interpretable Explanations of Black Boxes by Meaningful Perturbation}, - url = {https://arxiv.org/abs/1704.03296v1}, - year = {2017} -} - - -@article{1FDihfnM, - abstract = {Deep learning tools have gained tremendous attention in applied machine -learning. However such tools for regression and classification do not capture -model uncertainty. In comparison, Bayesian models offer a mathematically -grounded framework to reason about model uncertainty, but usually come with a -prohibitive computational cost. In this paper we develop a new theoretical -framework casting dropout training in deep neural networks (NNs) as approximate -Bayesian inference in deep Gaussian processes. A direct result of this theory -gives us tools to model uncertainty with dropout NNs -- extracting information -from existing models that has been thrown away so far. This mitigates the -problem of representing uncertainty in deep learning without sacrificing either -computational complexity or test accuracy. We perform an extensive study of the -properties of dropout's uncertainty. Various network architectures and -non-linearities are assessed on tasks of regression and classification, using -MNIST as an example. We show a considerable improvement in predictive -log-likelihood and RMSE compared to existing state-of-the-art methods, and -finish by using dropout's uncertainty in deep reinforcement learning.}, - archiveprefix = {arXiv}, - author = {Yarin Gal and Zoubin Ghahramani}, - eprint = {1506.02142v6}, - file = {1506.02142v6.pdf}, - month = {Jun}, - primaryclass = {stat.ML}, - title = {Dropout as a Bayesian Approximation: Representing Model Uncertainty in -Deep Learning}, - url = {https://arxiv.org/abs/1506.02142v6}, - year = {2015} -} - - -@article{2dU8f4XJ, - abstract = {We report a method to convert discrete representations of molecules to and -from a multidimensional continuous representation. This generative model allows -efficient search and optimization through open-ended spaces of chemical -compounds. We train deep neural networks on hundreds of thousands of existing -chemical structures to construct two coupled functions: an encoder and a -decoder. The encoder converts the discrete representation of a molecule into a -real-valued continuous vector, and the decoder converts these continuous -vectors back to the discrete representation from this latent space. Continuous -representations allow us to automatically generate novel chemical structures by -performing simple operations in the latent space, such as decoding random -vectors, perturbing known chemical structures, or interpolating between -molecules. Continuous representations also allow the use of powerful -gradient-based optimization to efficiently guide the search for optimized -functional compounds. We demonstrate our method in the design of drug-like -molecules as well as organic light-emitting diodes.}, - archiveprefix = {arXiv}, - author = {Rafael Gómez-Bombarelli and David Duvenaud and José Miguel Hernández-Lobato and Jorge Aguilera-Iparraguirre and Timothy D. Hirzel and Ryan P. Adams and Alán Aspuru-Guzik}, - eprint = {1610.02415v2}, - file = {1610.02415v2.pdf}, - month = {Nov}, - primaryclass = {cs.LG}, - title = {Automatic chemical design using a data-driven continuous representation -of molecules}, - url = {https://arxiv.org/abs/1610.02415v2}, - year = {2016} -} - - -@article{CKcJuj03, - abstract = {Training of large-scale deep neural networks is often constrained by the -available computational resources. We study the effect of limited precision -data representation and computation on neural network training. Within the -context of low-precision fixed-point computations, we observe the rounding -scheme to play a crucial role in determining the network's behavior during -training. Our results show that deep networks can be trained using only 16-bit -wide fixed-point number representation when using stochastic rounding, and -incur little to no degradation in the classification accuracy. We also -demonstrate an energy-efficient hardware accelerator that implements -low-precision fixed-point arithmetic with stochastic rounding.}, - archiveprefix = {arXiv}, - author = {Suyog Gupta and Ankur Agrawal and Kailash Gopalakrishnan and Pritish Narayanan}, - eprint = {1502.02551v1}, - file = {1502.02551v1.pdf}, - month = {Feb}, - primaryclass = {cs.LG}, - title = {Deep Learning with Limited Numerical Precision}, - url = {https://arxiv.org/abs/1502.02551v1}, - year = {2015} -} - - -@article{13KjSCKB2, - abstract = {We present Caffe con Troll (CcT), a fully compatible end-to-end version of -the popular framework Caffe with rebuilt internals. We built CcT to examine the -performance characteristics of training and deploying general-purpose -convolutional neural networks across different hardware architectures. We find -that, by employing standard batching optimizations for CPU training, we achieve -a 4.5x throughput improvement over Caffe on popular networks like CaffeNet. -Moreover, with these improvements, the end-to-end training time for CNNs is -directly proportional to the FLOPS delivered by the CPU, which enables us to -efficiently train hybrid CPU-GPU systems for CNNs.}, - archiveprefix = {arXiv}, - author = {Stefan Hadjis and Firas Abuzaid and Ce Zhang and Christopher Ré}, - eprint = {1504.04343v2}, - file = {1504.04343v2.pdf}, - month = {Apr}, - primaryclass = {cs.LG}, - title = {Caffe con Troll: Shallow Ideas to Speed Up Deep Learning}, - url = {https://arxiv.org/abs/1504.04343v2}, - year = {2015} -} - - -@article{j7KrVyi8, - abstract = {Deeper neural networks are more difficult to train. We present a residual -learning framework to ease the training of networks that are substantially -deeper than those used previously. We explicitly reformulate the layers as -learning residual functions with reference to the layer inputs, instead of -learning unreferenced functions. We provide comprehensive empirical evidence -showing that these residual networks are easier to optimize, and can gain -accuracy from considerably increased depth. On the ImageNet dataset we evaluate -residual nets with a depth of up to 152 layers---8x deeper than VGG nets but -still having lower complexity. An ensemble of these residual nets achieves -3.57% error on the ImageNet test set. This result won the 1st place on the -ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 -and 1000 layers. -The depth of representations is of central importance for many visual -recognition tasks. Solely due to our extremely deep representations, we obtain -a 28% relative improvement on the COCO object detection dataset. Deep residual -nets are foundations of our submissions to ILSVRC & COCO 2015 competitions, where we also won the 1st places on the tasks of ImageNet detection, ImageNet -localization, COCO detection, and COCO segmentation.}, - archiveprefix = {arXiv}, - author = {Kaiming He and Xiangyu Zhang and Shaoqing Ren and Jian Sun}, - eprint = {1512.03385v1}, - file = {1512.03385v1.pdf}, - month = {12}, - primaryclass = {cs.CV}, - title = {Deep Residual Learning for Image Recognition}, - url = {https://arxiv.org/abs/1512.03385v1}, - year = {2015} -} - - -@article{1CRF3gAV, - abstract = {A very simple way to improve the performance of almost any machine learning -algorithm is to train many different models on the same data and then to -average their predictions. Unfortunately, making predictions using a whole -ensemble of models is cumbersome and may be too computationally expensive to -allow deployment to a large number of users, especially if the individual -models are large neural nets. Caruana and his collaborators have shown that it -is possible to compress the knowledge in an ensemble into a single model which -is much easier to deploy and we develop this approach further using a different -compression technique. We achieve some surprising results on MNIST and we show -that we can significantly improve the acoustic model of a heavily used -commercial system by distilling the knowledge in an ensemble of models into a -single model. We also introduce a new type of ensemble composed of one or more -full models and many specialist models which learn to distinguish fine-grained -classes that the full models confuse. Unlike a mixture of experts, these -specialist models can be trained rapidly and in parallel.}, - archiveprefix = {arXiv}, - author = {Geoffrey Hinton and Oriol Vinyals and Jeff Dean}, - eprint = {1503.02531v1}, - file = {1503.02531v1.pdf}, - month = {Mar}, - primaryclass = {stat.ML}, - title = {Distilling the Knowledge in a Neural Network}, - url = {https://arxiv.org/abs/1503.02531v1}, - year = {2015} -} - - -@article{1GUizyE8e, - abstract = {We introduce a method to train Quantized Neural Networks (QNNs) --- neural -networks with extremely low precision (e.g., 1-bit) weights and activations, at -run-time. At train-time the quantized weights and activations are used for -computing the parameter gradients. During the forward pass, QNNs drastically -reduce memory size and accesses, and replace most arithmetic operations with -bit-wise operations. As a result, power consumption is expected to be -drastically reduced. We trained QNNs over the MNIST, CIFAR-10, SVHN and -ImageNet datasets. The resulting QNNs achieve prediction accuracy comparable to -their 32-bit counterparts. For example, our quantized version of AlexNet with -1-bit weights and 2-bit activations achieves $51\%$ top-1 accuracy. Moreover, we quantize the parameter gradients to 6-bits as well which enables gradients -computation using only bit-wise operation. Quantized recurrent neural networks -were tested over the Penn Treebank dataset, and achieved comparable accuracy as -their 32-bit counterparts using only 4-bits. Last but not least, we programmed -a binary matrix multiplication GPU kernel with which it is possible to run our -MNIST QNN 7 times faster than with an unoptimized GPU kernel, without suffering -any loss in classification accuracy. The QNN code is available online.}, - archiveprefix = {arXiv}, - author = {Itay Hubara and Matthieu Courbariaux and Daniel Soudry and Ran El-Yaniv and Yoshua Bengio}, - eprint = {1609.07061v1}, - file = {1609.07061v1.pdf}, - month = {Sep}, - primaryclass = {cs.NE}, - title = {Quantized Neural Networks: Training Neural Networks with Low Precision -Weights and Activations}, - url = {https://arxiv.org/abs/1609.07061v1}, - year = {2016} -} - - -@article{1AJUcl1KV, - abstract = {Melanoma is amongst most aggressive types of cancer. However, it is highly -curable if detected in its early stages. Prescreening of suspicious moles and -lesions for malignancy is of great importance. Detection can be done by images -captured by standard cameras, which are more preferable due to low cost and -availability. One important step in computerized evaluation of skin lesions is -accurate detection of lesion region, i.e. segmentation of an image into two -regions as lesion and normal skin. Accurate segmentation can be challenging due -to burdens such as illumination variation and low contrast between lesion and -healthy skin. In this paper, a method based on deep neural networks is proposed -for accurate extraction of a lesion region. The input image is preprocessed and -then its patches are fed to a convolutional neural network (CNN). Local texture -and global structure of the patches are processed in order to assign pixels to -lesion or normal classes. A method for effective selection of training patches -is used for more accurate detection of a lesion border. The output segmentation -mask is refined by some post processing operations. The experimental results of -qualitative and quantitative evaluations demonstrate that our method can -outperform other state-of-the-art algorithms exist in the literature.}, - archiveprefix = {arXiv}, - author = {Mohammad H. Jafari and Ebrahim Nasr-Esfahani and Nader Karimi and S. M. Reza Soroushmehr and Shadrokh Samavi and Kayvan Najarian}, - doi = {10.1007/s11548-017-1567-8}, - eprint = {1609.02374v1}, - file = {1609.02374v1.pdf}, - month = {Sep}, - primaryclass = {cs.CV}, - title = {Extraction of Skin Lesions from Non-Dermoscopic Images Using Deep -Learning}, - url = {https://arxiv.org/abs/1609.02374v1}, - year = {2016} -} - - -@article{71c6rs2z, - abstract = {We present a conditional generative model to learn variation in cell and -nuclear morphology and the location of subcellular structures from microscopy -images. Our model generalizes to a wide range of subcellular localization and -allows for a probabilistic interpretation of cell and nuclear morphology and -structure localization from fluorescence images. We demonstrate the -effectiveness of our approach by producing photo-realistic cell images using -our generative model. The conditional nature of the model provides the ability -to predict the localization of unobserved structures given cell and nuclear -morphology.}, - archiveprefix = {arXiv}, - author = {Gregory R. Johnson and Rory M. Donovan-Maiye and Mary M. Maleckar}, - eprint = {1705.00092v1}, - file = {1705.00092v1.pdf}, - month = {Apr}, - primaryclass = {stat.ML}, - title = {Generative Modeling with Conditional Autoencoders: Building an -Integrated Cell}, - url = {https://arxiv.org/abs/1705.00092v1}, - year = {2017} -} - - -@article{QphVo2P2, - abstract = {While deep learning models have achieved state-of-the-art accuracies for many -prediction tasks, understanding these models remains a challenge. Despite the -recent interest in developing visual tools to help users interpret deep -learning models, the complexity and wide variety of models deployed in -industry, and the large-scale datasets that they used, pose unique design -challenges that are inadequately addressed by existing work. Through -participatory design sessions with over 15 researchers and engineers at -Facebook, we have developed, deployed, and iteratively improved ActiVis, an -interactive visualization system for interpreting large-scale deep learning -models and results. By tightly integrating multiple coordinated views, such as -a computation graph overview of the model architecture, and a neuron activation -view for pattern discovery and comparison, users can explore complex deep -neural network models at both the instance- and subset-level. ActiVis has been -deployed on Facebook's machine learning platform. We present case studies with -Facebook researchers and engineers, and usage scenarios of how ActiVis may work -with different models.}, - archiveprefix = {arXiv}, - author = {Minsuk Kahng and Pierre Y. Andrews and Aditya Kalro and Duen Horng Chau}, - eprint = {1704.01942v2}, - file = {1704.01942v2.pdf}, - month = {Apr}, - primaryclass = {cs.HC}, - title = {ActiVis: Visual Exploration of Industry-Scale Deep Neural Network Models}, - url = {https://arxiv.org/abs/1704.01942v2}, - year = {2017} -} - - -@article{2cpYveR4, - abstract = {Recurrent Neural Networks (RNNs), and specifically a variant with Long -Short-Term Memory (LSTM), are enjoying renewed interest as a result of -successful applications in a wide range of machine learning problems that -involve sequential data. However, while LSTMs provide exceptional results in -practice, the source of their performance and their limitations remain rather -poorly understood. Using character-level language models as an interpretable -testbed, we aim to bridge this gap by providing an analysis of their -representations, predictions and error types. In particular, our experiments -reveal the existence of interpretable cells that keep track of long-range -dependencies such as line lengths, quotes and brackets. Moreover, our -comparative analysis with finite horizon n-gram models traces the source of the -LSTM improvements to long-range structural dependencies. Finally, we provide -analysis of the remaining errors and suggests areas for further study.}, - archiveprefix = {arXiv}, - author = {Andrej Karpathy and Justin Johnson and Li Fei-Fei}, - eprint = {1506.02078v2}, - file = {1506.02078v2.pdf}, - month = {Jun}, - primaryclass = {cs.LG}, - title = {Visualizing and Understanding Recurrent Networks}, - url = {https://arxiv.org/abs/1506.02078v2}, - year = {2015} -} - - -@article{uP7SgBVd, - abstract = {Deep learning methods such as multitask neural networks have recently been -applied to ligand-based virtual screening and other drug discovery -applications. Using a set of industrial ADMET datasets, we compare neural -networks to standard baseline models and analyze multitask learning effects -with both random cross-validation and a more relevant temporal validation -scheme. We confirm that multitask learning can provide modest benefits over -single-task models and show that smaller datasets tend to benefit more than -larger datasets from multitask learning. Additionally, we find that adding -massive amounts of side information is not guaranteed to improve performance -relative to simpler multitask learning. Our results emphasize that multitask -effects are highly dataset-dependent, suggesting the use of dataset-specific -models to maximize overall performance.}, - archiveprefix = {arXiv}, - author = {Steven Kearnes and Brian Goldman and Vijay Pande}, - eprint = {1606.08793v3}, - file = {1606.08793v3.pdf}, - month = {Jun}, - primaryclass = {stat.ML}, - title = {Modeling Industrial ADMET Data with Multitask Networks}, - url = {https://arxiv.org/abs/1606.08793v3}, - year = {2016} -} - - -@article{b1sc0cgP, - abstract = {Understanding neural networks is becoming increasingly important. Over the -last few years different types of visualisation and explanation methods have -been proposed. However, none of them explicitly considered the behaviour in the -presence of noise and distracting elements. In this work, we will show how -noise and distracting dimensions can influence the result of an explanation -model. This gives a new theoretical insights to aid selection of the most -appropriate explanation model within the deep-Taylor decomposition framework.}, - archiveprefix = {arXiv}, - author = {Pieter-Jan Kindermans and Kristof Schütt and Klaus-Robert Müller and Sven Dähne}, - eprint = {1611.07270v1}, - file = {1611.07270v1.pdf}, - month = {Dec}, - primaryclass = {stat.ML}, - title = {Investigating the influence of noise and distractors on the -interpretation of neural networks}, - url = {https://arxiv.org/abs/1611.07270v1}, - year = {2016} -} - - -@article{69wxD9y, - abstract = {How can we explain the predictions of a black-box model? In this paper, we -use influence functions -- a classic technique from robust statistics -- to -trace a model's prediction through the learning algorithm and back to its -training data, thereby identifying training points most responsible for a given -prediction. To scale up influence functions to modern machine learning -settings, we develop a simple, efficient implementation that requires only -oracle access to gradients and Hessian-vector products. We show that even on -non-convex and non-differentiable models where the theory breaks down, approximations to influence functions can still provide valuable information. -On linear models and convolutional neural networks, we demonstrate that -influence functions are useful for multiple purposes: understanding model -behavior, debugging models, detecting dataset errors, and even creating -visually-indistinguishable training-set attacks.}, - archiveprefix = {arXiv}, - author = {Pang Wei Koh and Percy Liang}, - eprint = {1703.04730v2}, - file = {1703.04730v2.pdf}, - month = {Mar}, - primaryclass = {stat.ML}, - title = {Understanding Black-box Predictions via Influence Functions}, - url = {https://arxiv.org/abs/1703.04730v2}, - year = {2017} -} - - -@article{ZSVsnPVO, - abstract = {I present a new way to parallelize the training of convolutional neural -networks across multiple GPUs. The method scales significantly better than all -alternatives when applied to modern convolutional neural networks.}, - archiveprefix = {arXiv}, - author = {Alex Krizhevsky}, - eprint = {1404.5997v2}, - file = {1404.5997v2.pdf}, - month = {Apr}, - primaryclass = {cs.NE}, - title = {One weird trick for parallelizing convolutional neural networks}, - url = {https://arxiv.org/abs/1404.5997v2}, - year = {2014} -} - - -@article{9NKsJjSw, - abstract = {The rapid growth of data size and accessibility in recent years has -instigated a shift of philosophy in algorithm design for artificial -intelligence. Instead of engineering algorithms by hand, the ability to learn -composable systems automatically from massive amounts of data has led to -ground-breaking performance in important domains such as computer vision, speech recognition, and natural language processing. The most popular class of -techniques used in these domains is called deep learning, and is seeing -significant attention from industry. However, these models require incredible -amounts of data and compute power to train, and are limited by the need for -better hardware acceleration to accommodate scaling beyond current data and -model sizes. While the current solution has been to use clusters of graphics -processing units (GPU) as general purpose processors (GPGPU), the use of field -programmable gate arrays (FPGA) provide an interesting alternative. Current -trends in design tools for FPGAs have made them more compatible with the -high-level software practices typically practiced in the deep learning -community, making FPGAs more accessible to those who build and deploy models. -Since FPGA architectures are flexible, this could also allow researchers the -ability to explore model-level optimizations beyond what is possible on fixed -architectures such as GPUs. As well, FPGAs tend to provide high performance per -watt of power consumption, which is of particular importance for application -scientists interested in large scale server-based deployment or -resource-limited embedded applications. This review takes a look at deep -learning and FPGAs from a hardware acceleration perspective, identifying trends -and innovations that make these technologies a natural fit, and motivates a -discussion on how FPGAs may best serve the needs of the deep learning community -moving forward.}, - archiveprefix = {arXiv}, - author = {Griffin Lacey and Graham W. Taylor and Shawki Areibi}, - eprint = {1602.04283v1}, - file = {1602.04283v1.pdf}, - month = {Feb}, - primaryclass = {cs.DC}, - title = {Deep Learning on FPGAs: Past, Present, and Future}, - url = {https://arxiv.org/abs/1602.04283v1}, - year = {2016} -} - - -@article{Dwi2eAvT, - abstract = {Deep neural network (DNN) models have recently obtained state-of-the-art -prediction accuracy for the transcription factor binding (TFBS) site -classification task. However, it remains unclear how these approaches identify -meaningful DNA sequence signals and give insights as to why TFs bind to certain -locations. In this paper, we propose a toolkit called the Deep Motif Dashboard -(DeMo Dashboard) which provides a suite of visualization strategies to extract -motifs, or sequence patterns from deep neural network models for TFBS -classification. We demonstrate how to visualize and understand three important -DNN models: convolutional, recurrent, and convolutional-recurrent networks. Our -first visualization method is finding a test sequence's saliency map which uses -first-order derivatives to describe the importance of each nucleotide in making -the final prediction. Second, considering recurrent models make predictions in -a temporal manner (from one end of a TFBS sequence to the other), we introduce -temporal output scores, indicating the prediction score of a model over time -for a sequential input. Lastly, a class-specific visualization strategy finds -the optimal input sequence for a given TFBS positive class via stochastic -gradient optimization. Our experimental results indicate that a -convolutional-recurrent architecture performs the best among the three -architectures. The visualization techniques indicate that CNN-RNN makes -predictions by modeling both motifs as well as dependencies among them.}, - archiveprefix = {arXiv}, - author = {Jack Lanchantin and Ritambhara Singh and Beilun Wang and Yanjun Qi}, - eprint = {1608.03644v4}, - file = {1608.03644v4.pdf}, - month = {Aug}, - primaryclass = {cs.LG}, - title = {Deep Motif Dashboard: Visualizing and Understanding Genomic Sequences -Using Deep Neural Networks}, - url = {https://arxiv.org/abs/1608.03644v4}, - year = {2016} -} - - -@article{1GwC1ll6h, - abstract = {MicroRNAs (miRNAs) are short sequences of ribonucleic acids that control the -expression of target messenger RNAs (mRNAs) by binding them. Robust prediction -of miRNA-mRNA pairs is of utmost importance in deciphering gene regulations but -has been challenging because of high false positive rates, despite a deluge of -computational tools that normally require laborious manual feature extraction. -This paper presents an end-to-end machine learning framework for miRNA target -prediction. Leveraged by deep recurrent neural networks-based auto-encoding and -sequence-sequence interaction learning, our approach not only delivers an -unprecedented level of accuracy but also eliminates the need for manual feature -extraction. The performance gap between the proposed method and existing -alternatives is substantial (over 25% increase in F-measure), and deepTarget -delivers a quantum leap in the long-standing challenge of robust miRNA target -prediction.}, - archiveprefix = {arXiv}, - author = {Byunghan Lee and Junghwan Baek and Seunghyun Park and Sungroh Yoon}, - eprint = {1603.09123v2}, - file = {1603.09123v2.pdf}, - month = {Mar}, - primaryclass = {cs.LG}, - title = {deepTarget: End-to-end Learning Framework for microRNA Target Prediction -using Deep Recurrent Neural Networks}, - url = {https://arxiv.org/abs/1603.09123v2}, - year = {2016} -} - - -@article{ZUCVI5eU, - abstract = {Prediction without justification has limited applicability. As a remedy, we -learn to extract pieces of input text as justifications -- rationales -- that -are tailored to be short and coherent, yet sufficient for making the same -prediction. Our approach combines two modular components, generator and -encoder, which are trained to operate well together. The generator specifies a -distribution over text fragments as candidate rationales and these are passed -through the encoder for prediction. Rationales are never given during training. -Instead, the model is regularized by desiderata for rationales. We evaluate the -approach on multi-aspect sentiment analysis against manually annotated test -cases. Our approach outperforms attention-based baseline by a significant -margin. We also successfully illustrate the method on the question retrieval -task.}, - archiveprefix = {arXiv}, - author = {Tao Lei and Regina Barzilay and Tommi Jaakkola}, - eprint = {1606.04155v2}, - file = {1606.04155v2.pdf}, - month = {Jun}, - primaryclass = {cs.CL}, - title = {Rationalizing Neural Predictions}, - url = {https://arxiv.org/abs/1606.04155v2}, - year = {2016} -} - - -@article{glyI7H6F, - abstract = {We present a novel application of LSTM recurrent neural networks to -multilabel classification of diagnoses given variable-length time series of -clinical measurements. Our method outperforms a strong baseline on a variety of -metrics.}, - archiveprefix = {arXiv}, - author = {Zachary C. Lipton and David C. Kale and Randall C. Wetzel}, - eprint = {1510.07641v2}, - file = {1510.07641v2.pdf}, - month = {Nov}, - primaryclass = {cs.LG}, - title = {Phenotyping of Clinical Time Series with LSTM Recurrent Neural Networks}, - url = {https://arxiv.org/abs/1510.07641v2}, - year = {2015} -} - - -@article{4zpZxjHR, - abstract = {We demonstrate a simple strategy to cope with missing data in sequential -inputs, addressing the task of multilabel classification of diagnoses given -clinical time series. Collected from the pediatric intensive care unit (PICU) -at Children's Hospital Los Angeles, our data consists of multivariate time -series of observations. The measurements are irregularly spaced, leading to -missingness patterns in temporally discretized sequences. While these artifacts -are typically handled by imputation, we achieve superior predictive performance -by treating the artifacts as features. Unlike linear models, recurrent neural -networks can realize this improvement using only simple binary indicators of -missingness. For linear models, we show an alternative strategy to capture this -signal. Training models on missingness patterns only, we show that for some -diseases, what tests are run can be as predictive as the results themselves.}, - archiveprefix = {arXiv}, - author = {Zachary C. Lipton and David C. Kale and Randall Wetzel}, - eprint = {1606.04130v5}, - file = {1606.04130v5.pdf}, - month = {Jun}, - primaryclass = {cs.LG}, - title = {Modeling Missing Data in Clinical Time Series with RNNs}, - url = {https://arxiv.org/abs/1606.04130v5}, - year = {2016} -} - - -@article{LL5huVs3, - abstract = {Deep learning algorithms, in particular convolutional networks, have rapidly -become a methodology of choice for analyzing medical images. This paper reviews -the major deep learning concepts pertinent to medical image analysis and -summarizes over 300 contributions to the field, most of which appeared in the -last year. We survey the use of deep learning for image classification, object -detection, segmentation, registration, and other tasks and provide concise -overviews of studies per application area. Open challenges and directions for -future research are discussed.}, - archiveprefix = {arXiv}, - author = {Geert Litjens and Thijs Kooi and Babak Ehteshami Bejnordi and Arnaud Arindra Adiyoso Setio and Francesco Ciompi and Mohsen Ghafoorian and Jeroen A. W. M. van der Laak and Bram van Ginneken and Clara I. Sánchez}, - doi = {10.1016/j.media.2017.07.005}, - eprint = {1702.05747v2}, - file = {1702.05747v2.pdf}, - month = {Feb}, - primaryclass = {cs.CV}, - title = {A Survey on Deep Learning in Medical Image Analysis}, - url = {https://arxiv.org/abs/1702.05747v2}, - year = {2017} -} - - -@article{AEc66xxR, - abstract = {Deep convolutional neural networks (CNNs) have achieved breakthrough -performance in many pattern recognition tasks such as image classification. -However, the development of high-quality deep models typically relies on a -substantial amount of trial-and-error, as there is still no clear understanding -of when and why a deep model works. In this paper, we present a visual -analytics approach for better understanding, diagnosing, and refining deep -CNNs. We formulate a deep CNN as a directed acyclic graph. Based on this -formulation, a hybrid visualization is developed to disclose the multiple -facets of each neuron and the interactions between them. In particular, we -introduce a hierarchical rectangle packing algorithm and a matrix reordering -algorithm to show the derived features of a neuron cluster. We also propose a -biclustering-based edge bundling method to reduce visual clutter caused by a -large number of connections between neurons. We evaluated our method on a set -of CNNs and the results are generally favorable.}, - archiveprefix = {arXiv}, - author = {Mengchen Liu and Jiaxin Shi and Zhen Li and Chongxuan Li and Jun Zhu and Shixia Liu}, - eprint = {1604.07043v3}, - file = {1604.07043v3.pdf}, - month = {Apr}, - primaryclass = {cs.CV}, - title = {Towards Better Analysis of Deep Convolutional Neural Networks}, - url = {https://arxiv.org/abs/1604.07043v3}, - year = {2016} -} - - -@article{DeOI1oGf, - abstract = {Understanding why a model made a certain prediction is crucial in many data -science fields. Interpretable predictions engender appropriate trust and -provide insight into how the model may be improved. However, with large modern -datasets the best accuracy is often achieved by complex models even experts -struggle to interpret, which creates a tension between accuracy and -interpretability. Recently, several methods have been proposed for interpreting -predictions from complex models by estimating the importance of input features. -Here, we present how a model-agnostic additive representation of the importance -of input features unifies current methods. This representation is optimal, in -the sense that it is the only set of additive values that satisfies important -properties. We show how we can leverage these properties to create novel visual -explanations of model predictions. The thread of unity that this representation -weaves through the literature indicates that there are common principles to be -learned about the interpretation of model predictions that apply in many -scenarios.}, - archiveprefix = {arXiv}, - author = {Scott Lundberg and Su-In Lee}, - eprint = {1611.07478v3}, - file = {1611.07478v3.pdf}, - month = {Dec}, - primaryclass = {cs.AI}, - title = {An unexpected unity among methods for interpreting model predictions}, - url = {https://arxiv.org/abs/1611.07478v3}, - year = {2016} -} - - -@article{19mGl6pfy, - abstract = {Image representations, from SIFT and Bag of Visual Words to Convolutional -Neural Networks (CNNs), are a crucial component of almost any image -understanding system. Nevertheless, our understanding of them remains limited. -In this paper we conduct a direct analysis of the visual information contained -in representations by asking the following question: given an encoding of an -image, to which extent is it possible to reconstruct the image itself? To -answer this question we contribute a general framework to invert -representations. We show that this method can invert representations such as -HOG and SIFT more accurately than recent alternatives while being applicable to -CNNs too. We then use this technique to study the inverse of recent -state-of-the-art CNN image representations for the first time. Among our -findings, we show that several layers in CNNs retain photographically accurate -information about the image, with different degrees of geometric and -photometric invariance.}, - archiveprefix = {arXiv}, - author = {Aravindh Mahendran and Andrea Vedaldi}, - eprint = {1412.0035v1}, - file = {1412.0035v1.pdf}, - month = {Dec}, - primaryclass = {cs.CV}, - title = {Understanding Deep Image Representations by Inverting Them}, - url = {https://arxiv.org/abs/1412.0035v1}, - year = {2014} -} - - -@article{rZnxDitd, - abstract = {Apache Spark is a popular open-source platform for large-scale data -processing that is well-suited for iterative machine learning tasks. In this -paper we present MLlib, Spark's open-source distributed machine learning -library. MLlib provides efficient functionality for a wide range of learning -settings and includes several underlying statistical, optimization, and linear -algebra primitives. Shipped with Spark, MLlib supports several languages and -provides a high-level API that leverages Spark's rich ecosystem to simplify the -development of end-to-end machine learning pipelines. MLlib has experienced a -rapid growth due to its vibrant open-source community of over 140 contributors, and includes extensive documentation to support further growth and to let users -quickly get up to speed.}, - archiveprefix = {arXiv}, - author = {Xiangrui Meng and Joseph Bradley and Burak Yavuz and Evan Sparks and Shivaram Venkataraman and Davies Liu and Jeremy Freeman and DB Tsai and Manish Amde and Sean Owen and Doris Xin and Reynold Xin and Michael J. Franklin and Reza Zadeh and Matei Zaharia and Ameet Talwalkar}, - eprint = {1505.06807v1}, - file = {1505.06807v1.pdf}, - month = {May}, - primaryclass = {cs.LG}, - title = {MLlib: Machine Learning in Apache Spark}, - url = {https://arxiv.org/abs/1505.06807v1}, - year = {2015} -} - - -@article{rmJZ2Aui, - abstract = {Training deep networks is a time-consuming process, with networks for object -recognition often requiring multiple days to train. For this reason, leveraging -the resources of a cluster to speed up training is an important area of work. -However, widely-popular batch-processing computational frameworks like -MapReduce and Spark were not designed to support the asynchronous and -communication-intensive workloads of existing distributed deep learning -systems. We introduce SparkNet, a framework for training deep networks in -Spark. Our implementation includes a convenient interface for reading data from -Spark RDDs, a Scala interface to the Caffe deep learning framework, and a -lightweight multi-dimensional tensor library. Using a simple parallelization -scheme for stochastic gradient descent, SparkNet scales well with the cluster -size and tolerates very high-latency communication. Furthermore, it is easy to -deploy and use with no parameter tuning, and it is compatible with existing -Caffe models. We quantify the dependence of the speedup obtained by SparkNet on -the number of machines, the communication frequency, and the cluster's -communication overhead, and we benchmark our system's performance on the -ImageNet dataset.}, - archiveprefix = {arXiv}, - author = {Philipp Moritz and Robert Nishihara and Ion Stoica and Michael I. Jordan}, - eprint = {1511.06051v4}, - file = {1511.06051v4.pdf}, - month = {Dec}, - primaryclass = {stat.ML}, - title = {SparkNet: Training Deep Networks in Spark}, - url = {https://arxiv.org/abs/1511.06051v4}, - year = {2015} -} - - -@article{10ViHstXn, - abstract = {Although deep learning models have proven effective at solving problems in -natural language processing, the mechanism by which they come to their -conclusions is often unclear. As a result, these models are generally treated -as black boxes, yielding no insight of the underlying learned patterns. In this -paper we consider Long Short Term Memory networks (LSTMs) and demonstrate a new -approach for tracking the importance of a given input to the LSTM for a given -output. By identifying consistently important patterns of words, we are able to -distill state of the art LSTMs on sentiment analysis and question answering -into a set of representative phrases. This representation is then -quantitatively validated by using the extracted phrases to construct a simple, rule-based classifier which approximates the output of the LSTM.}, - archiveprefix = {arXiv}, - author = {W. James Murdoch and Arthur Szlam}, - eprint = {1702.02540v2}, - file = {1702.02540v2.pdf}, - month = {Feb}, - primaryclass = {cs.CL}, - title = {Automatic Rule Extraction from Long Short Term Memory Networks}, - url = {https://arxiv.org/abs/1702.02540v2}, - year = {2017} -} - - -@article{1AkF8Wsv7, - abstract = {Deep neural networks (DNNs) have recently been achieving state-of-the-art -performance on a variety of pattern-recognition tasks, most notably visual -classification problems. Given that DNNs are now able to classify objects in -images with near-human-level performance, questions naturally arise as to what -differences remain between computer and human vision. A recent study revealed -that changing an image (e.g. of a lion) in a way imperceptible to humans can -cause a DNN to label the image as something else entirely (e.g. mislabeling a -lion a library). Here we show a related result: it is easy to produce images -that are completely unrecognizable to humans, but that state-of-the-art DNNs -believe to be recognizable objects with 99.99% confidence (e.g. labeling with -certainty that white noise static is a lion). Specifically, we take -convolutional neural networks trained to perform well on either the ImageNet or -MNIST datasets and then find images with evolutionary algorithms or gradient -ascent that DNNs label with high confidence as belonging to each dataset class. -It is possible to produce images totally unrecognizable to human eyes that DNNs -believe with near certainty are familiar objects, which we call "fooling -images" (more generally, fooling examples). Our results shed light on -interesting differences between human vision and current DNNs, and raise -questions about the generality of DNN computer vision.}, - archiveprefix = {arXiv}, - author = {Anh Nguyen and Jason Yosinski and Jeff Clune}, - eprint = {1412.1897v4}, - file = {1412.1897v4.pdf}, - month = {12}, - primaryclass = {cs.CV}, - title = {Deep Neural Networks are Easily Fooled: High Confidence Predictions for -Unrecognizable Images}, - url = {https://arxiv.org/abs/1412.1897v4}, - year = {2014} -} - - -@article{1EayJRsI, - abstract = {This work introduces a method to tune a sequence-based generative model for -molecular de novo design that through augmented episodic likelihood can learn -to generate structures with certain specified desirable properties. We -demonstrate how this model can execute a range of tasks such as generating -analogues to a query structure and generating compounds predicted to be active -against a biological target. As a proof of principle, the model is first -trained to generate molecules that do not contain sulphur. As a second example, the model is trained to generate analogues to the drug Celecoxib, a technique -that could be used for scaffold hopping or library expansion starting from a -single molecule. Finally, when tuning the model towards generating compounds -predicted to be active against the dopamine receptor type 2, the model -generates structures of which more than 95% are predicted to be active, including experimentally confirmed actives that have not been included in -either the generative model nor the activity prediction model.}, - archiveprefix = {arXiv}, - author = {Marcus Olivecrona and Thomas Blaschke and Ola Engkvist and Hongming Chen}, - eprint = {1704.07555v2}, - file = {1704.07555v2.pdf}, - month = {Apr}, - primaryclass = {cs.AI}, - title = {Molecular De Novo Design through Deep Reinforcement Learning}, - url = {https://arxiv.org/abs/1704.07555v2}, - year = {2017} -} - - -@article{1TeyWffV, - abstract = {Since microRNAs (miRNAs) play a crucial role in post-transcriptional gene -regulation, miRNA identification is one of the most essential problems in -computational biology. miRNAs are usually short in length ranging between 20 -and 23 base pairs. It is thus often difficult to distinguish miRNA-encoding -sequences from other non-coding RNAs and pseudo miRNAs that have a similar -length, and most previous studies have recommended using precursor miRNAs -instead of mature miRNAs for robust detection. A great number of conventional -machine-learning-based classification methods have been proposed, but they -often have the serious disadvantage of requiring manual feature engineering, and their performance is limited as well. In this paper, we propose a novel -miRNA precursor prediction algorithm, deepMiRGene, based on recurrent neural -networks, specifically long short-term memory networks. deepMiRGene -automatically learns suitable features from the data themselves without manual -feature engineering and constructs a model that can successfully reflect -structural characteristics of precursor miRNAs. For the performance evaluation -of our approach, we have employed several widely used evaluation metrics on -three recent benchmark datasets and verified that deepMiRGene delivered -comparable performance among the current state-of-the-art tools.}, - archiveprefix = {arXiv}, - author = {Seunghyun Park and Seonwoo Min and Hyunsoo Choi and Sungroh Yoon}, - eprint = {1605.00017v1}, - file = {1605.00017v1.pdf}, - month = {Apr}, - primaryclass = {cs.LG}, - title = {deepMiRGene: Deep Neural Network based Precursor microRNA Prediction}, - url = {https://arxiv.org/abs/1605.00017v1}, - year = {2016} -} - - -@article{bNBiIiTt, - abstract = {Computational approaches to drug discovery can reduce the time and cost -associated with experimental assays and enable the screening of novel -chemotypes. Structure-based drug design methods rely on scoring functions to -rank and predict binding affinities and poses. The ever-expanding amount of -protein-ligand binding and structural data enables the use of deep machine -learning techniques for protein-ligand scoring. -We describe convolutional neural network (CNN) scoring functions that take as -input a comprehensive 3D representation of a protein-ligand interaction. A CNN -scoring function automatically learns the key features of protein-ligand -interactions that correlate with binding. We train and optimize our CNN scoring -functions to discriminate between correct and incorrect binding poses and known -binders and non-binders. We find that our CNN scoring function outperforms the -AutoDock Vina scoring function when ranking poses both for pose prediction and -virtual screening.}, - archiveprefix = {arXiv}, - author = {Matthew Ragoza and Joshua Hochuli and Elisa Idrobo and Jocelyn Sunseri and David Ryan Koes}, - eprint = {1612.02751v1}, - file = {1612.02751v1.pdf}, - month = {12}, - primaryclass = {stat.ML}, - title = {Protein-Ligand Scoring with Convolutional Neural Networks}, - url = {https://arxiv.org/abs/1612.02751v1}, - year = {2016} -} - - -@article{yAoN5gTU, - abstract = {Massively multitask neural architectures provide a learning framework for -drug discovery that synthesizes information from many distinct biological -sources. To train these architectures at scale, we gather large amounts of data -from public sources to create a dataset of nearly 40 million measurements -across more than 200 biological targets. We investigate several aspects of the -multitask framework by performing a series of empirical studies and obtain some -interesting results: (1) massively multitask networks obtain predictive -accuracies significantly better than single-task methods, (2) the predictive -power of multitask networks improves as additional tasks and data are added, (3) the total amount of data and the total number of tasks both contribute -significantly to multitask improvement, and (4) multitask networks afford -limited transferability to tasks not in the training set. Our results -underscore the need for greater data sharing and further algorithmic innovation -to accelerate the drug discovery process.}, - archiveprefix = {arXiv}, - author = {Bharath Ramsundar and Steven Kearnes and Patrick Riley and Dale Webster and David Konerding and Vijay Pande}, - eprint = {1502.02072v1}, - file = {1502.02072v1.pdf}, - month = {Feb}, - primaryclass = {stat.ML}, - title = {Massively Multitask Networks for Drug Discovery}, - url = {https://arxiv.org/abs/1502.02072v1}, - year = {2015} -} - - -@article{QwXSJhr0, - abstract = {Despite widespread adoption, machine learning models remain mostly black -boxes. Understanding the reasons behind predictions is, however, quite -important in assessing trust, which is fundamental if one plans to take action -based on a prediction, or when choosing whether to deploy a new model. Such -understanding also provides insights into the model, which can be used to -transform an untrustworthy model or prediction into a trustworthy one. In this -work, we propose LIME, a novel explanation technique that explains the -predictions of any classifier in an interpretable and faithful manner, by -learning an interpretable model locally around the prediction. We also propose -a method to explain models by presenting representative individual predictions -and their explanations in a non-redundant way, framing the task as a submodular -optimization problem. We demonstrate the flexibility of these methods by -explaining different models for text (e.g. random forests) and image -classification (e.g. neural networks). We show the utility of explanations via -novel experiments, both simulated and with human subjects, on various scenarios -that require trust: deciding if one should trust a prediction, choosing between -models, improving an untrustworthy classifier, and identifying why a classifier -should not be trusted.}, - archiveprefix = {arXiv}, - author = {Marco Tulio Ribeiro and Sameer Singh and Carlos Guestrin}, - eprint = {1602.04938v3}, - file = {1602.04938v3.pdf}, - month = {Feb}, - primaryclass = {cs.LG}, - title = {"Why Should I Trust You?": Explaining the Predictions of Any Classifier}, - url = {https://arxiv.org/abs/1602.04938v3}, - year = {2016} -} - - -@article{w6CoVmFK, - abstract = {Stochastic gradient descent (SGD) is a ubiquitous algorithm for a variety of -machine learning problems. Researchers and industry have developed several -techniques to optimize SGD's runtime performance, including asynchronous -execution and reduced precision. Our main result is a martingale-based analysis -that enables us to capture the rich noise models that may arise from such -techniques. Specifically, we use our new analysis in three ways: (1) we derive -convergence rates for the convex case (Hogwild!) with relaxed assumptions on -the sparsity of the problem; (2) we analyze asynchronous SGD algorithms for -non-convex matrix problems including matrix completion; and (3) we design and -analyze an asynchronous SGD algorithm, called Buckwild!, that uses -lower-precision arithmetic. We show experimentally that our algorithms run -efficiently for a variety of problems on modern hardware.}, - archiveprefix = {arXiv}, - author = {Christopher De Sa and Ce Zhang and Kunle Olukotun and Christopher Ré}, - eprint = {1506.06438v2}, - file = {1506.06438v2.pdf}, - month = {Jun}, - primaryclass = {cs.LG}, - title = {Taming the Wild: A Unified Analysis of Hogwild!-Style Algorithms}, - url = {https://arxiv.org/abs/1506.06438v2}, - year = {2015} -} - - -@article{8LWFFeYg, - abstract = {In de novo drug design, computational strategies are used to generate novel -molecules with good affinity to the desired biological target. In this work, we -show that recurrent neural networks can be trained as generative models for -molecular structures, similar to statistical language models in natural -language processing. We demonstrate that the properties of the generated -molecules correlate very well with the properties of the molecules used to -train the model. In order to enrich libraries with molecules active towards a -given biological target, we propose to fine-tune the model with small sets of -molecules, which are known to be active against that target. -Against Staphylococcus aureus, the model reproduced 14% of 6051 hold-out test -molecules that medicinal chemists designed, whereas against Plasmodium -falciparum (Malaria) it reproduced 28% of 1240 test molecules. When coupled -with a scoring function, our model can perform the complete de novo drug design -cycle to generate large sets of novel molecules for drug discovery.}, - archiveprefix = {arXiv}, - author = {Marwin H. S. Segler and Thierry Kogej and Christian Tyrchan and Mark P. Waller}, - eprint = {1701.01329v1}, - file = {1701.01329v1.pdf}, - month = {Jan}, - primaryclass = {cs.NE}, - title = {Generating Focussed Molecule Libraries for Drug Discovery with Recurrent -Neural Networks}, - url = {https://arxiv.org/abs/1701.01329v1}, - year = {2017} -} - - -@article{RZsNSRDS, - abstract = {We propose a technique for producing "visual explanations" for decisions from -a large class of CNN-based models, making them more transparent. Our approach - -Gradient-weighted Class Activation Mapping (Grad-CAM), uses the gradients of -any target concept, flowing into the final convolutional layer to produce a -coarse localization map highlighting the important regions in the image for -predicting the concept. Unlike previous approaches, GradCAM is applicable to a -wide variety of CNN model-families: (1) CNNs with fully-connected layers (e.g. -VGG), (2) CNNs used for structured outputs (e.g. captioning), (3) CNNs used in -tasks with multimodal inputs (e.g. VQA) or reinforcement learning, without any -architectural changes or re-training. We combine GradCAM with fine-grained -visualizations to create a high-resolution class-discriminative visualization -and apply it to off-the-shelf image classification, captioning, and visual -question answering (VQA) models, including ResNet-based architectures. In the -context of image classification models, our visualizations (a) lend insights -into their failure modes (showing that seemingly unreasonable predictions have -reasonable explanations), (b) are robust to adversarial images, (c) outperform -previous methods on weakly-supervised localization, (d) are more faithful to -the underlying model and (e) help achieve generalization by identifying dataset -bias. For captioning and VQA, our visualizations show that even non-attention -based models can localize inputs. Finally, we conduct human studies to measure -if GradCAM explanations help users establish trust in predictions from deep -networks and show that GradCAM helps untrained users successfully discern a -"stronger" deep network from a "weaker" one. Our code is available at -https://github.com/ramprs/grad-cam. A demo and a video of the demo can be found -at http://gradcam.cloudcv.org and youtu.be/COjUB9Izk6E.}, - archiveprefix = {arXiv}, - author = {Ramprasaath R. Selvaraju and Michael Cogswell and Abhishek Das and Ramakrishna Vedantam and Devi Parikh and Dhruv Batra}, - eprint = {1610.02391v3}, - file = {1610.02391v3.pdf}, - month = {Nov}, - primaryclass = {cs.CV}, - title = {Grad-CAM: Visual Explanations from Deep Networks via Gradient-based -Localization}, - url = {https://arxiv.org/abs/1610.02391v3}, - year = {2016} -} - - -@article{T2Md9xLY, - abstract = {Sources of variability in experimentally derived data include measurement -error in addition to the physical phenomena of interest. This measurement error -is a combination of systematic components, originating from the measuring -instrument, and random measurement errors. Several novel biological -technologies, such as mass cytometry and single-cell RNA-seq, are plagued with -systematic errors that may severely affect statistical analysis if the data is -not properly calibrated. We propose a novel deep learning approach for removing -systematic batch effects. Our method is based on a residual network, trained to -minimize the Maximum Mean Discrepancy (MMD) between the multivariate -distributions of two replicates, measured in different batches. We apply our -method to mass cytometry and single-cell RNA-seq datasets, and demonstrate that -it effectively attenuates batch effects.}, - archiveprefix = {arXiv}, - author = {Uri Shaham and Kelly P. Stanton and Jun Zhao and Huamin Li and Khadir Raddassi and Ruth Montgomery and Yuval Kluger}, - doi = {10.1093/bioinformatics/btx196}, - eprint = {1610.04181v5}, - file = {1610.04181v5.pdf}, - month = {Nov}, - primaryclass = {stat.ML}, - title = {Removal of Batch Effects using Distribution-Matching Residual Networks}, - url = {https://arxiv.org/abs/1610.04181v5}, - year = {2016} -} - - -@article{zhmq9ktJ, - abstract = {The purported "black box"' nature of neural networks is a barrier to adoption -in applications where interpretability is essential. Here we present DeepLIFT -(Deep Learning Important FeaTures), a method for decomposing the output -prediction of a neural network on a specific input by backpropagating the -contributions of all neurons in the network to every feature of the input. -DeepLIFT compares the activation of each neuron to its 'reference activation' -and assigns contribution scores according to the difference. By optionally -giving separate consideration to positive and negative contributions, DeepLIFT -can also reveal dependencies which are missed by other approaches. Scores can -be computed efficiently in a single backward pass. We apply DeepLIFT to models -trained on MNIST and simulated genomic data, and show significant advantages -over gradient-based methods. A detailed video tutorial on the method is at -http://goo.gl/qKb7pL and code is at http://goo.gl/RM8jvH.}, - archiveprefix = {arXiv}, - author = {Avanti Shrikumar and Peyton Greenside and Anshul Kundaje}, - eprint = {1704.02685v1}, - file = {1704.02685v1.pdf}, - month = {Apr}, - primaryclass = {cs.CV}, - title = {Learning Important Features Through Propagating Activation Differences}, - url = {https://arxiv.org/abs/1704.02685v1}, - year = {2017} -} - - -@article{1YcKYTvO, - abstract = {This paper addresses the visualisation of image classification models, learnt -using deep Convolutional Networks (ConvNets). We consider two visualisation -techniques, based on computing the gradient of the class score with respect to -the input image. The first one generates an image, which maximises the class -score [Erhan et al., 2009], thus visualising the notion of the class, captured -by a ConvNet. The second technique computes a class saliency map, specific to a -given image and class. We show that such maps can be employed for weakly -supervised object segmentation using classification ConvNets. Finally, we -establish the connection between the gradient-based ConvNet visualisation -methods and deconvolutional networks [Zeiler et al., 2013].}, - archiveprefix = {arXiv}, - author = {Karen Simonyan and Andrea Vedaldi and Andrew Zisserman}, - eprint = {1312.6034v2}, - file = {1312.6034v2.pdf}, - month = {12}, - primaryclass = {cs.CV}, - title = {Deep Inside Convolutional Networks: Visualising Image Classification -Models and Saliency Maps}, - url = {https://arxiv.org/abs/1312.6034v2}, - year = {2013} -} - - -@article{G10wkFHt, - abstract = {Motivation: Histone modifications are among the most important factors that -control gene regulation. Computational methods that predict gene expression -from histone modification signals are highly desirable for understanding their -combinatorial effects in gene regulation. This knowledge can help in developing -'epigenetic drugs' for diseases like cancer. Previous studies for quantifying -the relationship between histone modifications and gene expression levels -either failed to capture combinatorial effects or relied on multiple methods -that separate predictions and combinatorial analysis. This paper develops a -unified discriminative framework using a deep convolutional neural network to -classify gene expression using histone modification data as input. Our system, called DeepChrome, allows automatic extraction of complex interactions among -important features. To simultaneously visualize the combinatorial interactions -among histone modifications, we propose a novel optimization-based technique -that generates feature pattern maps from the learnt deep model. This provides -an intuitive description of underlying epigenetic mechanisms that regulate -genes. Results: We show that DeepChrome outperforms state-of-the-art models -like Support Vector Machines and Random Forests for gene expression -classification task on 56 different cell-types from REMC database. The output -of our visualization technique not only validates the previous observations but -also allows novel insights about combinatorial interactions among histone -modification marks, some of which have recently been observed by experimental -studies.}, - archiveprefix = {arXiv}, - author = {Ritambhara Singh and Jack Lanchantin and Gabriel Robins and Yanjun Qi}, - eprint = {1607.02078v1}, - file = {1607.02078v1.pdf}, - month = {Jul}, - primaryclass = {cs.LG}, - title = {DeepChrome: Deep-learning for predicting gene expression from histone -modifications}, - url = {https://arxiv.org/abs/1607.02078v1}, - year = {2016} -} - - -@article{81Cl5QSM, - abstract = {Machine learning is widely used to analyze biological sequence data. -Non-sequential models such as SVMs or feed-forward neural networks are often -used although they have no natural way of handling sequences of varying length. -Recurrent neural networks such as the long short term memory (LSTM) model on -the other hand are designed to handle sequences. In this study we demonstrate -that LSTM networks predict the subcellular location of proteins given only the -protein sequence with high accuracy (0.902) outperforming current state of the -art algorithms. We further improve the performance by introducing convolutional -filters and experiment with an attention mechanism which lets the LSTM focus on -specific parts of the protein. Lastly we introduce new visualizations of both -the convolutional filters and the attention mechanisms and show how they can be -used to extract biological relevant knowledge from the LSTM networks.}, - archiveprefix = {arXiv}, - author = {Søren Kaae Sønderby and Casper Kaae Sønderby and Henrik Nielsen and Ole Winther}, - doi = {10.1007/978-3-319-21233-3_6}, - eprint = {1503.01919v1}, - file = {1503.01919v1.pdf}, - month = {Mar}, - note = {Algorithms for Computational Biology 9199 (2015) 68}, - primaryclass = {q-bio.QM}, - title = {Convolutional LSTM Networks for Subcellular Localization of Proteins}, - url = {https://arxiv.org/abs/1503.01919v1}, - year = {2015} -} - - -@article{f2L6isRj, - abstract = {Most modern convolutional neural networks (CNNs) used for object recognition -are built using the same principles: Alternating convolution and max-pooling -layers followed by a small number of fully connected layers. We re-evaluate the -state of the art for object recognition from small images with convolutional -networks, questioning the necessity of different components in the pipeline. We -find that max-pooling can simply be replaced by a convolutional layer with -increased stride without loss in accuracy on several image recognition -benchmarks. Following this finding -- and building on other recent work for -finding simple network structures -- we propose a new architecture that -consists solely of convolutional layers and yields competitive or state of the -art performance on several object recognition datasets (CIFAR-10, CIFAR-100, ImageNet). To analyze the network we introduce a new variant of the -"deconvolution approach" for visualizing features learned by CNNs, which can be -applied to a broader range of network structures than existing approaches.}, - archiveprefix = {arXiv}, - author = {Jost Tobias Springenberg and Alexey Dosovitskiy and Thomas Brox and Martin Riedmiller}, - eprint = {1412.6806v3}, - file = {1412.6806v3.pdf}, - month = {12}, - primaryclass = {cs.LG}, - title = {Striving for Simplicity: The All Convolutional Net}, - url = {https://arxiv.org/abs/1412.6806v3}, - year = {2014} -} - - -@article{1Ad3UOefc, - abstract = {Recurrent neural networks, and in particular long short-term memory networks -(LSTMs), are a remarkably effective tool for sequence modeling that learn a -dense black-box hidden representation of their sequential input. Researchers -interested in better understanding these models have studied the changes in -hidden state representations over time and noticed some interpretable patterns -but also significant noise. In this work, we present LSTMVis a visual analysis -tool for recurrent neural networks with a focus on understanding these hidden -state dynamics. The tool allows a user to select a hypothesis input range to -focus on local state changes, to match these states changes to similar patterns -in a large data set, and to align these results with domain specific structural -annotations. We further show several use cases of the tool for analyzing -specific hidden state properties on datasets containing nesting, phrase -structure, and chord progressions, and demonstrate how the tool can be used to -isolate patterns for further statistical analysis.}, - archiveprefix = {arXiv}, - author = {Hendrik Strobelt and Sebastian Gehrmann and Bernd Huber and Hanspeter Pfister and Alexander M. Rush}, - eprint = {1606.07461v1}, - file = {1606.07461v1.pdf}, - month = {Jun}, - primaryclass = {cs.CL}, - title = {Visual Analysis of Hidden State Dynamics in Recurrent Neural Networks}, - url = {https://arxiv.org/abs/1606.07461v1}, - year = {2016} -} - - -@article{aClNvbyM, - abstract = {In this work we apply model averaging to parallel training of deep neural -network (DNN). Parallelization is done in a model averaging manner. Data is -partitioned and distributed to different nodes for local model updates, and -model averaging across nodes is done every few minibatches. We use multiple -GPUs for data parallelization, and Message Passing Interface (MPI) for -communication between nodes, which allows us to perform model averaging -frequently without losing much time on communication. We investigate the -effectiveness of Natural Gradient Stochastic Gradient Descent (NG-SGD) and -Restricted Boltzmann Machine (RBM) pretraining for parallel training in -model-averaging framework, and explore the best setups in term of different -learning rate schedules, averaging frequencies and minibatch sizes. It is shown -that NG-SGD and RBM pretraining benefits parameter-averaging based model -training. On the 300h Switchboard dataset, a 9.3 times speedup is achieved -using 16 GPUs and 17 times speedup using 32 GPUs with limited decoding accuracy -loss.}, - archiveprefix = {arXiv}, - author = {Hang Su and Haoyu Chen}, - eprint = {1507.01239v2}, - file = {1507.01239v2.pdf}, - month = {Jul}, - primaryclass = {cs.LG}, - title = {Experiments on Parallel Training of Deep Neural Network using Model -Averaging}, - url = {https://arxiv.org/abs/1507.01239v2}, - year = {2015} -} - - -@article{JUF9VoRD, - abstract = {Parallelization framework has become a necessity to speed up the training of -deep neural networks (DNN) recently. Such framework typically employs the Model -Average approach, denoted as MA-DNN, in which parallel workers conduct -respective training based on their own local data while the parameters of local -models are periodically communicated and averaged to obtain a global model -which serves as the new start of local models. However, since DNN is a highly -non-convex model, averaging parameters cannot ensure that such global model can -perform better than those local models. To tackle this problem, we introduce a -new parallel training framework called Ensemble-Compression, denoted as EC-DNN. -In this framework, we propose to aggregate the local models by ensemble, i.e., averaging the outputs of local models instead of the parameters. As most of -prevalent loss functions are convex to the output of DNN, the performance of -ensemble-based global model is guaranteed to be at least as good as the average -performance of local models. However, a big challenge lies in the explosion of -model size since each round of ensemble can give rise to multiple times size -increment. Thus, we carry out model compression after each ensemble, specialized by a distillation based method in this paper, to reduce the size of -the global model to be the same as the local ones. Our experimental results -demonstrate the prominent advantage of EC-DNN over MA-DNN in terms of both -accuracy and speedup.}, - archiveprefix = {arXiv}, - author = {Shizhao Sun and Wei Chen and Jiang Bian and Xiaoguang Liu and Tie-Yan Liu}, - eprint = {1606.00575v2}, - file = {1606.00575v2.pdf}, - month = {Jun}, - primaryclass = {cs.DC}, - title = {Ensemble-Compression: A New Method for Parallel Training of Deep Neural -Networks}, - url = {https://arxiv.org/abs/1606.00575v2}, - year = {2016} -} - - -@article{WzFOJBiA, - abstract = {We study the problem of attributing the prediction of a deep network to its -input features, a problem previously studied by several other works. We -identify two fundamental axioms---Sensitivity and Implementation Invariance -that attribution methods ought to satisfy. We show that they are not satisfied -by most known attribution methods, which we consider to be a fundamental -weakness of those methods. We use the axioms to guide the design of a new -attribution method called Integrated Gradients. Our method requires no -modification to the original network and is extremely simple to implement; it -just needs a few calls to the standard gradient operator. We apply this method -to a couple of image models, a couple of text models and a chemistry model, demonstrating its ability to debug networks, to extract rules from a network, and to enable users to engage with models better.}, - archiveprefix = {arXiv}, - author = {Mukund Sundararajan and Ankur Taly and Qiqi Yan}, - eprint = {1703.01365v2}, - file = {1703.01365v2.pdf}, - month = {Mar}, - primaryclass = {cs.LG}, - title = {Axiomatic Attribution for Deep Networks}, - url = {https://arxiv.org/abs/1703.01365v2}, - year = {2017} -} - - -@article{2cMhMv5A, - abstract = {Deep Neural Networks (DNNs) are powerful models that have achieved excellent -performance on difficult learning tasks. Although DNNs work well whenever large -labeled training sets are available, they cannot be used to map sequences to -sequences. In this paper, we present a general end-to-end approach to sequence -learning that makes minimal assumptions on the sequence structure. Our method -uses a multilayered Long Short-Term Memory (LSTM) to map the input sequence to -a vector of a fixed dimensionality, and then another deep LSTM to decode the -target sequence from the vector. Our main result is that on an English to -French translation task from the WMT'14 dataset, the translations produced by -the LSTM achieve a BLEU score of 34.8 on the entire test set, where the LSTM's -BLEU score was penalized on out-of-vocabulary words. Additionally, the LSTM did -not have difficulty on long sentences. For comparison, a phrase-based SMT -system achieves a BLEU score of 33.3 on the same dataset. When we used the LSTM -to rerank the 1000 hypotheses produced by the aforementioned SMT system, its -BLEU score increases to 36.5, which is close to the previous best result on -this task. The LSTM also learned sensible phrase and sentence representations -that are sensitive to word order and are relatively invariant to the active and -the passive voice. Finally, we found that reversing the order of the words in -all source sentences (but not target sentences) improved the LSTM's performance -markedly, because doing so introduced many short term dependencies between the -source and the target sentence which made the optimization problem easier.}, - archiveprefix = {arXiv}, - author = {Ilya Sutskever and Oriol Vinyals and Quoc V. Le}, - eprint = {1409.3215v3}, - file = {1409.3215v3.pdf}, - month = {Sep}, - primaryclass = {cs.CL}, - title = {Sequence to Sequence Learning with Neural Networks}, - url = {https://arxiv.org/abs/1409.3215v3}, - year = {2014} -} - - -@article{hOeUlCvS, - abstract = {TensorFlow is an interface for expressing machine learning algorithms, and an -implementation for executing such algorithms. A computation expressed using -TensorFlow can be executed with little or no change on a wide variety of -heterogeneous systems, ranging from mobile devices such as phones and tablets -up to large-scale distributed systems of hundreds of machines and thousands of -computational devices such as GPU cards. The system is flexible and can be used -to express a wide variety of algorithms, including training and inference -algorithms for deep neural network models, and it has been used for conducting -research and for deploying machine learning systems into production across more -than a dozen areas of computer science and other fields, including speech -recognition, computer vision, robotics, information retrieval, natural language -processing, geographic information extraction, and computational drug -discovery. This paper describes the TensorFlow interface and an implementation -of that interface that we have built at Google. The TensorFlow API and a -reference implementation were released as an open-source package under the -Apache 2.0 license in November, 2015 and are available at www.tensorflow.org.}, - archiveprefix = {arXiv}, - author = {Martín Abadi and Ashish Agarwal and Paul Barham and Eugene Brevdo and Zhifeng Chen and Craig Citro and Greg S. Corrado and Andy Davis and Jeffrey Dean and Matthieu Devin and Sanjay Ghemawat and Ian Goodfellow and Andrew Harp and Geoffrey Irving and Michael Isard and Yangqing Jia and Rafal Jozefowicz and Lukasz Kaiser and Manjunath Kudlur and Josh Levenberg and Dan Mane and Rajat Monga and Sherry Moore and Derek Murray and Chris Olah and Mike Schuster and Jonathon Shlens and Benoit Steiner and Ilya Sutskever and Kunal Talwar and Paul Tucker and Vincent Vanhoucke and Vijay Vasudevan and Fernanda Viegas and Oriol Vinyals and Pete Warden and Martin Wattenberg and Martin Wicke and Yuan Yu and Xiaoqiang Zheng}, - eprint = {1603.04467v2}, - file = {1603.04467v2.pdf}, - month = {Mar}, - primaryclass = {cs.DC}, - title = {TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed -Systems}, - url = {https://arxiv.org/abs/1603.04467v2}, - year = {2016} -} - - -@article{1GhHIDxuW, - abstract = {We propose two novel model architectures for computing continuous vector -representations of words from very large data sets. The quality of these -representations is measured in a word similarity task, and the results are -compared to the previously best performing techniques based on different types -of neural networks. We observe large improvements in accuracy at much lower -computational cost, i.e. it takes less than a day to learn high quality word -vectors from a 1.6 billion words data set. Furthermore, we show that these -vectors provide state-of-the-art performance on our test set for measuring -syntactic and semantic word similarities.}, - archiveprefix = {arXiv}, - author = {Tomas Mikolov and Kai Chen and Greg Corrado and Jeffrey Dean}, - eprint = {1301.3781v3}, - file = {1301.3781v3.pdf}, - month = {Jan}, - primaryclass = {cs.CL}, - title = {Efficient Estimation of Word Representations in Vector Space}, - url = {https://arxiv.org/abs/1301.3781v3}, - year = {2013} -} - - -@article{16OPHvAij, - abstract = {Molecular machine learning has been maturing rapidly over the last few years. -Improved methods and the presence of larger datasets have enabled machine -learning algorithms to make increasingly accurate predictions about molecular -properties. However, algorithmic progress has been limited due to the lack of a -standard benchmark to compare the efficacy of proposed methods; most new -algorithms are benchmarked on different datasets making it challenging to gauge -the quality of proposed methods. This work introduces MoleculeNet, a large -scale benchmark for molecular machine learning. MoleculeNet curates multiple -public datasets, establishes metrics for evaluation, and offers high quality -open-source implementations of multiple previously proposed molecular -featurization and learning algorithms (released as part of the DeepChem open -source library). MoleculeNet benchmarks demonstrate that learnable -representations, and in particular graph convolutional networks, are powerful -tools for molecular machine learning and broadly offer the best performance. -However, for quantum mechanical and biophysical datasets, the use of -physics-aware featurizations can be significantly more important than choice of -particular learning algorithm.}, - archiveprefix = {arXiv}, - author = {Zhenqin Wu and Bharath Ramsundar and Evan N. Feinberg and Joseph Gomes and Caleb Geniesse and Aneesh S. Pappu and Karl Leswing and Vijay Pande}, - eprint = {1703.00564v1}, - file = {1703.00564v1.pdf}, - month = {Mar}, - primaryclass = {cs.LG}, - title = {MoleculeNet: A Benchmark for Molecular Machine Learning}, - url = {https://arxiv.org/abs/1703.00564v1}, - year = {2017} -} - - -@article{yHn4SDRI, - abstract = {Inspired by recent work in machine translation and object detection, we -introduce an attention based model that automatically learns to describe the -content of images. We describe how we can train this model in a deterministic -manner using standard backpropagation techniques and stochastically by -maximizing a variational lower bound. We also show through visualization how -the model is able to automatically learn to fix its gaze on salient objects -while generating the corresponding words in the output sequence. We validate -the use of attention with state-of-the-art performance on three benchmark -datasets: Flickr8k, Flickr30k and MS COCO.}, - archiveprefix = {arXiv}, - author = {Kelvin Xu and Jimmy Ba and Ryan Kiros and Kyunghyun Cho and Aaron Courville and Ruslan Salakhutdinov and Richard Zemel and Yoshua Bengio}, - eprint = {1502.03044v3}, - file = {1502.03044v3.pdf}, - month = {Feb}, - primaryclass = {cs.LG}, - title = {Show, Attend and Tell: Neural Image Caption Generation with Visual -Attention}, - url = {https://arxiv.org/abs/1502.03044v3}, - year = {2015} -} - - -@article{17i18PMkR, - abstract = {Recent years have produced great advances in training large, deep neural -networks (DNNs), including notable successes in training convolutional neural -networks (convnets) to recognize natural images. However, our understanding of -how these models work, especially what computations they perform at -intermediate layers, has lagged behind. Progress in the field will be further -accelerated by the development of better tools for visualizing and interpreting -neural nets. We introduce two such tools here. The first is a tool that -visualizes the activations produced on each layer of a trained convnet as it -processes an image or video (e.g. a live webcam stream). We have found that -looking at live activations that change in response to user input helps build -valuable intuitions about how convnets work. The second tool enables -visualizing features at each layer of a DNN via regularized optimization in -image space. Because previous versions of this idea produced less recognizable -images, here we introduce several new regularization methods that combine to -produce qualitatively clearer, more interpretable visualizations. Both tools -are open source and work on a pre-trained convnet with minimal setup.}, - archiveprefix = {arXiv}, - author = {Jason Yosinski and Jeff Clune and Anh Nguyen and Thomas Fuchs and Hod Lipson}, - eprint = {1506.06579v1}, - file = {1506.06579v1.pdf}, - month = {Jun}, - primaryclass = {cs.CV}, - title = {Understanding Neural Networks Through Deep Visualization}, - url = {https://arxiv.org/abs/1506.06579v1}, - year = {2015} -} - - -@article{voh0OiT2, - abstract = {Large Convolutional Network models have recently demonstrated impressive -classification performance on the ImageNet benchmark. However there is no clear -understanding of why they perform so well, or how they might be improved. In -this paper we address both issues. We introduce a novel visualization technique -that gives insight into the function of intermediate feature layers and the -operation of the classifier. We also perform an ablation study to discover the -performance contribution from different model layers. This enables us to find -model architectures that outperform Krizhevsky \etal on the ImageNet -classification benchmark. We show our ImageNet model generalizes well to other -datasets: when the softmax classifier is retrained, it convincingly beats the -current state-of-the-art results on Caltech-101 and Caltech-256 datasets.}, - archiveprefix = {arXiv}, - author = {Matthew D Zeiler and Rob Fergus}, - eprint = {1311.2901v3}, - file = {1311.2901v3.pdf}, - month = {Dec}, - primaryclass = {cs.CV}, - title = {Visualizing and Understanding Convolutional Networks}, - url = {https://arxiv.org/abs/1311.2901v3}, - year = {2013} -} - - -@article{Kk20paR7, - abstract = {This article presents the prediction difference analysis method for -visualizing the response of a deep neural network to a specific input. When -classifying images, the method highlights areas in a given input image that -provide evidence for or against a certain class. It overcomes several -shortcoming of previous methods and provides great additional insight into the -decision making process of classifiers. Making neural network decisions -interpretable through visualization is important both to improve models and to -accelerate the adoption of black-box classifiers in application areas such as -medicine. We illustrate the method in experiments on natural images (ImageNet -data), as well as medical images (MRI brain scans).}, - archiveprefix = {arXiv}, - author = {Luisa M Zintgraf and Taco S Cohen and Tameem Adel and Max Welling}, - eprint = {1702.04595v1}, - file = {1702.04595v1.pdf}, - month = {Feb}, - primaryclass = {cs.CV}, - title = {Visualizing Deep Neural Network Decisions: Prediction Difference -Analysis}, - url = {https://arxiv.org/abs/1702.04595v1}, - year = {2017} -} - diff --git a/citations.json b/citations.json deleted file mode 100644 index 32a8007e..00000000 --- a/citations.json +++ /dev/null @@ -1,46092 +0,0 @@ -{ - "arxiv:1106.5730": { - "source": "arxiv", - "identifer": "1106.5730", - "standard_citation": "arxiv:1106.5730", - "bibtex": "@article{3qm8sXnB,\n abstract = {Stochastic Gradient Descent (SGD) is a popular algorithm that can achieve\nstate-of-the-art performance on a variety of machine learning tasks. Several\nresearchers have recently proposed schemes to parallelize SGD, but all require\nperformance-destroying memory locking and synchronization. This work aims to\nshow using novel theoretical analysis, algorithms, and implementation that SGD\ncan be implemented without any locking. We present an update scheme called\nHOGWILD! which allows processors access to shared memory with the possibility\nof overwriting each other's work. We show that when the associated optimization\nproblem is sparse, meaning most gradient updates only modify small parts of the\ndecision variable, then HOGWILD! achieves a nearly optimal rate of convergence.\nWe demonstrate experimentally that HOGWILD! outperforms alternative schemes\nthat use locking by an order of magnitude.},\n archiveprefix = {arXiv},\n author = {Feng Niu and Benjamin Recht and Christopher Re and Stephen J. Wright},\n eprint = {1106.5730v2},\n file = {1106.5730v2.pdf},\n month = {Jun},\n primaryclass = {math.OC},\n title = {HOGWILD!: A Lock-Free Approach to Parallelizing Stochastic Gradient\nDescent},\n url = {https://arxiv.org/abs/1106.5730v2},\n year = {2011}\n}\n\n", - "citation_id": "3qm8sXnB" - }, - "arxiv:1206.7051": { - "source": "arxiv", - "identifer": "1206.7051", - "standard_citation": "arxiv:1206.7051", - "bibtex": "@article{8RAYEOPl,\n abstract = {We develop stochastic variational inference, a scalable algorithm for\napproximating posterior distributions. We develop this technique for a large\nclass of probabilistic models and we demonstrate it with two probabilistic\ntopic models, latent Dirichlet allocation and the hierarchical Dirichlet\nprocess topic model. Using stochastic variational inference, we analyze several\nlarge collections of documents: 300K articles from Nature, 1.8M articles from\nThe New York Times, and 3.8M articles from Wikipedia. Stochastic inference can\neasily handle data sets of this size and outperforms traditional variational\ninference, which can only handle a smaller subset. (We also show that the\nBayesian nonparametric topic model outperforms its parametric counterpart.)\nStochastic variational inference lets us apply complex Bayesian models to\nmassive data sets.},\n archiveprefix = {arXiv},\n author = {Matt Hoffman and David M. Blei and Chong Wang and John Paisley},\n eprint = {1206.7051v3},\n file = {1206.7051v3.pdf},\n month = {Jun},\n primaryclass = {stat.ML},\n title = {Stochastic Variational Inference},\n url = {https://arxiv.org/abs/1206.7051v3},\n year = {2012}\n}\n\n", - "citation_id": "8RAYEOPl" - }, - "arxiv:1212.0901v2": { - "source": "arxiv", - "identifer": "1212.0901v2", - "standard_citation": "arxiv:1212.0901v2", - "bibtex": "@article{g2vvbB91,\n abstract = {After a more than decade-long period of relatively little research activity\nin the area of recurrent neural networks, several new developments will be\nreviewed here that have allowed substantial progress both in understanding and\nin technical solutions towards more efficient training of recurrent networks.\nThese advances have been motivated by and related to the optimization issues\nsurrounding deep learning. Although recurrent networks are extremely powerful\nin what they can in principle represent in terms of modelling sequences,their\ntraining is plagued by two aspects of the same issue regarding the learning of\nlong-term dependencies. Experiments reported here evaluate the use of clipping\ngradients, spanning longer time ranges with leaky integration, advanced\nmomentum techniques, using more powerful output probability models, and\nencouraging sparser gradients to help symmetry breaking and credit assignment.\nThe experiments are performed on text and music data and show off the combined\neffects of these techniques in generally improving both training and test\nerror.},\n archiveprefix = {arXiv},\n author = {Yoshua Bengio and Nicolas Boulanger-Lewandowski and Razvan Pascanu},\n eprint = {1212.0901v2},\n file = {1212.0901v2.pdf},\n month = {12},\n primaryclass = {cs.LG},\n title = {Advances in Optimizing Recurrent Networks},\n url = {https://arxiv.org/abs/1212.0901v2},\n year = {2012}\n}\n\n", - "citation_id": "g2vvbB91" - }, - "arxiv:1308.0850": { - "source": "arxiv", - "identifer": "1308.0850", - "standard_citation": "arxiv:1308.0850", - "bibtex": "@article{15y7iq6HF,\n abstract = {This paper shows how Long Short-term Memory recurrent neural networks can be\nused to generate complex sequences with long-range structure, simply by\npredicting one data point at a time. The approach is demonstrated for text\n(where the data are discrete) and online handwriting (where the data are\nreal-valued). It is then extended to handwriting synthesis by allowing the\nnetwork to condition its predictions on a text sequence. The resulting system\nis able to generate highly realistic cursive handwriting in a wide variety of\nstyles.},\n archiveprefix = {arXiv},\n author = {Alex Graves},\n eprint = {1308.0850v5},\n file = {1308.0850v5.pdf},\n month = {Aug},\n primaryclass = {cs.NE},\n title = {Generating Sequences With Recurrent Neural Networks},\n url = {https://arxiv.org/abs/1308.0850v5},\n year = {2013}\n}\n\n", - "citation_id": "15y7iq6HF" - }, - "arxiv:1312.6199": { - "source": "arxiv", - "identifer": "1312.6199", - "standard_citation": "arxiv:1312.6199", - "bibtex": "@article{1Fel6Bdb8,\n abstract = {Deep neural networks are highly expressive models that have recently achieved\nstate of the art performance on speech and visual recognition tasks. While\ntheir expressiveness is the reason they succeed, it also causes them to learn\nuninterpretable solutions that could have counter-intuitive properties. In this\npaper we report two such properties.\nFirst, we find that there is no distinction between individual high level\nunits and random linear combinations of high level units, according to various\nmethods of unit analysis. It suggests that it is the space, rather than the\nindividual units, that contains of the semantic information in the high layers\nof neural networks.\nSecond, we find that deep neural networks learn input-output mappings that\nare fairly discontinuous to a significant extend. We can cause the network to\nmisclassify an image by applying a certain imperceptible perturbation, which is\nfound by maximizing the network's prediction error. In addition, the specific\nnature of these perturbations is not a random artifact of learning: the same\nperturbation can cause a different network, that was trained on a different\nsubset of the dataset, to misclassify the same input.},\n archiveprefix = {arXiv},\n author = {Christian Szegedy and Wojciech Zaremba and Ilya Sutskever and Joan Bruna and Dumitru Erhan and Ian Goodfellow and Rob Fergus},\n eprint = {1312.6199v4},\n file = {1312.6199v4.pdf},\n month = {12},\n primaryclass = {cs.CV},\n title = {Intriguing properties of neural networks},\n url = {https://arxiv.org/abs/1312.6199v4},\n year = {2013}\n}\n\n", - "citation_id": "1Fel6Bdb8" - }, - "arxiv:1403.1347": { - "source": "arxiv", - "identifer": "1403.1347", - "standard_citation": "arxiv:1403.1347", - "bibtex": "@article{8t43CQ9m,\n abstract = {Predicting protein secondary structure is a fundamental problem in protein\nstructure prediction. Here we present a new supervised generative stochastic\nnetwork (GSN) based method to predict local secondary structure with deep\nhierarchical representations. GSN is a recently proposed deep learning\ntechnique (Bengio & Thibodeau-Laufer, 2013) to globally train deep generative\nmodel. We present the supervised extension of GSN, which learns a Markov chain\nto sample from a conditional distribution, and applied it to protein structure\nprediction. To scale the model to full-sized, high-dimensional data, like\nprotein sequences with hundreds of amino acids, we introduce a convolutional\narchitecture, which allows efficient learning across multiple layers of\nhierarchical representations. Our architecture uniquely focuses on predicting\nstructured low-level labels informed with both low and high-level\nrepresentations learned by the model. In our application this corresponds to\nlabeling the secondary structure state of each amino-acid residue. We trained\nand tested the model on separate sets of non-homologous proteins sharing less\nthan 30% sequence identity. Our model achieves 66.4% Q8 accuracy on the CB513\ndataset, better than the previously reported best performance 64.9% (Wang et\nal., 2011) for this challenging secondary structure prediction problem.},\n archiveprefix = {arXiv},\n author = {Jian Zhou and Olga G. Troyanskaya},\n eprint = {1403.1347v1},\n file = {1403.1347v1.pdf},\n month = {Mar},\n primaryclass = {q-bio.QM},\n title = {Deep Supervised and Convolutional Generative Stochastic Network for\nProtein Secondary Structure Prediction},\n url = {https://arxiv.org/abs/1403.1347v1},\n year = {2014}\n}\n\n", - "citation_id": "8t43CQ9m" - }, - "arxiv:1411.2581v1": { - "source": "arxiv", - "identifer": "1411.2581v1", - "standard_citation": "arxiv:1411.2581v1", - "bibtex": "@article{pxdeuhMS,\n abstract = {We describe \\textit{deep exponential families} (DEFs), a class of latent\nvariable models that are inspired by the hidden structures used in deep neural\nnetworks. DEFs capture a hierarchy of dependencies between latent variables, and are easily generalized to many settings through exponential families. We\nperform inference using recent \"black box\" variational inference techniques. We\nthen evaluate various DEFs on text and combine multiple DEFs into a model for\npairwise recommendation data. In an extensive study, we show that going beyond\none layer improves predictions for DEFs. We demonstrate that DEFs find\ninteresting exploratory structure in large data sets, and give better\npredictive performance than state-of-the-art models.},\n archiveprefix = {arXiv},\n author = {Rajesh Ranganath and Linpeng Tang and Laurent Charlin and David M. Blei},\n eprint = {1411.2581v1},\n file = {1411.2581v1.pdf},\n month = {Dec},\n primaryclass = {stat.ML},\n title = {Deep Exponential Families},\n url = {https://arxiv.org/abs/1411.2581v1},\n year = {2014}\n}\n\n", - "citation_id": "pxdeuhMS" - }, - "arxiv:1412.6572": { - "source": "arxiv", - "identifer": "1412.6572", - "standard_citation": "arxiv:1412.6572", - "bibtex": "@article{UtcyntjF,\n abstract = {Several machine learning models, including neural networks, consistently\nmisclassify adversarial examples---inputs formed by applying small but\nintentionally worst-case perturbations to examples from the dataset, such that\nthe perturbed input results in the model outputting an incorrect answer with\nhigh confidence. Early attempts at explaining this phenomenon focused on\nnonlinearity and overfitting. We argue instead that the primary cause of neural\nnetworks' vulnerability to adversarial perturbation is their linear nature.\nThis explanation is supported by new quantitative results while giving the\nfirst explanation of the most intriguing fact about them: their generalization\nacross architectures and training sets. Moreover, this view yields a simple and\nfast method of generating adversarial examples. Using this approach to provide\nexamples for adversarial training, we reduce the test set error of a maxout\nnetwork on the MNIST dataset.},\n archiveprefix = {arXiv},\n author = {Ian J. Goodfellow and Jonathon Shlens and Christian Szegedy},\n eprint = {1412.6572v3},\n file = {1412.6572v3.pdf},\n month = {12},\n primaryclass = {stat.ML},\n title = {Explaining and Harnessing Adversarial Examples},\n url = {https://arxiv.org/abs/1412.6572v3},\n year = {2014}\n}\n\n", - "citation_id": "UtcyntjF" - }, - "arxiv:1510.02855": { - "source": "arxiv", - "identifer": "1510.02855", - "standard_citation": "arxiv:1510.02855", - "bibtex": "@article{Z7fd0BYf,\n abstract = {Deep convolutional neural networks comprise a subclass of deep neural\nnetworks (DNN) with a constrained architecture that leverages the spatial and\ntemporal structure of the domain they model. Convolutional networks achieve the\nbest predictive performance in areas such as speech and image recognition by\nhierarchically composing simple local features into complex models. Although\nDNNs have been used in drug discovery for QSAR and ligand-based bioactivity\npredictions, none of these models have benefited from this powerful\nconvolutional architecture. This paper introduces AtomNet, the first\nstructure-based, deep convolutional neural network designed to predict the\nbioactivity of small molecules for drug discovery applications. We demonstrate\nhow to apply the convolutional concepts of feature locality and hierarchical\ncomposition to the modeling of bioactivity and chemical interactions. In\nfurther contrast to existing DNN techniques, we show that AtomNet's application\nof local convolutional filters to structural target information successfully\npredicts new active molecules for targets with no previously known modulators.\nFinally, we show that AtomNet outperforms previous docking approaches on a\ndiverse set of benchmarks by a large margin, achieving an AUC greater than 0.9\non 57.8% of the targets in the DUDE benchmark.},\n archiveprefix = {arXiv},\n author = {Izhar Wallach and Michael Dzamba and Abraham Heifets},\n eprint = {1510.02855v1},\n file = {1510.02855v1.pdf},\n month = {Nov},\n primaryclass = {cs.LG},\n title = {AtomNet: A Deep Convolutional Neural Network for Bioactivity Prediction\nin Structure-based Drug Discovery},\n url = {https://arxiv.org/abs/1510.02855v1},\n year = {2015}\n}\n\n", - "citation_id": "Z7fd0BYf" - }, - "arxiv:1511.02386": { - "source": "arxiv", - "identifer": "1511.02386", - "standard_citation": "arxiv:1511.02386", - "bibtex": "@article{15lbUf0as,\n abstract = {Black box variational inference allows researchers to easily prototype and\nevaluate an array of models. Recent advances allow such algorithms to scale to\nhigh dimensions. However, a central question remains: How to specify an\nexpressive variational distribution that maintains efficient computation? To\naddress this, we develop hierarchical variational models (HVMs). HVMs augment a\nvariational approximation with a prior on its parameters, which allows it to\ncapture complex structure for both discrete and continuous latent variables.\nThe algorithm we develop is black box, can be used for any HVM, and has the\nsame computational efficiency as the original approximation. We study HVMs on a\nvariety of deep discrete latent variable models. HVMs generalize other\nexpressive variational distributions and maintains higher fidelity to the\nposterior.},\n archiveprefix = {arXiv},\n author = {Rajesh Ranganath and Dustin Tran and David M. Blei},\n eprint = {1511.02386v2},\n file = {1511.02386v2.pdf},\n month = {Dec},\n primaryclass = {stat.ML},\n title = {Hierarchical Variational Models},\n url = {https://arxiv.org/abs/1511.02386v2},\n year = {2015}\n}\n\n", - "citation_id": "15lbUf0as" - }, - "arxiv:1602.00357": { - "source": "arxiv", - "identifer": "1602.00357", - "standard_citation": "arxiv:1602.00357", - "bibtex": "@article{HRXii6Ni,\n abstract = {Personalized predictive medicine necessitates the modeling of patient illness\nand care processes, which inherently have long-term temporal dependencies.\nHealthcare observations, recorded in electronic medical records, are episodic\nand irregular in time. We introduce DeepCare, an end-to-end deep dynamic neural\nnetwork that reads medical records, stores previous illness history, infers\ncurrent illness states and predicts future medical outcomes. At the data level, DeepCare represents care episodes as vectors in space, models patient health\nstate trajectories through explicit memory of historical records. Built on Long\nShort-Term Memory (LSTM), DeepCare introduces time parameterizations to handle\nirregular timed events by moderating the forgetting and consolidation of memory\ncells. DeepCare also incorporates medical interventions that change the course\nof illness and shape future medical risk. Moving up to the health state level, historical and present health states are then aggregated through multiscale\ntemporal pooling, before passing through a neural network that estimates future\noutcomes. We demonstrate the efficacy of DeepCare for disease progression\nmodeling, intervention recommendation, and future risk prediction. On two\nimportant cohorts with heavy social and economic burden -- diabetes and mental\nhealth -- the results show improved modeling and risk prediction accuracy.},\n archiveprefix = {arXiv},\n author = {Trang Pham and Truyen Tran and Dinh Phung and Svetha Venkatesh},\n eprint = {1602.00357v2},\n file = {1602.00357v2.pdf},\n month = {Feb},\n primaryclass = {stat.ML},\n title = {DeepCare: A Deep Dynamic Memory Model for Predictive Medicine},\n url = {https://arxiv.org/abs/1602.00357v2},\n year = {2016}\n}\n\n", - "citation_id": "HRXii6Ni" - }, - "arxiv:1605.03661": { - "source": "arxiv", - "identifer": "1605.03661", - "standard_citation": "arxiv:1605.03661", - "bibtex": "@article{173ftiSzF,\n abstract = {Observational studies are rising in importance due to the widespread\naccumulation of data in fields such as healthcare, education, employment and\necology. We consider the task of answering counterfactual questions such as, \"Would this patient have lower blood sugar had she received a different\nmedication?\". We propose a new algorithmic framework for counterfactual\ninference which brings together ideas from domain adaptation and representation\nlearning. In addition to a theoretical justification, we perform an empirical\ncomparison with previous approaches to causal inference from observational\ndata. Our deep learning algorithm significantly outperforms the previous\nstate-of-the-art.},\n archiveprefix = {arXiv},\n author = {Fredrik D. Johansson and Uri Shalit and David Sontag},\n eprint = {1605.03661v2},\n file = {1605.03661v2.pdf},\n month = {May},\n primaryclass = {stat.ML},\n title = {Learning Representations for Counterfactual Inference},\n url = {https://arxiv.org/abs/1605.03661v2},\n year = {2016}\n}\n\n", - "citation_id": "173ftiSzF" - }, - "arxiv:1605.07723": { - "source": "arxiv", - "identifer": "1605.07723", - "standard_citation": "arxiv:1605.07723", - "bibtex": "@article{5Il3kN32,\n abstract = {Large labeled training sets are the critical building blocks of supervised\nlearning methods and are key enablers of deep learning techniques. For some\napplications, creating labeled training sets is the most time-consuming and\nexpensive part of applying machine learning. We therefore propose a paradigm\nfor the programmatic creation of training sets called data programming in which\nusers express weak supervision strategies or domain heuristics as labeling\nfunctions, which are programs that label subsets of the data, but that are\nnoisy and may conflict. We show that by explicitly representing this training\nset labeling process as a generative model, we can \"denoise\" the generated\ntraining set, and establish theoretically that we can recover the parameters of\nthese generative models in a handful of settings. We then show how to modify a\ndiscriminative loss function to make it noise-aware, and demonstrate our method\nover a range of discriminative models including logistic regression and LSTMs.\nExperimentally, on the 2014 TAC-KBP Slot Filling challenge, we show that data\nprogramming would have led to a new winning score, and also show that applying\ndata programming to an LSTM model leads to a TAC-KBP score almost 6 F1 points\nover a state-of-the-art LSTM baseline (and into second place in the\ncompetition). Additionally, in initial user studies we observed that data\nprogramming may be an easier way for non-experts to create machine learning\nmodels when training data is limited or unavailable.},\n archiveprefix = {arXiv},\n author = {Alexander Ratner and Christopher De Sa and Sen Wu and Daniel Selsam and Christopher Ré},\n eprint = {1605.07723v3},\n file = {1605.07723v3.pdf},\n month = {May},\n primaryclass = {stat.ML},\n title = {Data Programming: Creating Large Training Sets, Quickly},\n url = {https://arxiv.org/abs/1605.07723v3},\n year = {2016}\n}\n\n", - "citation_id": "5Il3kN32" - }, - "arxiv:1606.00931": { - "source": "arxiv", - "identifer": "1606.00931", - "standard_citation": "arxiv:1606.00931", - "bibtex": "@article{1FE0F2pQ,\n abstract = {Medical practitioners use survival models to explore and understand the\nrelationships between patients' covariates (e.g. clinical and genetic features)\nand the effectiveness of various treatment options. Standard survival models\nlike the linear Cox proportional hazards model require extensive feature\nengineering or prior medical knowledge to model treatment interaction at an\nindividual level. While nonlinear survival methods, such as neural networks and\nsurvival forests, can inherently model these high-level interaction terms, they\nhave yet to be shown as effective treatment recommender systems. We introduce\nDeepSurv, a Cox proportional hazards deep neural network and state-of-the-art\nsurvival method for modeling interactions between a patient's covariates and\ntreatment effectiveness in order to provide personalized treatment\nrecommendations. We perform a number of experiments training DeepSurv on\nsimulated and real survival data. We demonstrate that DeepSurv performs as well\nas or better than other state-of-the-art survival models and validate that\nDeepSurv successfully models increasingly complex relationships between a\npatient's covariates and their risk of failure. We then show how DeepSurv\nmodels the relationship between a patient's features and effectiveness of\ndifferent treatment options to show how DeepSurv can be used to provide\nindividual treatment recommendations. Finally, we train DeepSurv on real\nclinical studies to demonstrate how it's personalized treatment recommendations\nwould increase the survival time of a set of patients. The predictive and\nmodeling capabilities of DeepSurv will enable medical researchers to use deep\nneural networks as a tool in their exploration, understanding, and prediction\nof the effects of a patient's characteristics on their risk of failure.},\n archiveprefix = {arXiv},\n author = {Jared Katzman and Uri Shaham and Jonathan Bates and Alexander Cloninger and Tingting Jiang and Yuval Kluger},\n eprint = {1606.00931v3},\n file = {1606.00931v3.pdf},\n month = {Jun},\n primaryclass = {stat.ML},\n title = {DeepSurv: Personalized Treatment Recommender System Using A Cox\nProportional Hazards Deep Neural Network},\n url = {https://arxiv.org/abs/1606.00931v3},\n year = {2016}\n}\n\n", - "citation_id": "1FE0F2pQ" - }, - "arxiv:1606.05718": { - "source": "arxiv", - "identifer": "1606.05718", - "standard_citation": "arxiv:1606.05718", - "bibtex": "@article{mbEp6jNr,\n abstract = {The International Symposium on Biomedical Imaging (ISBI) held a grand\nchallenge to evaluate computational systems for the automated detection of\nmetastatic breast cancer in whole slide images of sentinel lymph node biopsies.\nOur team won both competitions in the grand challenge, obtaining an area under\nthe receiver operating curve (AUC) of 0.925 for the task of whole slide image\nclassification and a score of 0.7051 for the tumor localization task. A\npathologist independently reviewed the same images, obtaining a whole slide\nimage classification AUC of 0.966 and a tumor localization score of 0.733.\nCombining our deep learning system's predictions with the human pathologist's\ndiagnoses increased the pathologist's AUC to 0.995, representing an\napproximately 85 percent reduction in human error rate. These results\ndemonstrate the power of using deep learning to produce significant\nimprovements in the accuracy of pathological diagnoses.},\n archiveprefix = {arXiv},\n author = {Dayong Wang and Aditya Khosla and Rishab Gargeya and Humayun Irshad and Andrew H. Beck},\n eprint = {1606.05718v1},\n file = {1606.05718v1.pdf},\n month = {Jun},\n primaryclass = {q-bio.QM},\n title = {Deep Learning for Identifying Metastatic Breast Cancer},\n url = {https://arxiv.org/abs/1606.05718v1},\n year = {2016}\n}\n\n", - "citation_id": "mbEp6jNr" - }, - "arxiv:1606.08813v3": { - "source": "arxiv", - "identifer": "1606.08813v3", - "standard_citation": "arxiv:1606.08813v3", - "bibtex": "@article{7yE9K08a,\n abstract = {We summarize the potential impact that the European Union's new General Data\nProtection Regulation will have on the routine use of machine learning\nalgorithms. Slated to take effect as law across the EU in 2018, it will\nrestrict automated individual decision-making (that is, algorithms that make\ndecisions based on user-level predictors) which \"significantly affect\" users.\nThe law will also effectively create a \"right to explanation,\" whereby a user\ncan ask for an explanation of an algorithmic decision that was made about them.\nWe argue that while this law will pose large challenges for industry, it\nhighlights opportunities for computer scientists to take the lead in designing\nalgorithms and evaluation frameworks which avoid discrimination and enable\nexplanation.},\n archiveprefix = {arXiv},\n author = {Bryce Goodman and Seth Flaxman},\n eprint = {1606.08813v3},\n file = {1606.08813v3.pdf},\n month = {Jun},\n primaryclass = {stat.ML},\n title = {European Union regulations on algorithmic decision-making and a \"right\nto explanation\"},\n url = {https://arxiv.org/abs/1606.08813v3},\n year = {2016}\n}\n\n", - "citation_id": "7yE9K08a" - }, - "arxiv:1607.00133": { - "source": "arxiv", - "identifer": "1607.00133", - "standard_citation": "arxiv:1607.00133", - "bibtex": "@article{ucHUOABT,\n abstract = {Machine learning techniques based on neural networks are achieving remarkable\nresults in a wide variety of domains. Often, the training of models requires\nlarge, representative datasets, which may be crowdsourced and contain sensitive\ninformation. The models should not expose private information in these\ndatasets. Addressing this goal, we develop new algorithmic techniques for\nlearning and a refined analysis of privacy costs within the framework of\ndifferential privacy. Our implementation and experiments demonstrate that we\ncan train deep neural networks with non-convex objectives, under a modest\nprivacy budget, and at a manageable cost in software complexity, training\nefficiency, and model quality.},\n archiveprefix = {arXiv},\n author = {Martín Abadi and Andy Chu and Ian Goodfellow and H. Brendan McMahan and Ilya Mironov and Kunal Talwar and Li Zhang},\n doi = {10.1145/2976749.2978318},\n eprint = {1607.00133v2},\n file = {1607.00133v2.pdf},\n month = {Jul},\n primaryclass = {stat.ML},\n title = {Deep Learning with Differential Privacy},\n url = {https://arxiv.org/abs/1607.00133v2},\n year = {2016}\n}\n\n", - "citation_id": "ucHUOABT" - }, - "arxiv:1607.07519": { - "source": "arxiv", - "identifer": "1607.07519", - "standard_citation": "arxiv:1607.07519", - "bibtex": "@article{Ohd1Q9Xw,\n abstract = {Feature engineering remains a major bottleneck when creating predictive\nsystems from electronic medical records. At present, an important missing\nelement is detecting predictive regular clinical motifs from irregular episodic\nrecords. We present Deepr (short for Deep record), a new end-to-end deep\nlearning system that learns to extract features from medical records and\npredicts future risk automatically. Deepr transforms a record into a sequence\nof discrete elements separated by coded time gaps and hospital transfers. On\ntop of the sequence is a convolutional neural net that detects and combines\npredictive local clinical motifs to stratify the risk. Deepr permits\ntransparent inspection and visualization of its inner working. We validate\nDeepr on hospital data to predict unplanned readmission after discharge. Deepr\nachieves superior accuracy compared to traditional techniques, detects\nmeaningful clinical motifs, and uncovers the underlying structure of the\ndisease and intervention space.},\n archiveprefix = {arXiv},\n author = {Phuoc Nguyen and Truyen Tran and Nilmini Wickramasinghe and Svetha Venkatesh},\n eprint = {1607.07519v1},\n file = {1607.07519v1.pdf},\n month = {Jul},\n primaryclass = {stat.ML},\n title = {Deepr: A Convolutional Net for Medical Records},\n url = {https://arxiv.org/abs/1607.07519v1},\n year = {2016}\n}\n\n", - "citation_id": "Ohd1Q9Xw" - }, - "arxiv:1608.00647": { - "source": "arxiv", - "identifer": "1608.00647", - "standard_citation": "arxiv:1608.00647", - "bibtex": "@article{c6MfDdWP,\n abstract = {Disparate areas of machine learning have benefited from models that can take\nraw data with little preprocessing as input and learn rich representations of\nthat raw data in order to perform well on a given prediction task. We evaluate\nthis approach in healthcare by using longitudinal measurements of lab tests, one of the more raw signals of a patient's health state widely available in\nclinical data, to predict disease onsets. In particular, we train a Long\nShort-Term Memory (LSTM) recurrent neural network and two novel convolutional\nneural networks for multi-task prediction of disease onset for 133 conditions\nbased on 18 common lab tests measured over time in a cohort of 298K patients\nderived from 8 years of administrative claims data. We compare the neural\nnetworks to a logistic regression with several hand-engineered, clinically\nrelevant features. We find that the representation-based learning approaches\nsignificantly outperform this baseline. We believe that our work suggests a new\navenue for patient risk stratification based solely on lab results.},\n archiveprefix = {arXiv},\n author = {Narges Razavian and Jake Marcus and David Sontag},\n eprint = {1608.00647v3},\n file = {1608.00647v3.pdf},\n month = {Aug},\n primaryclass = {cs.LG},\n title = {Multi-task Prediction of Disease Onsets from Longitudinal Lab Tests},\n url = {https://arxiv.org/abs/1608.00647v3},\n year = {2016}\n}\n\n", - "citation_id": "c6MfDdWP" - }, - "arxiv:1608.02158": { - "source": "arxiv", - "identifer": "1608.02158", - "standard_citation": "arxiv:1608.02158", - "bibtex": "@article{qXdO2aMm,\n abstract = {The electronic health record (EHR) provides an unprecedented opportunity to\nbuild actionable tools to support physicians at the point of care. In this\npaper, we investigate survival analysis in the context of EHR data. We\nintroduce deep survival analysis, a hierarchical generative approach to\nsurvival analysis. It departs from previous approaches in two primary ways: (1)\nall observations, including covariates, are modeled jointly conditioned on a\nrich latent structure; and (2) the observations are aligned by their failure\ntime, rather than by an arbitrary time zero as in traditional survival\nanalysis. Further, it (3) scalably handles heterogeneous (continuous and\ndiscrete) data types that occur in the EHR. We validate deep survival analysis\nmodel by stratifying patients according to risk of developing coronary heart\ndisease (CHD). Specifically, we study a dataset of 313,000 patients\ncorresponding to 5.5 million months of observations. When compared to the\nclinically validated Framingham CHD risk score, deep survival analysis is\nsignificantly superior in stratifying patients according to their risk.},\n archiveprefix = {arXiv},\n author = {Rajesh Ranganath and Adler Perotte and Noémie Elhadad and David Blei},\n eprint = {1608.02158v2},\n file = {1608.02158v2.pdf},\n month = {Aug},\n primaryclass = {stat.ML},\n title = {Deep Survival Analysis},\n url = {https://arxiv.org/abs/1608.02158v2},\n year = {2016}\n}\n\n", - "citation_id": "qXdO2aMm" - }, - "arxiv:1609.02943": { - "source": "arxiv", - "identifer": "1609.02943", - "standard_citation": "arxiv:1609.02943", - "bibtex": "@article{ULSPV0rh,\n abstract = {Machine learning (ML) models may be deemed confidential due to their\nsensitive training data, commercial value, or use in security applications.\nIncreasingly often, confidential ML models are being deployed with publicly\naccessible query interfaces. ML-as-a-service (\"predictive analytics\") systems\nare an example: Some allow users to train models on potentially sensitive data\nand charge others for access on a pay-per-query basis.\nThe tension between model confidentiality and public access motivates our\ninvestigation of model extraction attacks. In such attacks, an adversary with\nblack-box access, but no prior knowledge of an ML model's parameters or\ntraining data, aims to duplicate the functionality of (i.e., \"steal\") the\nmodel. Unlike in classical learning theory settings, ML-as-a-service offerings\nmay accept partial feature vectors as inputs and include confidence values with\npredictions. Given these practices, we show simple, efficient attacks that\nextract target ML models with near-perfect fidelity for popular model classes\nincluding logistic regression, neural networks, and decision trees. We\ndemonstrate these attacks against the online services of BigML and Amazon\nMachine Learning. We further show that the natural countermeasure of omitting\nconfidence values from model outputs still admits potentially harmful model\nextraction attacks. Our results highlight the need for careful ML model\ndeployment and new model extraction countermeasures.},\n archiveprefix = {arXiv},\n author = {Florian Tramèr and Fan Zhang and Ari Juels and Michael K. Reiter and Thomas Ristenpart},\n eprint = {1609.02943v2},\n file = {1609.02943v2.pdf},\n month = {Sep},\n primaryclass = {cs.CR},\n title = {Stealing Machine Learning Models via Prediction APIs},\n url = {https://arxiv.org/abs/1609.02943v2},\n year = {2016}\n}\n\n", - "citation_id": "ULSPV0rh" - }, - "arxiv:1609.08144": { - "source": "arxiv", - "identifer": "1609.08144", - "standard_citation": "arxiv:1609.08144", - "bibtex": "@article{4TK06zOf,\n abstract = {Neural Machine Translation (NMT) is an end-to-end learning approach for\nautomated translation, with the potential to overcome many of the weaknesses of\nconventional phrase-based translation systems. Unfortunately, NMT systems are\nknown to be computationally expensive both in training and in translation\ninference. Also, most NMT systems have difficulty with rare words. These issues\nhave hindered NMT's use in practical deployments and services, where both\naccuracy and speed are essential. In this work, we present GNMT, Google's\nNeural Machine Translation system, which attempts to address many of these\nissues. Our model consists of a deep LSTM network with 8 encoder and 8 decoder\nlayers using attention and residual connections. To improve parallelism and\ntherefore decrease training time, our attention mechanism connects the bottom\nlayer of the decoder to the top layer of the encoder. To accelerate the final\ntranslation speed, we employ low-precision arithmetic during inference\ncomputations. To improve handling of rare words, we divide words into a limited\nset of common sub-word units (\"wordpieces\") for both input and output. This\nmethod provides a good balance between the flexibility of \"character\"-delimited\nmodels and the efficiency of \"word\"-delimited models, naturally handles\ntranslation of rare words, and ultimately improves the overall accuracy of the\nsystem. Our beam search technique employs a length-normalization procedure and\nuses a coverage penalty, which encourages generation of an output sentence that\nis most likely to cover all the words in the source sentence. On the WMT'14\nEnglish-to-French and English-to-German benchmarks, GNMT achieves competitive\nresults to state-of-the-art. Using a human side-by-side evaluation on a set of\nisolated simple sentences, it reduces translation errors by an average of 60%\ncompared to Google's phrase-based production system.},\n archiveprefix = {arXiv},\n author = {Yonghui Wu and Mike Schuster and Zhifeng Chen and Quoc V. Le and Mohammad Norouzi and Wolfgang Macherey and Maxim Krikun and Yuan Cao and Qin Gao and Klaus Macherey and Jeff Klingner and Apurva Shah and Melvin Johnson and Xiaobing Liu and Łukasz Kaiser and Stephan Gouws and Yoshikiyo Kato and Taku Kudo and Hideto Kazawa and Keith Stevens and George Kurian and Nishant Patil and Wei Wang and Cliff Young and Jason Smith and Jason Riesa and Alex Rudnick and Oriol Vinyals and Greg Corrado and Macduff Hughes and Jeffrey Dean},\n eprint = {1609.08144v2},\n file = {1609.08144v2.pdf},\n month = {Sep},\n primaryclass = {cs.CL},\n title = {Google's Neural Machine Translation System: Bridging the Gap between\nHuman and Machine Translation},\n url = {https://arxiv.org/abs/1609.08144v2},\n year = {2016}\n}\n\n", - "citation_id": "4TK06zOf" - }, - "arxiv:1610.02413": { - "source": "arxiv", - "identifer": "1610.02413", - "standard_citation": "arxiv:1610.02413", - "bibtex": "@article{1ENxzq6pT,\n abstract = {We propose a criterion for discrimination against a specified sensitive\nattribute in supervised learning, where the goal is to predict some target\nbased on available features. Assuming data about the predictor, target, and\nmembership in the protected group are available, we show how to optimally\nadjust any learned predictor so as to remove discrimination according to our\ndefinition. Our framework also improves incentives by shifting the cost of poor\nclassification from disadvantaged groups to the decision maker, who can respond\nby improving the classification accuracy.\nIn line with other studies, our notion is oblivious: it depends only on the\njoint statistics of the predictor, the target and the protected attribute, but\nnot on interpretation of individualfeatures. We study the inherent limits of\ndefining and identifying biases based on such oblivious measures, outlining\nwhat can and cannot be inferred from different oblivious tests.\nWe illustrate our notion using a case study of FICO credit scores.},\n archiveprefix = {arXiv},\n author = {Moritz Hardt and Eric Price and Nathan Srebro},\n eprint = {1610.02413v1},\n file = {1610.02413v1.pdf},\n month = {Nov},\n primaryclass = {cs.LG},\n title = {Equality of Opportunity in Supervised Learning},\n url = {https://arxiv.org/abs/1610.02413v1},\n year = {2016}\n}\n\n", - "citation_id": "1ENxzq6pT" - }, - "arxiv:1610.05256": { - "source": "arxiv", - "identifer": "1610.05256", - "standard_citation": "arxiv:1610.05256", - "bibtex": "@article{M2OLWojE,\n abstract = {Conversational speech recognition has served as a flagship speech recognition\ntask since the release of the Switchboard corpus in the 1990s. In this paper, we measure the human error rate on the widely used NIST 2000 test set, and find\nthat our latest automated system has reached human parity. The error rate of\nprofessional transcribers is 5.9% for the Switchboard portion of the data, in\nwhich newly acquainted pairs of people discuss an assigned topic, and 11.3% for\nthe CallHome portion where friends and family members have open-ended\nconversations. In both cases, our automated system establishes a new state of\nthe art, and edges past the human benchmark, achieving error rates of 5.8% and\n11.0%, respectively. The key to our system's performance is the use of various\nconvolutional and LSTM acoustic model architectures, combined with a novel\nspatial smoothing method and lattice-free MMI acoustic training, multiple\nrecurrent neural network language modeling approaches, and a systematic use of\nsystem combination.},\n archiveprefix = {arXiv},\n author = {W. Xiong and J. Droppo and X. Huang and F. Seide and M. Seltzer and A. Stolcke and D. Yu and G. Zweig},\n eprint = {1610.05256v2},\n file = {1610.05256v2.pdf},\n month = {Nov},\n primaryclass = {cs.CL},\n title = {Achieving Human Parity in Conversational Speech Recognition},\n url = {https://arxiv.org/abs/1610.05256v2},\n year = {2016}\n}\n\n", - "citation_id": "M2OLWojE" - }, - "arxiv:1610.05820": { - "source": "arxiv", - "identifer": "1610.05820", - "standard_citation": "arxiv:1610.05820", - "bibtex": "@article{1HbRTExaU,\n abstract = {We quantitatively investigate how machine learning models leak information\nabout the individual data records on which they were trained. We focus on the\nbasic membership inference attack: given a data record and black-box access to\na model, determine if the record was in the model's training dataset. To\nperform membership inference against a target model, we make adversarial use of\nmachine learning and train our own inference model to recognize differences in\nthe target model's predictions on the inputs that it trained on versus the\ninputs that it did not train on.\nWe empirically evaluate our inference techniques on classification models\ntrained by commercial \"machine learning as a service\" providers such as Google\nand Amazon. Using realistic datasets and classification tasks, including a\nhospital discharge dataset whose membership is sensitive from the privacy\nperspective, we show that these models can be vulnerable to membership\ninference attacks. We then investigate the factors that influence this leakage\nand evaluate mitigation strategies.},\n archiveprefix = {arXiv},\n author = {Reza Shokri and Marco Stronati and Congzheng Song and Vitaly Shmatikov},\n eprint = {1610.05820v2},\n file = {1610.05820v2.pdf},\n month = {Nov},\n primaryclass = {cs.CR},\n title = {Membership Inference Attacks against Machine Learning Models},\n url = {https://arxiv.org/abs/1610.05820v2},\n year = {2016}\n}\n\n", - "citation_id": "1HbRTExaU" - }, - "arxiv:1610.09559": { - "source": "arxiv", - "identifer": "1610.09559", - "standard_citation": "arxiv:1610.09559", - "bibtex": "@article{11aqfNfQx,\n abstract = {We study fairness in linear bandit problems. Starting from the notion of\nmeritocratic fairness introduced in Joseph et al. [2016], we carry out a more\nrefined analysis of a more general problem, achieving better performance\nguarantees with fewer modelling assumptions on the number and structure of\navailable choices as well as the number selected. We also analyze the\npreviously-unstudied question of fairness in infinite linear bandit problems, obtaining instance-dependent regret upper bounds as well as lower bounds\ndemonstrating that this instance-dependence is necessary. The result is a\nframework for meritocratic fairness in an online linear setting that is\nsubstantially more powerful, general, and realistic than the current state of\nthe art.},\n archiveprefix = {arXiv},\n author = {Matthew Joseph and Michael Kearns and Jamie Morgenstern and Seth Neel and Aaron Roth},\n eprint = {1610.09559v4},\n file = {1610.09559v4.pdf},\n month = {Nov},\n primaryclass = {cs.LG},\n title = {Fair Algorithms for Infinite and Contextual Bandits},\n url = {https://arxiv.org/abs/1610.09559v4},\n year = {2016}\n}\n\n", - "citation_id": "11aqfNfQx" - }, - "arxiv:1611.02796": { - "source": "arxiv", - "identifer": "1611.02796", - "standard_citation": "arxiv:1611.02796", - "bibtex": "@article{lERqKdZJ,\n abstract = {This paper proposes a general method for improving the structure and quality\nof sequences generated by a recurrent neural network (RNN), while maintaining\ninformation originally learned from data, as well as sample diversity. An RNN\nis first pre-trained on data using maximum likelihood estimation (MLE), and the\nprobability distribution over the next token in the sequence learned by this\nmodel is treated as a prior policy. Another RNN is then trained using\nreinforcement learning (RL) to generate higher-quality outputs that account for\ndomain-specific incentives while retaining proximity to the prior policy of the\nMLE RNN. To formalize this objective, we derive novel off-policy RL methods for\nRNNs from KL-control. The effectiveness of the approach is demonstrated on two\napplications; 1) generating novel musical melodies, and 2) computational\nmolecular generation. For both problems, we show that the proposed method\nimproves the desired properties and structure of the generated sequences, while\nmaintaining information learned from data.},\n archiveprefix = {arXiv},\n author = {Natasha Jaques and Shixiang Gu and Dzmitry Bahdanau and José Miguel Hernández-Lobato and Richard E. Turner and Douglas Eck},\n eprint = {1611.02796v8},\n file = {1611.02796v8.pdf},\n month = {Dec},\n primaryclass = {cs.LG},\n title = {Sequence Tutor: Conservative Fine-Tuning of Sequence Generation Models\nwith KL-control},\n url = {https://arxiv.org/abs/1611.02796v8},\n year = {2016}\n}\n\n", - "citation_id": "lERqKdZJ" - }, - "arxiv:1611.03814": { - "source": "arxiv", - "identifer": "1611.03814", - "standard_citation": "arxiv:1611.03814", - "bibtex": "@article{AsLAb71x,\n abstract = {Advances in machine learning (ML) in recent years have enabled a dizzying\narray of applications such as data analytics, autonomous systems, and security\ndiagnostics. ML is now pervasive---new systems and models are being deployed in\nevery domain imaginable, leading to rapid and widespread deployment of software\nbased inference and decision making. There is growing recognition that ML\nexposes new vulnerabilities in software systems, yet the technical community's\nunderstanding of the nature and extent of these vulnerabilities remains\nlimited. We systematize recent findings on ML security and privacy, focusing on\nattacks identified on these systems and defenses crafted to date. We articulate\na comprehensive threat model for ML, and categorize attacks and defenses within\nan adversarial framework. Key insights resulting from works both in the ML and\nsecurity communities are identified and the effectiveness of approaches are\nrelated to structural elements of ML algorithms and the data used to train\nthem. We conclude by formally exploring the opposing relationship between model\naccuracy and resilience to adversarial manipulation. Through these\nexplorations, we show that there are (possibly unavoidable) tensions between\nmodel complexity, accuracy, and resilience that must be calibrated for the\nenvironments in which they will be used.},\n archiveprefix = {arXiv},\n author = {Nicolas Papernot and Patrick McDaniel and Arunesh Sinha and Michael Wellman},\n eprint = {1611.03814v1},\n file = {1611.03814v1.pdf},\n month = {Dec},\n primaryclass = {cs.CR},\n title = {Towards the Science of Security and Privacy in Machine Learning},\n url = {https://arxiv.org/abs/1611.03814v1},\n year = {2016}\n}\n\n", - "citation_id": "AsLAb71x" - }, - "arxiv:1611.08373": { - "source": "arxiv", - "identifer": "1611.08373", - "standard_citation": "arxiv:1611.08373", - "bibtex": "@article{dO844vZn,\n abstract = {Automated extraction of concepts from patient clinical records is an\nessential facilitator of clinical research. For this reason, the 2010 i2b2/VA\nNatural Language Processing Challenges for Clinical Records introduced a\nconcept extraction task aimed at identifying and classifying concepts into\npredefined categories (i.e., treatments, tests and problems). State-of-the-art\nconcept extraction approaches heavily rely on handcrafted features and\ndomain-specific resources which are hard to collect and define. For this\nreason, this paper proposes an alternative, streamlined approach: a recurrent\nneural network (the bidirectional LSTM with CRF decoding) initialized with\ngeneral-purpose, off-the-shelf word embeddings. The experimental results\nachieved on the 2010 i2b2/VA reference corpora using the proposed framework\noutperform all recent methods and ranks closely to the best submission from the\noriginal 2010 i2b2/VA challenge.},\n archiveprefix = {arXiv},\n author = {Raghavendra Chalapathy and Ehsan Zare Borzeshi and Massimo Piccardi},\n eprint = {1611.08373v1},\n file = {1611.08373v1.pdf},\n month = {Dec},\n primaryclass = {stat.ML},\n title = {Bidirectional LSTM-CRF for Clinical Concept Extraction},\n url = {https://arxiv.org/abs/1611.08373v1},\n year = {2016}\n}\n\n", - "citation_id": "dO844vZn" - }, - "arxiv:1701.06599": { - "source": "arxiv", - "identifer": "1701.06599", - "standard_citation": "arxiv:1701.06599", - "bibtex": "@article{apBChoyF,\n abstract = {The recent rapid and tremendous success of deep convolutional neural networks\n(CNN) on many challenging computer vision tasks largely derives from the\naccessibility of the well-annotated ImageNet and PASCAL VOC datasets.\nNevertheless, unsupervised image categorization (i.e., without the ground-truth\nlabeling) is much less investigated, yet critically important and difficult\nwhen annotations are extremely hard to obtain in the conventional way of\n\"Google Search\" and crowd sourcing. We address this problem by presenting a\nlooped deep pseudo-task optimization (LDPO) framework for joint mining of deep\nCNN features and image labels. Our method is conceptually simple and rests upon\nthe hypothesized \"convergence\" of better labels leading to better trained CNN\nmodels which in turn feed more discriminative image representations to\nfacilitate more meaningful clusters/labels. Our proposed method is validated in\ntackling two important applications: 1) Large-scale medical image annotation\nhas always been a prohibitively expensive and easily-biased task even for\nwell-trained radiologists. Significantly better image categorization results\nare achieved via our proposed approach compared to the previous\nstate-of-the-art method. 2) Unsupervised scene recognition on representative\nand publicly available datasets with our proposed technique is examined. The\nLDPO achieves excellent quantitative scene classification results. On the MIT\nindoor scene dataset, it attains a clustering accuracy of 75.3%, compared to\nthe state-of-the-art supervised classification accuracy of 81.0% (when both are\nbased on the VGG-VD model).},\n archiveprefix = {arXiv},\n author = {Xiaosong Wang and Le Lu and Hoo-chang Shin and Lauren Kim and Mohammadhadi Bagheri and Isabella Nogues and Jianhua Yao and Ronald M. Summers},\n eprint = {1701.06599v1},\n file = {1701.06599v1.pdf},\n month = {Jan},\n primaryclass = {cs.CV},\n title = {Unsupervised Joint Mining of Deep Features and Image Labels for\nLarge-scale Radiology Image Categorization and Scene Recognition},\n url = {https://arxiv.org/abs/1701.06599v1},\n year = {2017}\n}\n\n", - "citation_id": "apBChoyF" - }, - "arxiv:1703.01925": { - "source": "arxiv", - "identifer": "1703.01925", - "standard_citation": "arxiv:1703.01925", - "bibtex": "@article{AQ3N6Ayw,\n abstract = {Deep generative models have been wildly successful at learning coherent\nlatent representations for continuous data such as video and audio. However, generative modeling of discrete data such as arithmetic expressions and\nmolecular structures still poses significant challenges. Crucially, state-of-the-art methods often produce outputs that are not valid. We make the\nkey observation that frequently, discrete data can be represented as a parse\ntree from a context-free grammar. We propose a variational autoencoder which\nencodes and decodes directly to and from these parse trees, ensuring the\ngenerated outputs are always valid. Surprisingly, we show that not only does\nour model more often generate valid outputs, it also learns a more coherent\nlatent space in which nearby points decode to similar discrete outputs. We\ndemonstrate the effectiveness of our learned models by showing their improved\nperformance in Bayesian optimization for symbolic regression and molecular\nsynthesis.},\n archiveprefix = {arXiv},\n author = {Matt J. Kusner and Brooks Paige and José Miguel Hernández-Lobato},\n eprint = {1703.01925v1},\n file = {1703.01925v1.pdf},\n month = {Mar},\n primaryclass = {stat.ML},\n title = {Grammar Variational Autoencoder},\n url = {https://arxiv.org/abs/1703.01925v1},\n year = {2017}\n}\n\n", - "citation_id": "AQ3N6Ayw" - }, - "arxiv:1703.02136": { - "source": "arxiv", - "identifer": "1703.02136", - "standard_citation": "arxiv:1703.02136", - "bibtex": "@article{wKioubsT,\n abstract = {One of the most difficult speech recognition tasks is accurate recognition of\nhuman to human communication. Advances in deep learning over the last few years\nhave produced major speech recognition improvements on the representative\nSwitchboard conversational corpus. Word error rates that just a few years ago\nwere 14% have dropped to 8.0%, then 6.6% and most recently 5.8%, and are now\nbelieved to be within striking range of human performance. This then raises two\nissues - what IS human performance, and how far down can we still drive speech\nrecognition error rates? A recent paper by Microsoft suggests that we have\nalready achieved human performance. In trying to verify this statement, we\nperformed an independent set of human performance measurements on two\nconversational tasks and found that human performance may be considerably\nbetter than what was earlier reported, giving the community a significantly\nharder goal to achieve. We also report on our own efforts in this area, presenting a set of acoustic and language modeling techniques that lowered the\nword error rate of our own English conversational telephone LVCSR system to the\nlevel of 5.5%/10.3% on the Switchboard/CallHome subsets of the Hub5 2000\nevaluation, which - at least at the writing of this paper - is a new\nperformance milestone (albeit not at what we measure to be human performance!).\nOn the acoustic side, we use a score fusion of three models: one LSTM with\nmultiple feature inputs, a second LSTM trained with speaker-adversarial\nmulti-task learning and a third residual net (ResNet) with 25 convolutional\nlayers and time-dilated convolutions. On the language modeling side, we use\nword and character LSTMs and convolutional WaveNet-style language models.},\n archiveprefix = {arXiv},\n author = {George Saon and Gakuto Kurata and Tom Sercu and Kartik Audhkhasi and Samuel Thomas and Dimitrios Dimitriadis and Xiaodong Cui and Bhuvana Ramabhadran and Michael Picheny and Lynn-Li Lim and Bergul Roomi and Phil Hall},\n eprint = {1703.02136v1},\n file = {1703.02136v1.pdf},\n month = {Mar},\n primaryclass = {cs.CL},\n title = {English Conversational Telephone Speech Recognition by Humans and\nMachines},\n url = {https://arxiv.org/abs/1703.02136v1},\n year = {2017}\n}\n\n", - "citation_id": "wKioubsT" - }, - "arxiv:1703.06490v1": { - "source": "arxiv", - "identifer": "1703.06490v1", - "standard_citation": "arxiv:1703.06490v1", - "bibtex": "@article{xl1ijigK,\n abstract = {Access to electronic health records (EHR) data has motivated computational\nadvances in medical research. However, various concerns, particularly over\nprivacy, can limit access to and collaborative use of EHR data. Sharing\nsynthetic EHR data could mitigate risk. In this paper, we propose a new\napproach, medical Generative Adversarial Network (medGAN), to generate\nrealistic synthetic EHRs. Based on an input EHR dataset, medGAN can generate\nhigh-dimensional discrete variables (e.g., binary and count features) via a\ncombination of an autoencoder and generative adversarial networks. We also\npropose minibatch averaging to efficiently avoid mode collapse, and increase\nthe learning efficiency with batch normalization and shortcut connections. To\ndemonstrate feasibility, we showed that medGAN generates synthetic EHR datasets\nthat achieve comparable performance to real data on many experiments including\ndistribution statistics, predictive modeling tasks and medical expert review.},\n archiveprefix = {arXiv},\n author = {Edward Choi and Siddharth Biswal and Bradley Malin and Jon Duke and Walter F. Stewart and Jimeng Sun},\n eprint = {1703.06490v1},\n file = {1703.06490v1.pdf},\n month = {Mar},\n primaryclass = {cs.LG},\n title = {Generating Multi-label Discrete Electronic Health Records using\nGenerative Adversarial Networks},\n url = {https://arxiv.org/abs/1703.06490v1},\n year = {2017}\n}\n\n", - "citation_id": "xl1ijigK" - }, - "arxiv:1703.10603": { - "source": "arxiv", - "identifer": "1703.10603", - "standard_citation": "arxiv:1703.10603", - "bibtex": "@article{17YaKNLKk,\n abstract = {Empirical scoring functions based on either molecular force fields or\ncheminformatics descriptors are widely used, in conjunction with molecular\ndocking, during the early stages of drug discovery to predict potency and\nbinding affinity of a drug-like molecule to a given target. These models\nrequire expert-level knowledge of physical chemistry and biology to be encoded\nas hand-tuned parameters or features rather than allowing the underlying model\nto select features in a data-driven procedure. Here, we develop a general\n3-dimensional spatial convolution operation for learning atomic-level chemical\ninteractions directly from atomic coordinates and demonstrate its application\nto structure-based bioactivity prediction. The atomic convolutional neural\nnetwork is trained to predict the experimentally determined binding affinity of\na protein-ligand complex by direct calculation of the energy associated with\nthe complex, protein, and ligand given the crystal structure of the binding\npose. Non-covalent interactions present in the complex that are absent in the\nprotein-ligand sub-structures are identified and the model learns the\ninteraction strength associated with these features. We test our model by\npredicting the binding free energy of a subset of protein-ligand complexes\nfound in the PDBBind dataset and compare with state-of-the-art cheminformatics\nand machine learning-based approaches. We find that all methods achieve\nexperimental accuracy and that atomic convolutional networks either outperform\nor perform competitively with the cheminformatics based methods. Unlike all\nprevious protein-ligand prediction systems, atomic convolutional networks are\nend-to-end and fully-differentiable. They represent a new data-driven, physics-based deep learning model paradigm that offers a strong foundation for\nfuture improvements in structure-based bioactivity prediction.},\n archiveprefix = {arXiv},\n author = {Joseph Gomes and Bharath Ramsundar and Evan N. Feinberg and Vijay S. Pande},\n eprint = {1703.10603v1},\n file = {1703.10603v1.pdf},\n month = {Mar},\n primaryclass = {cs.LG},\n title = {Atomic Convolutional Networks for Predicting Protein-Ligand Binding\nAffinity},\n url = {https://arxiv.org/abs/1703.10603v1},\n year = {2017}\n}\n\n", - "citation_id": "17YaKNLKk" - }, - "arxiv:1704.01155": { - "source": "arxiv", - "identifer": "1704.01155", - "standard_citation": "arxiv:1704.01155", - "bibtex": "@article{18lZK7fxH,\n abstract = {Although deep neural networks (DNNs) have achieved great success in many\ncomputer vision tasks, recent studies have shown they are vulnerable to\nadversarial examples. Such examples, typically generated by adding small but\npurposeful distortions, can frequently fool DNN models. Previous studies to\ndefend against adversarial examples mostly focused on refining the DNN models.\nThey have either shown limited success or suffer from the expensive\ncomputation. We propose a new strategy, \\emph{feature squeezing}, that can be\nused to harden DNN models by detecting adversarial examples. Feature squeezing\nreduces the search space available to an adversary by coalescing samples that\ncorrespond to many different feature vectors in the original space into a\nsingle sample. By comparing a DNN model's prediction on the original input with\nthat on the squeezed input, feature squeezing detects adversarial examples with\nhigh accuracy and few false positives. This paper explores two instances of\nfeature squeezing: reducing the color bit depth of each pixel and smoothing\nusing a spatial filter. These strategies are straightforward, inexpensive, and\ncomplementary to defensive methods that operate on the underlying model, such\nas adversarial training.},\n archiveprefix = {arXiv},\n author = {Weilin Xu and David Evans and Yanjun Qi},\n eprint = {1704.01155v1},\n file = {1704.01155v1.pdf},\n month = {Apr},\n primaryclass = {cs.CV},\n title = {Feature Squeezing: Detecting Adversarial Examples in Deep Neural\nNetworks},\n url = {https://arxiv.org/abs/1704.01155v1},\n year = {2017}\n}\n\n", - "citation_id": "18lZK7fxH" - }, - "arxiv:1704.04760": { - "source": "arxiv", - "identifer": "1704.04760", - "standard_citation": "arxiv:1704.04760", - "bibtex": "@article{ULagTifF,\n abstract = {Many architects believe that major improvements in cost-energy-performance\nmust now come from domain-specific hardware. This paper evaluates a custom\nASIC---called a Tensor Processing Unit (TPU)---deployed in datacenters since\n2015 that accelerates the inference phase of neural networks (NN). The heart of\nthe TPU is a 65,536 8-bit MAC matrix multiply unit that offers a peak\nthroughput of 92 TeraOps/second (TOPS) and a large (28 MiB) software-managed\non-chip memory. The TPU's deterministic execution model is a better match to\nthe 99th-percentile response-time requirement of our NN applications than are\nthe time-varying optimizations of CPUs and GPUs (caches, out-of-order\nexecution, multithreading, multiprocessing, prefetching, ...) that help average\nthroughput more than guaranteed latency. The lack of such features helps\nexplain why, despite having myriad MACs and a big memory, the TPU is relatively\nsmall and low power. We compare the TPU to a server-class Intel Haswell CPU and\nan Nvidia K80 GPU, which are contemporaries deployed in the same datacenters.\nOur workload, written in the high-level TensorFlow framework, uses production\nNN applications (MLPs, CNNs, and LSTMs) that represent 95% of our datacenters'\nNN inference demand. Despite low utilization for some applications, the TPU is\non average about 15X - 30X faster than its contemporary GPU or CPU, with\nTOPS/Watt about 30X - 80X higher. Moreover, using the GPU's GDDR5 memory in the\nTPU would triple achieved TOPS and raise TOPS/Watt to nearly 70X the GPU and\n200X the CPU.},\n archiveprefix = {arXiv},\n author = {Norman P. Jouppi and Cliff Young and Nishant Patil and David Patterson and Gaurav Agrawal and Raminder Bajwa and Sarah Bates and Suresh Bhatia and Nan Boden and Al Borchers and Rick Boyle and Pierre-luc Cantin and Clifford Chao and Chris Clark and Jeremy Coriell and Mike Daley and Matt Dau and Jeffrey Dean and Ben Gelb and Tara Vazir Ghaemmaghami and Rajendra Gottipati and William Gulland and Robert Hagmann and C. Richard Ho and Doug Hogberg and John Hu and Robert Hundt and Dan Hurt and Julian Ibarz and Aaron Jaffey and Alek Jaworski and Alexander Kaplan and Harshit Khaitan and Andy Koch and Naveen Kumar and Steve Lacy and James Laudon and James Law and Diemthu Le and Chris Leary and Zhuyuan Liu and Kyle Lucke and Alan Lundin and Gordon MacKean and Adriana Maggiore and Maire Mahony and Kieran Miller and Rahul Nagarajan and Ravi Narayanaswami and Ray Ni and Kathy Nix and Thomas Norrie and Mark Omernick and Narayana Penukonda and Andy Phelps and Jonathan Ross and Matt Ross and Amir Salek and Emad Samadiani and Chris Severn and Gregory Sizikov and Matthew Snelham and Jed Souter and Dan Steinberg and Andy Swing and Mercedes Tan and Gregory Thorson and Bo Tian and Horia Toma and Erick Tuttle and Vijay Vasudevan and Richard Walter and Walter Wang and Eric Wilcox and Doe Hyun Yoon},\n eprint = {1704.04760v1},\n file = {1704.04760v1.pdf},\n month = {Apr},\n primaryclass = {cs.AR},\n title = {In-Datacenter Performance Analysis of a Tensor Processing Unit},\n url = {https://arxiv.org/abs/1704.04760v1},\n year = {2017}\n}\n\n", - "citation_id": "ULagTifF" - }, - "arxiv:1704.07207": { - "source": "arxiv", - "identifer": "1704.07207", - "standard_citation": "arxiv:1704.07207", - "bibtex": "@article{39RPiE10,\n abstract = {Computational prediction of membrane protein (MP) structures is very\nchallenging partially due to lack of sufficient solved structures for homology\nmodeling. Recently direct evolutionary coupling analysis (DCA) sheds some light\non protein contact prediction and accordingly, contact-assisted folding, but\nDCA is effective only on some very large-sized families since it uses\ninformation only in a single protein family. This paper presents a deep\ntransfer learning method that can significantly improve MP contact prediction\nby learning contact patterns and complex sequence-contact relationship from\nthousands of non-membrane proteins (non-MPs). Tested on 510 non-redundant MPs, our deep model (learned from only non-MPs) has top L/10 long-range contact\nprediction accuracy 0.69, better than our deep model trained by only MPs (0.63)\nand much better than a representative DCA method CCMpred (0.47) and the CASP11\nwinner MetaPSICOV (0.55). The accuracy of our deep model can be further\nimproved to 0.72 when trained by a mix of non-MPs and MPs. When only contacts\nin transmembrane regions are evaluated, our method has top L/10 long-range\naccuracy 0.62, 0.57, and 0.53 when trained by a mix of non-MPs and MPs, by\nnon-MPs only, and by MPs only, respectively, still much better than MetaPSICOV\n(0.45) and CCMpred (0.40). All these results suggest that sequence-structure\nrelationship learned by our deep model from non-MPs generalizes well to MP\ncontact prediction. Improved contact prediction also leads to better\ncontact-assisted folding. Using only top predicted contacts as restraints, our\ndeep learning method can fold 160 and 200 of 510 MPs with TMscore>0.6 when\ntrained by non-MPs only and by a mix of non-MPs and MPs, respectively, while\nCCMpred and MetaPSICOV can do so for only 56 and 77 MPs, respectively. Our\ncontact-assisted folding also greatly outperforms homology modeling.},\n archiveprefix = {arXiv},\n author = {Zhen Li and Sheng Wang and Yizhou Yu and Jinbo Xu},\n eprint = {1704.07207v1},\n file = {1704.07207v1.pdf},\n month = {Apr},\n primaryclass = {q-bio.BM},\n title = {Predicting membrane protein contacts from non-membrane proteins by deep\ntransfer learning},\n url = {https://arxiv.org/abs/1704.07207v1},\n year = {2017}\n}\n\n", - "citation_id": "39RPiE10" - }, - "arxiv:1705.02315": { - "source": "arxiv", - "identifer": "1705.02315", - "standard_citation": "arxiv:1705.02315", - "bibtex": "@article{PGi9g7yV,\n abstract = {The chest X-ray is one of the most commonly accessible radiological\nexaminations for screening and diagnosis of many lung diseases. A tremendous\nnumber of X-ray imaging studies accompanied by radiological reports are\naccumulated and stored in many modern hospitals' Picture Archiving and\nCommunication Systems (PACS). On the other side, it is still an open question\nhow this type of hospital-size knowledge database containing invaluable imaging\ninformatics (i.e., loosely labeled) can be used to facilitate the data-hungry\ndeep learning paradigms in building truly large-scale high precision\ncomputer-aided diagnosis (CAD) systems.\nIn this paper, we present a new chest X-ray database, namely \"ChestX-ray8\", which comprises 108,948 frontal-view X-ray images of 32,717 unique patients\nwith the text-mined eight disease image labels (where each image can have\nmulti-labels), from the associated radiological reports using natural language\nprocessing. Importantly, we demonstrate that these commonly occurring thoracic\ndiseases can be detected and even spatially-located via a unified\nweakly-supervised multi-label image classification and disease localization\nframework, which is validated using our proposed dataset. Although the initial\nquantitative results are promising as reported, deep convolutional neural\nnetwork based \"reading chest X-rays\" (i.e., recognizing and locating the common\ndisease patterns trained with only image-level labels) remains a strenuous task\nfor fully-automated high precision CAD systems. Data download link:\nhttps://nihcc.app.box.com/v/ChestXray-NIHCC},\n archiveprefix = {arXiv},\n author = {Xiaosong Wang and Yifan Peng and Le Lu and Zhiyong Lu and Mohammadhadi Bagheri and Ronald M. Summers},\n eprint = {1705.02315v4},\n file = {1705.02315v4.pdf},\n month = {May},\n primaryclass = {cs.CV},\n title = {ChestX-ray8: Hospital-scale Chest X-ray Database and Benchmarks on\nWeakly-Supervised Classification and Localization of Common Thorax Diseases},\n url = {https://arxiv.org/abs/1705.02315v4},\n year = {2017}\n}\n\n", - "citation_id": "PGi9g7yV" - }, - "doi:10.1001/jama.2016.17216": { - "source": "doi", - "identifer": "10.1001/jama.2016.17216", - "standard_citation": "doi:10.1001/jama.2016.17216", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T11:02:15Z", - "timestamp": 1507114935019 - }, - "reference-count": 0, - "publisher": "American Medical Association (AMA)", - "issue": "22", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 12, - 13 - ] - ] - }, - "DOI": "10.1001/jama.2016.17216", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 11, - 29 - ] - ], - "date-time": "2016-11-29T17:33:53Z", - "timestamp": 1480440833000 - }, - "page": "2402", - "source": "Crossref", - "is-referenced-by-count": 51, - "title": "Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs", - "prefix": "10.1001", - "volume": "316", - "author": [ - { - "given": "Varun", - "family": "Gulshan", - "affiliation": [ - { - "name": "Google Inc, Mountain View, California" - } - ] - }, - { - "given": "Lily", - "family": "Peng", - "affiliation": [ - { - "name": "Google Inc, Mountain View, California" - } - ] - }, - { - "given": "Marc", - "family": "Coram", - "affiliation": [ - { - "name": "Google Inc, Mountain View, California" - } - ] - }, - { - "given": "Martin C.", - "family": "Stumpe", - "affiliation": [ - { - "name": "Google Inc, Mountain View, California" - } - ] - }, - { - "given": "Derek", - "family": "Wu", - "affiliation": [ - { - "name": "Google Inc, Mountain View, California" - } - ] - }, - { - "given": "Arunachalam", - "family": "Narayanaswamy", - "affiliation": [ - { - "name": "Google Inc, Mountain View, California" - } - ] - }, - { - "given": "Subhashini", - "family": "Venugopalan", - "affiliation": [ - { - "name": "Google Inc, Mountain View, California2Department of Computer Science, University of Texas, Austin" - } - ] - }, - { - "given": "Kasumi", - "family": "Widner", - "affiliation": [ - { - "name": "Google Inc, Mountain View, California" - } - ] - }, - { - "given": "Tom", - "family": "Madams", - "affiliation": [ - { - "name": "Google Inc, Mountain View, California" - } - ] - }, - { - "given": "Jorge", - "family": "Cuadros", - "affiliation": [ - { - "name": "EyePACS LLC, San Jose, California4School of Optometry, Vision Science Graduate Group, University of California, Berkeley" - } - ] - }, - { - "given": "Ramasamy", - "family": "Kim", - "affiliation": [ - { - "name": "Aravind Medical Research Foundation, Aravind Eye Care System, Madurai, India" - } - ] - }, - { - "given": "Rajiv", - "family": "Raman", - "affiliation": [ - { - "name": "Shri Bhagwan Mahavir Vitreoretinal Services, Sankara Nethralaya, Chennai, Tamil Nadu, India" - } - ] - }, - { - "given": "Philip C.", - "family": "Nelson", - "affiliation": [ - { - "name": "Google Inc, Mountain View, California" - } - ] - }, - { - "given": "Jessica L.", - "family": "Mega", - "affiliation": [ - { - "name": "Verily Life Sciences, Mountain View, California8Cardiovascular Division, Department of Medicine, Brigham and Women’s Hospital and Harvard Medical School, Boston, Massachusetts" - } - ] - }, - { - "given": "Dale R.", - "family": "Webster", - "affiliation": [ - { - "name": "Google Inc, Mountain View, California" - } - ] - } - ], - "member": "10", - "container-title": "JAMA", - "original-title": [], - "link": [ - { - "URL": "http://jamanetwork.com/journals/jama/fullarticle/2588763", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 12, - 15 - ] - ], - "date-time": "2016-12-15T17:42:01Z", - "timestamp": 1481823721000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 12, - 13 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1001/jama.2016.17216", - "relation": {}, - "subject": [ - "General Medicine" - ], - "container-title-short": "JAMA", - "id": "1mJW6umJ" - }, - "citation_id": "1mJW6umJ" - }, - "doi:10.1002/9783527628766": { - "source": "doi", - "identifer": "10.1002/9783527628766", - "standard_citation": "doi:10.1002/9783527628766", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T07:42:15Z", - "timestamp": 1507102935645 - }, - "publisher-location": "Weinheim, Germany", - "reference-count": 0, - "publisher": "Wiley-VCH Verlag GmbH & Co. KGaA", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2009, - 7, - 15 - ] - ] - }, - "DOI": "10.1002/9783527628766", - "type": "book", - "created": { - "date-parts": [ - [ - 2010, - 2, - 12 - ] - ], - "date-time": "2010-02-12T09:46:41Z", - "timestamp": 1265968001000 - }, - "source": "Crossref", - "is-referenced-by-count": 285, - "title": "Molecular Descriptors for Chemoinformatics", - "prefix": "10.1002", - "member": "311", - "container-title": "Methods and Principles in Medicinal Chemistry", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2011, - 8, - 12 - ] - ], - "date-time": "2011-08-12T08:37:32Z", - "timestamp": 1313138252000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Roberto", - "family": "Todeschini", - "affiliation": [] - }, - { - "given": "Viviana", - "family": "Consonni", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2009, - 7, - 15 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1002/9783527628766", - "relation": {}, - "id": "17eGl2pn9" - }, - "citation_id": "17eGl2pn9" - }, - "doi:10.1002/cyto.a.22909": { - "source": "doi", - "identifer": "10.1002/cyto.a.22909", - "standard_citation": "doi:10.1002/cyto.a.22909", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 23 - ] - ], - "date-time": "2017-09-23T11:22:16Z", - "timestamp": 1506165736611 - }, - "reference-count": 119, - "publisher": "Wiley-Blackwell", - "issue": "2", - "license": [ - { - "URL": "http://doi.wiley.com/10.1002/tdm_license_1", - "start": { - "date-parts": [ - [ - 2016, - 7, - 19 - ] - ], - "date-time": "2016-07-19T00:00:00Z", - "timestamp": 1468886400000 - }, - "delay-in-days": 0, - "content-version": "tdm" - }, - { - "URL": "http://onlinelibrary.wiley.com/termsAndConditions", - "start": { - "date-parts": [ - [ - 2016, - 7, - 19 - ] - ], - "date-time": "2016-07-19T00:00:00Z", - "timestamp": 1468886400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/501100001441", - "name": "Bioinformatics Institute", - "doi-asserted-by": "crossref", - "award": [] - }, - { - "name": "Joint Council Office", - "award": [ - "1431AFG105" - ] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2017, - 2 - ] - ] - }, - "DOI": "10.1002/cyto.a.22909", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 7, - 19 - ] - ], - "date-time": "2016-07-19T15:17:52Z", - "timestamp": 1468941472000 - }, - "page": "115-125", - "source": "Crossref", - "is-referenced-by-count": 4, - "title": "Large-scale image-based screening and profiling of cellular phenotypes", - "prefix": "10.1002", - "volume": "91", - "author": [ - { - "given": "Nicola", - "family": "Bougen-Zhukov", - "affiliation": [ - { - "name": "Bioinformatics Institute, Agency for Science, Technology and Research; Singapore 138671 Singapore" - } - ] - }, - { - "given": "Sheng Yang", - "family": "Loh", - "affiliation": [ - { - "name": "Bioinformatics Institute, Agency for Science, Technology and Research; Singapore 138671 Singapore" - } - ] - }, - { - "given": "Hwee Kuan", - "family": "Lee", - "affiliation": [ - { - "name": "Bioinformatics Institute, Agency for Science, Technology and Research; Singapore 138671 Singapore" - } - ] - }, - { - "given": "Lit-Hsin", - "family": "Loo", - "affiliation": [ - { - "name": "Bioinformatics Institute, Agency for Science, Technology and Research; Singapore 138671 Singapore" - }, - { - "name": "Department of Pharmacology; School of Medicine, National University of Singapore; Singapore 117600 Singapore" - } - ] - } - ], - "member": "311", - "published-online": { - "date-parts": [ - [ - 2016, - 7, - 19 - ] - ] - }, - "container-title": "Cytometry Part A", - "original-title": [], - "link": [ - { - "URL": "https://api.wiley.com/onlinelibrary/tdm/v1/articles/10.1002%2Fcyto.a.22909", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "https://api.wiley.com/onlinelibrary/tdm/v1/articles/10.1002%2Fcyto.a.22909", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://onlinelibrary.wiley.com/wol1/doi/10.1002/cyto.a.22909/fullpdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T18:44:03Z", - "timestamp": 1498329843000 - }, - "score": 1.0, - "subtitle": [ - "Phenotypic Screening and Profiling" - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 7, - 19 - ] - ] - }, - "references-count": 119, - "URL": "https://doi.org/10.1002/cyto.a.22909", - "relation": { - "cites": [] - }, - "subject": [ - "Pathology and Forensic Medicine", - "Cell Biology", - "Histology" - ], - "container-title-short": "Cytometry", - "id": "m3Ij21U8" - }, - "citation_id": "m3Ij21U8" - }, - "doi:10.1002/jcc.24764": { - "source": "doi", - "identifer": "10.1002/jcc.24764", - "standard_citation": "doi:10.1002/jcc.24764", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 2 - ] - ], - "date-time": "2017-10-02T18:42:17Z", - "timestamp": 1506969737208 - }, - "reference-count": 150, - "publisher": "Wiley-Blackwell", - "issue": "16", - "license": [ - { - "URL": "http://doi.wiley.com/10.1002/tdm_license_1", - "start": { - "date-parts": [ - [ - 2017, - 3, - 8 - ] - ], - "date-time": "2017-03-08T00:00:00Z", - "timestamp": 1488931200000 - }, - "delay-in-days": 0, - "content-version": "tdm" - }, - { - "URL": "http://onlinelibrary.wiley.com/termsAndConditions", - "start": { - "date-parts": [ - [ - 2017, - 3, - 8 - ] - ], - "date-time": "2017-03-08T00:00:00Z", - "timestamp": 1488931200000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "name": "Pacific Northwest Laboratory Directed Research and Development (LDRD) Program and the Linus Pauling Distinguished Postdoctoral Fellowship", - "award": [] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2017, - 6, - 15 - ] - ] - }, - "DOI": "10.1002/jcc.24764", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2017, - 3, - 8 - ] - ], - "date-time": "2017-03-08T14:07:27Z", - "timestamp": 1488982047000 - }, - "page": "1291-1307", - "source": "Crossref", - "is-referenced-by-count": 5, - "title": "Deep learning for computational chemistry", - "prefix": "10.1002", - "volume": "38", - "author": [ - { - "ORCID": "http://orcid.org/0000-0002-8858-0868", - "authenticated-orcid": false, - "given": "Garrett B.", - "family": "Goh", - "affiliation": [ - { - "name": "Advanced Computing, Mathematics, and Data Division, Pacific Northwest National Laboratory; 902 Battelle Blvd Richland Washington 99354" - } - ] - }, - { - "given": "Nathan O.", - "family": "Hodas", - "affiliation": [ - { - "name": "Advanced Computing, Mathematics, and Data Division; Pacific Northwest National Laboratory; 902 Battelle Blvd Richland Washington 99354" - } - ] - }, - { - "given": "Abhinav", - "family": "Vishnu", - "affiliation": [ - { - "name": "Advanced Computing, Mathematics, and Data Division, Pacific Northwest National Laboratory; 902 Battelle Blvd Richland Washington 99354" - } - ] - } - ], - "member": "311", - "published-online": { - "date-parts": [ - [ - 2017, - 3, - 8 - ] - ] - }, - "container-title": "Journal of Computational Chemistry", - "original-title": [], - "link": [ - { - "URL": "https://api.wiley.com/onlinelibrary/tdm/v1/articles/10.1002%2Fjcc.24764", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://onlinelibrary.wiley.com/wol1/doi/10.1002/jcc.24764/fullpdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 4, - 27 - ] - ], - "date-time": "2017-04-27T01:30:55Z", - "timestamp": 1493256655000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 3, - 8 - ] - ] - }, - "references-count": 150, - "URL": "https://doi.org/10.1002/jcc.24764", - "relation": { - "cites": [] - }, - "subject": [ - "General Chemistry", - "Computational Mathematics" - ], - "container-title-short": "J. Comput. Chem.", - "id": "zCt6PUXj" - }, - "citation_id": "zCt6PUXj" - }, - "doi:10.1002/minf.201501008": { - "source": "doi", - "identifer": "10.1002/minf.201501008", - "standard_citation": "doi:10.1002/minf.201501008", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 22 - ] - ], - "date-time": "2017-09-22T06:22:12Z", - "timestamp": 1506061332525 - }, - "reference-count": 199, - "publisher": "Wiley-Blackwell", - "issue": "1", - "license": [ - { - "URL": "http://doi.wiley.com/10.1002/tdm_license_1.1", - "start": { - "date-parts": [ - [ - 2015, - 9, - 1 - ] - ], - "date-time": "2015-09-01T00:00:00Z", - "timestamp": 1441065600000 - }, - "delay-in-days": 0, - "content-version": "tdm" - } - ], - "funder": [ - { - "DOI": "10.13039/501100001711", - "name": "Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung", - "doi-asserted-by": "crossref", - "award": [ - "200021_157190", - "CR32I2_159737" - ] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 1 - ] - ] - }, - "DOI": "10.1002/minf.201501008", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 12, - 30 - ] - ], - "date-time": "2015-12-30T08:46:39Z", - "timestamp": 1451465199000 - }, - "page": "3-14", - "source": "Crossref", - "is-referenced-by-count": 41, - "title": "Deep Learning in Drug Discovery", - "prefix": "10.1002", - "volume": "35", - "author": [ - { - "given": "Erik", - "family": "Gawehn", - "affiliation": [] - }, - { - "given": "Jan A.", - "family": "Hiss", - "affiliation": [] - }, - { - "given": "Gisbert", - "family": "Schneider", - "affiliation": [] - } - ], - "member": "311", - "published-online": { - "date-parts": [ - [ - 2015, - 12, - 30 - ] - ] - }, - "container-title": "Molecular Informatics", - "original-title": [], - "link": [ - { - "URL": "https://api.wiley.com/onlinelibrary/tdm/v1/articles/10.1002%2Fminf.201501008", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T02:45:12Z", - "timestamp": 1498272312000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 12, - 30 - ] - ] - }, - "references-count": 199, - "URL": "https://doi.org/10.1002/minf.201501008", - "relation": { - "cites": [] - }, - "subject": [ - "Organic Chemistry", - "Molecular Medicine", - "Structural Biology", - "Drug Discovery", - "Computer Science Applications" - ], - "container-title-short": "Mol. Inf.", - "id": "gJE0ExFr" - }, - "citation_id": "gJE0ExFr" - }, - "doi:10.1002/wcms.1104": { - "source": "doi", - "identifer": "10.1002/wcms.1104", - "standard_citation": "doi:10.1002/wcms.1104", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 19 - ] - ], - "date-time": "2017-09-19T04:02:24Z", - "timestamp": 1505793744512 - }, - "reference-count": 167, - "publisher": "Wiley-Blackwell", - "issue": "5", - "license": [ - { - "URL": "http://doi.wiley.com/10.1002/tdm_license_1.1", - "start": { - "date-parts": [ - [ - 2015, - 9, - 1 - ] - ], - "date-time": "2015-09-01T00:00:00Z", - "timestamp": 1441065600000 - }, - "delay-in-days": 1231, - "content-version": "tdm" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2012, - 9 - ] - ] - }, - "DOI": "10.1002/wcms.1104", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2012, - 4, - 18 - ] - ], - "date-time": "2012-04-18T18:19:50Z", - "timestamp": 1334773190000 - }, - "page": "717-733", - "source": "Crossref", - "is-referenced-by-count": 26, - "title": "The enumeration of chemical space", - "prefix": "10.1002", - "volume": "2", - "author": [ - { - "given": "Jean-Louis", - "family": "Reymond", - "affiliation": [] - }, - { - "given": "Lars", - "family": "Ruddigkeit", - "affiliation": [] - }, - { - "given": "Lorenz", - "family": "Blum", - "affiliation": [] - }, - { - "given": "Ruud", - "family": "van Deursen", - "affiliation": [] - } - ], - "member": "311", - "published-online": { - "date-parts": [ - [ - 2012, - 4, - 18 - ] - ] - }, - "container-title": "Wiley Interdisciplinary Reviews: Computational Molecular Science", - "original-title": [], - "link": [ - { - "URL": "https://api.wiley.com/onlinelibrary/tdm/v1/articles/10.1002%2Fwcms.1104", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 20 - ] - ], - "date-time": "2017-06-20T17:27:14Z", - "timestamp": 1497979634000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2012, - 4, - 18 - ] - ] - }, - "references-count": 167, - "URL": "https://doi.org/10.1002/wcms.1104", - "relation": { - "cites": [] - }, - "subject": [ - "Physical and Theoretical Chemistry", - "Materials Chemistry", - "Biochemistry", - "Computational Mathematics", - "Computer Science Applications" - ], - "container-title-short": "WIREs Comput Mol Sci", - "id": "WeiyYhfy" - }, - "citation_id": "WeiyYhfy" - }, - "doi:10.1002/wcms.49": { - "source": "doi", - "identifer": "10.1002/wcms.49", - "standard_citation": "doi:10.1002/wcms.49", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 11 - ] - ], - "date-time": "2017-08-11T09:26:18Z", - "timestamp": 1502443578182 - }, - "reference-count": 132, - "publisher": "Wiley-Blackwell", - "issue": "5", - "license": [ - { - "URL": "http://doi.wiley.com/10.1002/tdm_license_1.1", - "start": { - "date-parts": [ - [ - 2015, - 9, - 1 - ] - ], - "date-time": "2015-09-01T00:00:00Z", - "timestamp": 1441065600000 - }, - "delay-in-days": 1590, - "content-version": "tdm" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2011, - 9 - ] - ] - }, - "DOI": "10.1002/wcms.49", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2011, - 6, - 28 - ] - ], - "date-time": "2011-06-28T12:49:17Z", - "timestamp": 1309265357000 - }, - "page": "742-759", - "source": "Crossref", - "is-referenced-by-count": 22, - "title": "Enabling future drug discovery byde novodesign", - "prefix": "10.1002", - "volume": "1", - "author": [ - { - "given": "Markus", - "family": "Hartenfeller", - "affiliation": [] - }, - { - "given": "Gisbert", - "family": "Schneider", - "affiliation": [] - } - ], - "member": "311", - "published-online": { - "date-parts": [ - [ - 2011, - 4, - 25 - ] - ] - }, - "container-title": "Wiley Interdisciplinary Reviews: Computational Molecular Science", - "original-title": [], - "link": [ - { - "URL": "https://api.wiley.com/onlinelibrary/tdm/v1/articles/10.1002%2Fwcms.49", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "https://api.wiley.com/onlinelibrary/tdm/v1/articles/10.1002%2Fwcms.49", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://onlinelibrary.wiley.com/wol1/doi/10.1002/wcms.49/fullpdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 20 - ] - ], - "date-time": "2017-06-20T02:27:59Z", - "timestamp": 1497925679000 - }, - "score": 1.0, - "subtitle": [ - "De novodrug design" - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2011, - 4, - 25 - ] - ] - }, - "references-count": 132, - "URL": "https://doi.org/10.1002/wcms.49", - "relation": { - "cites": [] - }, - "subject": [ - "Physical and Theoretical Chemistry", - "Materials Chemistry", - "Biochemistry", - "Computational Mathematics", - "Computer Science Applications" - ], - "container-title-short": "WIREs Comput Mol Sci", - "id": "kJ4hy7E" - }, - "citation_id": "kJ4hy7E" - }, - "doi:10.1006/jmbi.1999.3091": { - "source": "doi", - "identifer": "10.1006/jmbi.1999.3091", - "standard_citation": "doi:10.1006/jmbi.1999.3091", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 30 - ] - ], - "date-time": "2017-09-30T00:22:12Z", - "timestamp": 1506730932771 - }, - "reference-count": 30, - "publisher": "Elsevier BV", - "issue": "2", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 1999, - 9, - 1 - ] - ], - "date-time": "1999-09-01T00:00:00Z", - "timestamp": 936144000000 - }, - "delay-in-days": 0, - "content-version": "tdm" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 1999, - 9 - ] - ] - }, - "DOI": "10.1006/jmbi.1999.3091", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2002, - 10, - 7 - ] - ], - "date-time": "2002-10-07T16:21:31Z", - "timestamp": 1034007691000 - }, - "page": "195-202", - "source": "Crossref", - "is-referenced-by-count": 2460, - "title": "Protein secondary structure prediction based on position-specific scoring matrices 1 1Edited by G. Von Heijne", - "prefix": "10.1006", - "volume": "292", - "author": [ - { - "given": "David T", - "family": "Jones", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Journal of Molecular Biology", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S0022283699930917?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S0022283699930917?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 8 - ] - ], - "date-time": "2017-06-08T17:17:56Z", - "timestamp": 1496942276000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 1999, - 9 - ] - ] - }, - "references-count": 30, - "alternative-id": [ - "S0022283699930917" - ], - "URL": "https://doi.org/10.1006/jmbi.1999.3091", - "relation": {}, - "subject": [ - "Molecular Biology" - ], - "container-title-short": "Journal of Molecular Biology", - "id": "Aic7UyXM" - }, - "citation_id": "Aic7UyXM" - }, - "doi:10.1007/978-3-319-16706-0_20": { - "source": "doi", - "identifer": "10.1007/978-3-319-16706-0_20", - "standard_citation": "doi:10.1007/978-3-319-16706-0_20", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T00:46:15Z", - "timestamp": 1502325975840 - }, - "publisher-location": "Cham", - "reference-count": 0, - "publisher": "Springer International Publishing", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015 - ] - ] - }, - "DOI": "10.1007/978-3-319-16706-0_20", - "type": "chapter", - "created": { - "date-parts": [ - [ - 2015, - 3, - 25 - ] - ], - "date-time": "2015-03-25T08:54:01Z", - "timestamp": 1427273641000 - }, - "page": "205-217", - "source": "Crossref", - "is-referenced-by-count": 1, - "title": "Deep Feature Selection: Theory and Application to Identify Enhancers and Promoters", - "prefix": "10.1007", - "author": [ - { - "given": "Yifeng", - "family": "Li", - "affiliation": [] - }, - { - "given": "Chih-Yu", - "family": "Chen", - "affiliation": [] - }, - { - "given": "Wyeth W.", - "family": "Wasserman", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2015, - 3, - 26 - ] - ] - }, - "container-title": "Lecture Notes in Computer Science", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1007/978-3-319-16706-0_20", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2015, - 3, - 25 - ] - ], - "date-time": "2015-03-25T08:54:16Z", - "timestamp": 1427273656000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1007/978-3-319-16706-0_20", - "relation": {}, - "id": "17B2QAA1k" - }, - "citation_id": "17B2QAA1k" - }, - "doi:10.1007/978-3-319-24574-4_28": { - "source": "doi", - "identifer": "10.1007/978-3-319-24574-4_28", - "standard_citation": "doi:10.1007/978-3-319-24574-4_28", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T07:42:12Z", - "timestamp": 1507016532945 - }, - "publisher-location": "Cham", - "reference-count": 0, - "publisher": "Springer International Publishing", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2015, - 1, - 1 - ] - ], - "date-time": "2015-01-01T00:00:00Z", - "timestamp": 1420070400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015 - ] - ] - }, - "DOI": "10.1007/978-3-319-24574-4_28", - "type": "chapter", - "created": { - "date-parts": [ - [ - 2015, - 9, - 24 - ] - ], - "date-time": "2015-09-24T07:03:02Z", - "timestamp": 1443078182000 - }, - "page": "234-241", - "source": "Crossref", - "is-referenced-by-count": 150, - "title": "U-Net: Convolutional Networks for Biomedical Image Segmentation", - "prefix": "10.1007", - "author": [ - { - "given": "Olaf", - "family": "Ronneberger", - "affiliation": [] - }, - { - "given": "Philipp", - "family": "Fischer", - "affiliation": [] - }, - { - "given": "Thomas", - "family": "Brox", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2015, - 11, - 18 - ] - ] - }, - "container-title": "Lecture Notes in Computer Science", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1007/978-3-319-24574-4_28", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 3, - 9 - ] - ], - "date-time": "2016-03-09T23:04:13Z", - "timestamp": 1457564653000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1007/978-3-319-24574-4_28", - "relation": {}, - "id": "TutLhFSz" - }, - "citation_id": "TutLhFSz" - }, - "doi:10.1007/978-3-319-44839-8": { - "source": "doi", - "identifer": "10.1007/978-3-319-44839-8", - "standard_citation": "doi:10.1007/978-3-319-44839-8", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T05:36:16Z", - "timestamp": 1502343376633 - }, - "publisher-location": "Cham", - "reference-count": 0, - "publisher": "Springer International Publishing", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2016, - 1, - 1 - ] - ], - "date-time": "2016-01-01T00:00:00Z", - "timestamp": 1451606400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016 - ] - ] - }, - "DOI": "10.1007/978-3-319-44839-8", - "type": "book", - "created": { - "date-parts": [ - [ - 2016, - 12, - 19 - ] - ], - "date-time": "2016-12-19T10:24:55Z", - "timestamp": 1482143095000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Healthcare Interoperability Standards Compliance Handbook", - "prefix": "10.1007", - "author": [ - { - "given": "Frank", - "family": "Oemig", - "affiliation": [] - }, - { - "given": "Robert", - "family": "Snelick", - "affiliation": [] - } - ], - "member": "297", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1007/978-3-319-44839-8.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://link.springer.com/content/pdf/10.1007/978-3-319-44839-8", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 12, - 19 - ] - ], - "date-time": "2016-12-19T10:24:55Z", - "timestamp": 1482143095000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1007/978-3-319-44839-8", - "relation": {}, - "id": "1CWhXZxos" - }, - "citation_id": "1CWhXZxos" - }, - "doi:10.1007/978-3-319-46227-1_1": { - "source": "doi", - "identifer": "10.1007/978-3-319-46227-1_1", - "standard_citation": "doi:10.1007/978-3-319-46227-1_1", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 14 - ] - ], - "date-time": "2017-08-14T20:54:35Z", - "timestamp": 1502744075948 - }, - "publisher-location": "Cham", - "reference-count": 48, - "publisher": "Springer International Publishing", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2016, - 1, - 1 - ] - ], - "date-time": "2016-01-01T00:00:00Z", - "timestamp": 1451606400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016 - ] - ] - }, - "DOI": "10.1007/978-3-319-46227-1_1", - "type": "chapter", - "created": { - "date-parts": [ - [ - 2016, - 9, - 3 - ] - ], - "date-time": "2016-09-03T05:34:10Z", - "timestamp": 1472880850000 - }, - "page": "1-16", - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "AUC-Maximized Deep Convolutional Neural Fields for Protein Sequence Labeling", - "prefix": "10.1007", - "author": [ - { - "given": "Sheng", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Siqi", - "family": "Sun", - "affiliation": [] - }, - { - "given": "Jinbo", - "family": "Xu", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2016, - 9, - 4 - ] - ] - }, - "container-title": "Machine Learning and Knowledge Discovery in Databases", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1007/978-3-319-46227-1_1", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T21:58:01Z", - "timestamp": 1498341481000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016 - ] - ] - }, - "references-count": 48, - "URL": "https://doi.org/10.1007/978-3-319-46227-1_1", - "relation": { - "cites": [] - }, - "id": "pNoAbBEu" - }, - "citation_id": "pNoAbBEu" - }, - "doi:10.1007/978-3-642-40763-5_51": { - "source": "doi", - "identifer": "10.1007/978-3-642-40763-5_51", - "standard_citation": "doi:10.1007/978-3-642-40763-5_51", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T04:22:16Z", - "timestamp": 1507004536514 - }, - "publisher-location": "Berlin, Heidelberg", - "reference-count": 0, - "publisher": "Springer Berlin Heidelberg", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2013 - ] - ] - }, - "DOI": "10.1007/978-3-642-40763-5_51", - "type": "chapter", - "created": { - "date-parts": [ - [ - 2013, - 9, - 20 - ] - ], - "date-time": "2013-09-20T10:21:37Z", - "timestamp": 1379672497000 - }, - "page": "411-418", - "source": "Crossref", - "is-referenced-by-count": 115, - "title": "Mitosis Detection in Breast Cancer Histology Images with Deep Neural Networks", - "prefix": "10.1007", - "author": [ - { - "given": "Dan C.", - "family": "Cireşan", - "affiliation": [] - }, - { - "given": "Alessandro", - "family": "Giusti", - "affiliation": [] - }, - { - "given": "Luca M.", - "family": "Gambardella", - "affiliation": [] - }, - { - "given": "Jürgen", - "family": "Schmidhuber", - "affiliation": [] - } - ], - "member": "297", - "container-title": "Medical Image Computing and Computer-Assisted Intervention – MICCAI 2013", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1007/978-3-642-40763-5_51", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2013, - 9, - 20 - ] - ], - "date-time": "2013-09-20T10:21:50Z", - "timestamp": 1379672510000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2013 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1007/978-3-642-40763-5_51", - "relation": {}, - "id": "koEdZRcY" - }, - "citation_id": "koEdZRcY" - }, - "doi:10.1007/bf02478259": { - "source": "doi", - "identifer": "10.1007/bf02478259", - "standard_citation": "doi:10.1007/bf02478259", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T10:02:18Z", - "timestamp": 1507111338766 - }, - "reference-count": 3, - "publisher": "Springer Nature", - "issue": "4", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 1943, - 12 - ] - ] - }, - "DOI": "10.1007/bf02478259", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2006, - 9, - 18 - ] - ], - "date-time": "2006-09-18T11:10:54Z", - "timestamp": 1158577854000 - }, - "page": "115-133", - "source": "Crossref", - "is-referenced-by-count": 4052, - "title": "A logical calculus of the ideas immanent in nervous activity", - "prefix": "10.1007", - "volume": "5", - "author": [ - { - "given": "Warren S.", - "family": "McCulloch", - "affiliation": [] - }, - { - "given": "Walter", - "family": "Pitts", - "affiliation": [] - } - ], - "member": "297", - "container-title": "The Bulletin of Mathematical Biophysics", - "original-title": [], - "link": [ - { - "URL": "http://www.springerlink.com/index/pdf/10.1007/BF02478259", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2013, - 2, - 12 - ] - ], - "date-time": "2013-02-12T03:45:33Z", - "timestamp": 1360640733000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 1943, - 12 - ] - ] - }, - "references-count": 3, - "alternative-id": [ - "BF02478259" - ], - "URL": "https://doi.org/10.1007/bf02478259", - "relation": { - "cites": [] - }, - "subject": [ - "Immunology", - "General Biochemistry, Genetics and Molecular Biology", - "Computational Theory and Mathematics", - "General Neuroscience", - "Pharmacology", - "General Agricultural and Biological Sciences", - "General Mathematics", - "General Environmental Science", - "General Medicine" - ], - "container-title-short": "Bulletin of Mathematical Biophysics", - "id": "1HVDhhwpK" - }, - "citation_id": "1HVDhhwpK" - }, - "doi:10.1007/s10278-016-9914-9": { - "source": "doi", - "identifer": "10.1007/s10278-016-9914-9", - "standard_citation": "doi:10.1007/s10278-016-9914-9", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 21 - ] - ], - "date-time": "2017-08-21T12:22:18Z", - "timestamp": 1503318138231 - }, - "reference-count": 24, - "publisher": "Springer Nature", - "issue": "1", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0", - "start": { - "date-parts": [ - [ - 2016, - 10, - 11 - ] - ], - "date-time": "2016-10-11T00:00:00Z", - "timestamp": 1476144000000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "link.springer.com" - ], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2017, - 2 - ] - ] - }, - "DOI": "10.1007/s10278-016-9914-9", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 10, - 11 - ] - ], - "date-time": "2016-10-11T19:50:18Z", - "timestamp": 1476215418000 - }, - "page": "95-101", - "update-policy": "http://dx.doi.org/10.1007/springer_crossmark_policy", - "source": "Crossref", - "is-referenced-by-count": 6, - "title": "High-Throughput Classification of Radiographs Using Deep Convolutional Neural Networks", - "prefix": "10.1007", - "volume": "30", - "author": [ - { - "given": "Alvin", - "family": "Rajkomar", - "affiliation": [] - }, - { - "given": "Sneha", - "family": "Lingam", - "affiliation": [] - }, - { - "given": "Andrew G.", - "family": "Taylor", - "affiliation": [] - }, - { - "given": "Michael", - "family": "Blum", - "affiliation": [] - }, - { - "given": "John", - "family": "Mongan", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2016, - 10, - 11 - ] - ] - }, - "container-title": "Journal of Digital Imaging", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1007/s10278-016-9914-9.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://link.springer.com/article/10.1007/s10278-016-9914-9/fulltext.html", - "content-type": "text/html", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://link.springer.com/content/pdf/10.1007/s10278-016-9914-9.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T00:52:11Z", - "timestamp": 1498351931000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 10, - 11 - ] - ] - }, - "references-count": 24, - "alternative-id": [ - "9914" - ], - "URL": "https://doi.org/10.1007/s10278-016-9914-9", - "relation": { - "cites": [] - }, - "subject": [ - "Radiological and Ultrasound Technology", - "Radiology Nuclear Medicine and imaging", - "Computer Science Applications" - ], - "container-title-short": "J Digit Imaging", - "id": "x6HXFAS4" - }, - "citation_id": "x6HXFAS4" - }, - "doi:10.1016/s0097-8485(96)80015-5": { - "source": "doi", - "identifer": "10.1016/s0097-8485(96)80015-5", - "standard_citation": "doi:10.1016/s0097-8485(96)80015-5", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 8 - ] - ], - "date-time": "2017-08-08T02:06:55Z", - "timestamp": 1502158015617 - }, - "reference-count": 16, - "publisher": "Elsevier BV", - "issue": "1", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 1996, - 3, - 1 - ] - ], - "date-time": "1996-03-01T00:00:00Z", - "timestamp": 825638400000 - }, - "delay-in-days": 0, - "content-version": "tdm" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 1996, - 3 - ] - ] - }, - "DOI": "10.1016/s0097-8485(96)80015-5", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2002, - 10, - 10 - ] - ], - "date-time": "2002-10-10T21:58:24Z", - "timestamp": 1034287104000 - }, - "page": "135-140", - "source": "Crossref", - "is-referenced-by-count": 17, - "title": "Detection of RNA polymerase II promoters and polyadenylation sites in human DNA sequence", - "prefix": "10.1016", - "volume": "20", - "author": [ - { - "given": "Sherri", - "family": "Matis", - "affiliation": [] - }, - { - "given": "Ying", - "family": "Xu", - "affiliation": [] - }, - { - "given": "Manesh", - "family": "Shah", - "affiliation": [] - }, - { - "given": "Xiaojun", - "family": "Guan", - "affiliation": [] - }, - { - "given": "J.Ralph", - "family": "Einstein", - "affiliation": [] - }, - { - "given": "Richard", - "family": "Mural", - "affiliation": [] - }, - { - "given": "Edward", - "family": "Uberbacher", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Computers & Chemistry", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S0097848596800155?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S0097848596800155?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 15 - ] - ], - "date-time": "2017-06-15T03:51:34Z", - "timestamp": 1497498694000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 1996, - 3 - ] - ] - }, - "references-count": 16, - "alternative-id": [ - "S0097848596800155" - ], - "URL": "https://doi.org/10.1016/s0097-8485(96)80015-5", - "relation": {}, - "subject": [ - "Biotechnology", - "Applied Microbiology and Biotechnology", - "General Chemical Engineering" - ], - "container-title-short": "Computers & Chemistry", - "id": "3Ew5V1iC" - }, - "citation_id": "3Ew5V1iC" - }, - "doi:10.1016/s0140-6736(13)61752-3": { - "source": "doi", - "identifer": "10.1016/s0140-6736(13)61752-3", - "standard_citation": "doi:10.1016/s0140-6736(13)61752-3", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T09:42:13Z", - "timestamp": 1507110133737 - }, - "reference-count": 83, - "publisher": "Elsevier BV", - "issue": "9921", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2014, - 3, - 1 - ] - ], - "date-time": "2014-03-01T00:00:00Z", - "timestamp": 1393632000000 - }, - "delay-in-days": 0, - "content-version": "tdm" - } - ], - "funder": [ - { - "name": "NIH", - "award": [ - "N01-HC-25195" - ] - } - ], - "content-domain": { - "domain": [ - "thelancet.com", - "em-consulte.com", - "elsevier.com", - "sciencedirect.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2014, - 3 - ] - ] - }, - "DOI": "10.1016/s0140-6736(13)61752-3", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2013, - 9, - 29 - ] - ], - "date-time": "2013-09-29T01:52:08Z", - "timestamp": 1380419528000 - }, - "page": "999-1008", - "update-policy": "http://dx.doi.org/10.1016/elsevier_cm_policy", - "source": "Crossref", - "is-referenced-by-count": 100, - "title": "The Framingham Heart Study and the epidemiology of cardiovascular disease: a historical perspective", - "prefix": "10.1016", - "volume": "383", - "author": [ - { - "given": "Syed S", - "family": "Mahmood", - "affiliation": [] - }, - { - "given": "Daniel", - "family": "Levy", - "affiliation": [] - }, - { - "given": "Ramachandran S", - "family": "Vasan", - "affiliation": [] - }, - { - "given": "Thomas J", - "family": "Wang", - "affiliation": [] - } - ], - "member": "78", - "container-title": "The Lancet", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S0140673613617523?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S0140673613617523?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 21 - ] - ], - "date-time": "2017-06-21T22:46:55Z", - "timestamp": 1498085215000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 3 - ] - ] - }, - "references-count": 83, - "alternative-id": [ - "S0140673613617523" - ], - "URL": "https://doi.org/10.1016/s0140-6736(13)61752-3", - "relation": {}, - "subject": [ - "General Medicine" - ], - "container-title-short": "The Lancet", - "assertion": [ - { - "value": "Elsevier", - "name": "publisher", - "label": "This article is maintained by" - }, - { - "value": "The Framingham Heart Study and the epidemiology of cardiovascular disease: a historical perspective", - "name": "articletitle", - "label": "Article Title" - }, - { - "value": "The Lancet", - "name": "journaltitle", - "label": "Journal Title" - }, - { - "value": "http://dx.doi.org/10.1016/S0140-6736(13)61752-3", - "name": "articlelink", - "label": "CrossRef DOI link to publisher maintained version" - }, - { - "value": "article", - "name": "content_type", - "label": "Content Type" - }, - { - "value": "Copyright © 2014 Elsevier Ltd. All rights reserved.", - "name": "copyright", - "label": "Copyright" - } - ], - "id": "N96QKgly" - }, - "citation_id": "N96QKgly" - }, - "doi:10.1016/s0167-9473(99)00098-5": { - "source": "doi", - "identifer": "10.1016/s0167-9473(99)00098-5", - "standard_citation": "doi:10.1016/s0167-9473(99)00098-5", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 13 - ] - ], - "date-time": "2017-09-13T04:42:13Z", - "timestamp": 1505277733366 - }, - "reference-count": 11, - "publisher": "Elsevier BV", - "issue": "2", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2000, - 8, - 1 - ] - ], - "date-time": "2000-08-01T00:00:00Z", - "timestamp": 965088000000 - }, - "delay-in-days": 0, - "content-version": "tdm" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2000, - 8 - ] - ] - }, - "DOI": "10.1016/s0167-9473(99)00098-5", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2002, - 7, - 25 - ] - ], - "date-time": "2002-07-25T14:10:41Z", - "timestamp": 1027606241000 - }, - "page": "243-257", - "source": "Crossref", - "is-referenced-by-count": 20, - "title": "Comparison of the performance of neural network methods and Cox regression for censored survival data", - "prefix": "10.1016", - "volume": "34", - "author": [ - { - "given": "Anny", - "family": "Xiang", - "affiliation": [] - }, - { - "given": "Pablo", - "family": "Lapuerta", - "affiliation": [] - }, - { - "given": "Alex", - "family": "Ryutov", - "affiliation": [] - }, - { - "given": "Jonathan", - "family": "Buckley", - "affiliation": [] - }, - { - "given": "Stanley", - "family": "Azen", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Computational Statistics & Data Analysis", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S0167947399000985?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S0167947399000985?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 14 - ] - ], - "date-time": "2017-06-14T21:15:44Z", - "timestamp": 1497474944000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2000, - 8 - ] - ] - }, - "references-count": 11, - "alternative-id": [ - "S0167947399000985" - ], - "URL": "https://doi.org/10.1016/s0167-9473(99)00098-5", - "relation": {}, - "subject": [ - "Statistics and Probability", - "Computational Theory and Mathematics", - "Applied Mathematics", - "Computational Mathematics" - ], - "container-title-short": "Computational Statistics & Data Analysis", - "id": "1921Mctzh" - }, - "citation_id": "1921Mctzh" - }, - "doi:10.1016/s0168-8510(02)00208-7": { - "source": "doi", - "identifer": "10.1016/s0168-8510(02)00208-7", - "standard_citation": "doi:10.1016/s0168-8510(02)00208-7", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T02:02:43Z", - "timestamp": 1502244163869 - }, - "reference-count": 14, - "publisher": "Elsevier BV", - "issue": "2", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2003, - 8, - 1 - ] - ], - "date-time": "2003-08-01T00:00:00Z", - "timestamp": 1059696000000 - }, - "delay-in-days": 0, - "content-version": "tdm" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2003, - 8 - ] - ] - }, - "DOI": "10.1016/s0168-8510(02)00208-7", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2003, - 2, - 12 - ] - ], - "date-time": "2003-02-12T03:07:18Z", - "timestamp": 1045019238000 - }, - "page": "101-107", - "source": "Crossref", - "is-referenced-by-count": 21, - "title": "Have DRG-based prospective payment systems influenced the number of secondary diagnoses in health care administrative data?", - "prefix": "10.1016", - "volume": "65", - "author": [ - { - "given": "Lisbeth", - "family": "Serdén", - "affiliation": [] - }, - { - "given": "Rikard", - "family": "Lindqvist", - "affiliation": [] - }, - { - "given": "Måns", - "family": "Rosén", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Health Policy", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S0168851002002087?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S0168851002002087?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 15 - ] - ], - "date-time": "2017-06-15T18:22:18Z", - "timestamp": 1497550938000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2003, - 8 - ] - ] - }, - "references-count": 14, - "alternative-id": [ - "S0168851002002087" - ], - "URL": "https://doi.org/10.1016/s0168-8510(02)00208-7", - "relation": {}, - "subject": [ - "Health Policy" - ], - "container-title-short": "Health Policy", - "id": "7BctyA7f" - }, - "citation_id": "7BctyA7f" - }, - "doi:10.1016/j.bjid.2015.08.013": { - "source": "doi", - "identifer": "10.1016/j.bjid.2015.08.013", - "standard_citation": "doi:10.1016/j.bjid.2015.08.013", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 11 - ] - ], - "date-time": "2017-08-11T01:41:33Z", - "timestamp": 1502415693026 - }, - "reference-count": 49, - "publisher": "Elsevier BV", - "issue": "6", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2015, - 11, - 1 - ] - ], - "date-time": "2015-11-01T00:00:00Z", - "timestamp": 1446336000000 - }, - "delay-in-days": 0, - "content-version": "tdm" - }, - { - "URL": "http://creativecommons.org/licenses/by-nc-nd/4.0/", - "start": { - "date-parts": [ - [ - 2015, - 10, - 9 - ] - ], - "date-time": "2015-10-09T00:00:00Z", - "timestamp": 1444348800000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/501100006435", - "name": "Central European Institute of Technology", - "doi-asserted-by": "publisher", - "award": [ - "CZ.1.05/1.1.00/02.0068" - ] - } - ], - "content-domain": { - "domain": [ - "elsevier.com", - "sciencedirect.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2015, - 11 - ] - ] - }, - "DOI": "10.1016/j.bjid.2015.08.013", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 10, - 27 - ] - ], - "date-time": "2015-10-27T16:07:37Z", - "timestamp": 1445962057000 - }, - "page": "604-613", - "update-policy": "http://dx.doi.org/10.1016/elsevier_cm_policy", - "source": "Crossref", - "is-referenced-by-count": 2, - "title": "Influence of microbiome species in hard-to-heal wounds on disease severity and treatment duration", - "prefix": "10.1016", - "volume": "19", - "author": [ - { - "given": "Dagmar", - "family": "Chudobova", - "affiliation": [] - }, - { - "given": "Kristyna", - "family": "Cihalova", - "affiliation": [] - }, - { - "given": "Roman", - "family": "Guran", - "affiliation": [] - }, - { - "given": "Simona", - "family": "Dostalova", - "affiliation": [] - }, - { - "given": "Kristyna", - "family": "Smerkova", - "affiliation": [] - }, - { - "given": "Radek", - "family": "Vesely", - "affiliation": [] - }, - { - "given": "Jaromir", - "family": "Gumulec", - "affiliation": [] - }, - { - "given": "Michal", - "family": "Masarik", - "affiliation": [] - }, - { - "given": "Zbynek", - "family": "Heger", - "affiliation": [] - }, - { - "given": "Vojtech", - "family": "Adam", - "affiliation": [] - }, - { - "given": "Rene", - "family": "Kizek", - "affiliation": [] - } - ], - "member": "78", - "container-title": "The Brazilian Journal of Infectious Diseases", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S1413867015001889?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S1413867015001889?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 23 - ] - ], - "date-time": "2017-06-23T22:20:24Z", - "timestamp": 1498256424000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 11 - ] - ] - }, - "references-count": 49, - "alternative-id": [ - "S1413867015001889" - ], - "URL": "https://doi.org/10.1016/j.bjid.2015.08.013", - "relation": {}, - "subject": [ - "Microbiology (medical)", - "Infectious Diseases" - ], - "container-title-short": "The Brazilian Journal of Infectious Diseases", - "assertion": [ - { - "value": "Elsevier", - "name": "publisher", - "label": "This article is maintained by" - }, - { - "value": "Influence of microbiome species in hard-to-heal wounds on disease severity and treatment duration", - "name": "articletitle", - "label": "Article Title" - }, - { - "value": "The Brazilian Journal of Infectious Diseases", - "name": "journaltitle", - "label": "Journal Title" - }, - { - "value": "http://dx.doi.org/10.1016/j.bjid.2015.08.013", - "name": "articlelink", - "label": "CrossRef DOI link to publisher maintained version" - }, - { - "value": "article", - "name": "content_type", - "label": "Content Type" - }, - { - "value": "Copyright © 2015 Elsevier Editora Ltda.", - "name": "copyright", - "label": "Copyright" - } - ], - "id": "oas5tbC7" - }, - "citation_id": "oas5tbC7" - }, - "doi:10.1016/j.cels.2016.04.013": { - "source": "doi", - "identifer": "10.1016/j.cels.2016.04.013", - "standard_citation": "doi:10.1016/j.cels.2016.04.013", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T15:44:39Z", - "timestamp": 1502293479441 - }, - "reference-count": 48, - "publisher": "Elsevier BV", - "issue": "1", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2016, - 7, - 1 - ] - ], - "date-time": "2016-07-01T00:00:00Z", - "timestamp": 1467331200000 - }, - "delay-in-days": 0, - "content-version": "tdm" - }, - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2016, - 4, - 29 - ] - ], - "date-time": "2016-04-29T00:00:00Z", - "timestamp": 1461888000000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "cell.com", - "elsevier.com", - "sciencedirect.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2016, - 7 - ] - ] - }, - "DOI": "10.1016/j.cels.2016.04.013", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 7, - 21 - ] - ], - "date-time": "2016-07-21T23:07:08Z", - "timestamp": 1469142428000 - }, - "page": "54-61", - "update-policy": "http://dx.doi.org/10.1016/elsevier_cm_policy", - "source": "Crossref", - "is-referenced-by-count": 4, - "title": "Enabling Privacy-Preserving GWASs in Heterogeneous Human Populations", - "prefix": "10.1016", - "volume": "3", - "author": [ - { - "given": "Sean", - "family": "Simmons", - "affiliation": [] - }, - { - "given": "Cenk", - "family": "Sahinalp", - "affiliation": [] - }, - { - "given": "Bonnie", - "family": "Berger", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Cell Systems", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S2405471216301211?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S2405471216301211?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T18:57:49Z", - "timestamp": 1498330669000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 7 - ] - ] - }, - "references-count": 48, - "alternative-id": [ - "S2405471216301211" - ], - "URL": "https://doi.org/10.1016/j.cels.2016.04.013", - "relation": {}, - "container-title-short": "Cell Systems", - "assertion": [ - { - "value": "Elsevier", - "name": "publisher", - "label": "This article is maintained by" - }, - { - "value": "Enabling Privacy-Preserving GWASs in Heterogeneous Human Populations", - "name": "articletitle", - "label": "Article Title" - }, - { - "value": "Cell Systems", - "name": "journaltitle", - "label": "Journal Title" - }, - { - "value": "http://dx.doi.org/10.1016/j.cels.2016.04.013", - "name": "articlelink", - "label": "CrossRef DOI link to publisher maintained version" - }, - { - "value": "article", - "name": "content_type", - "label": "Content Type" - }, - { - "value": "© 2016 The Authors. Published by Elsevier Inc.", - "name": "copyright", - "label": "Copyright" - } - ], - "id": "6XtEfQMC" - }, - "citation_id": "6XtEfQMC" - }, - "doi:10.1016/j.copbio.2016.04.003": { - "source": "doi", - "identifer": "10.1016/j.copbio.2016.04.003", - "standard_citation": "doi:10.1016/j.copbio.2016.04.003", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 25 - ] - ], - "date-time": "2017-09-25T14:42:26Z", - "timestamp": 1506350546593 - }, - "reference-count": 104, - "publisher": "Elsevier BV", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2016, - 6, - 1 - ] - ], - "date-time": "2016-06-01T00:00:00Z", - "timestamp": 1464739200000 - }, - "delay-in-days": 0, - "content-version": "tdm" - }, - { - "URL": "http://creativecommons.org/licenses/by-nc-nd/4.0/", - "start": { - "date-parts": [ - [ - 2016, - 4, - 5 - ] - ], - "date-time": "2016-04-05T00:00:00Z", - "timestamp": 1459814400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/100000001", - "name": "National Science Foundation", - "doi-asserted-by": "publisher", - "award": [] - } - ], - "content-domain": { - "domain": [ - "clinicalkey.jp", - "clinicalkey.com", - "clinicalkey.es", - "clinicalkey.com.au", - "clinicalkey.fr", - "bmn.com", - "elsevier.com", - "sciencedirect.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2016, - 6 - ] - ] - }, - "DOI": "10.1016/j.copbio.2016.04.003", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 4, - 19 - ] - ], - "date-time": "2016-04-19T06:45:14Z", - "timestamp": 1461048314000 - }, - "page": "134-142", - "update-policy": "http://dx.doi.org/10.1016/elsevier_cm_policy", - "source": "Crossref", - "is-referenced-by-count": 14, - "title": "Applications in image-based profiling of perturbations", - "prefix": "10.1016", - "volume": "39", - "author": [ - { - "given": "Juan C", - "family": "Caicedo", - "affiliation": [] - }, - { - "given": "Shantanu", - "family": "Singh", - "affiliation": [] - }, - { - "given": "Anne E", - "family": "Carpenter", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Current Opinion in Biotechnology", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S0958166916301112?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S0958166916301112?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T11:15:39Z", - "timestamp": 1498302939000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 6 - ] - ] - }, - "references-count": 104, - "alternative-id": [ - "S0958166916301112" - ], - "URL": "https://doi.org/10.1016/j.copbio.2016.04.003", - "relation": {}, - "subject": [ - "Biotechnology", - "Bioengineering", - "Biomedical Engineering" - ], - "container-title-short": "Current Opinion in Biotechnology", - "assertion": [ - { - "value": "Elsevier", - "name": "publisher", - "label": "This article is maintained by" - }, - { - "value": "Applications in image-based profiling of perturbations", - "name": "articletitle", - "label": "Article Title" - }, - { - "value": "Current Opinion in Biotechnology", - "name": "journaltitle", - "label": "Journal Title" - }, - { - "value": "http://dx.doi.org/10.1016/j.copbio.2016.04.003", - "name": "articlelink", - "label": "CrossRef DOI link to publisher maintained version" - }, - { - "value": "article", - "name": "content_type", - "label": "Content Type" - }, - { - "value": "© 2016 The Authors. Published by Elsevier Ltd.", - "name": "copyright", - "label": "Copyright" - } - ], - "id": "hkKO4QYl" - }, - "citation_id": "hkKO4QYl" - }, - "doi:10.1016/j.drudis.2006.06.016": { - "source": "doi", - "identifer": "10.1016/j.drudis.2006.06.016", - "standard_citation": "doi:10.1016/j.drudis.2006.06.016", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 30 - ] - ], - "date-time": "2017-09-30T20:42:14Z", - "timestamp": 1506804134761 - }, - "reference-count": 80, - "publisher": "Elsevier BV", - "issue": "15-16", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2006, - 8, - 1 - ] - ], - "date-time": "2006-08-01T00:00:00Z", - "timestamp": 1154390400000 - }, - "delay-in-days": 0, - "content-version": "tdm" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2006, - 8 - ] - ] - }, - "DOI": "10.1016/j.drudis.2006.06.016", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2006, - 7, - 18 - ] - ], - "date-time": "2006-07-18T11:13:15Z", - "timestamp": 1153221195000 - }, - "page": "741-748", - "source": "Crossref", - "is-referenced-by-count": 96, - "title": "Hit discovery and hit-to-lead approaches", - "prefix": "10.1016", - "volume": "11", - "author": [ - { - "given": "György M.", - "family": "Keserű", - "affiliation": [] - }, - { - "given": "Gergely M.", - "family": "Makara", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Drug Discovery Today", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S1359644606002285?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S1359644606002285?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 17 - ] - ], - "date-time": "2017-06-17T07:34:25Z", - "timestamp": 1497684865000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2006, - 8 - ] - ] - }, - "references-count": 80, - "alternative-id": [ - "S1359644606002285" - ], - "URL": "https://doi.org/10.1016/j.drudis.2006.06.016", - "relation": {}, - "subject": [ - "Pharmacology", - "Drug Discovery" - ], - "container-title-short": "Drug Discovery Today", - "id": "1D6emOV6q" - }, - "citation_id": "1D6emOV6q" - }, - "doi:10.1016/j.ijmedinf.2016.09.014": { - "source": "doi", - "identifer": "10.1016/j.ijmedinf.2016.09.014", - "standard_citation": "doi:10.1016/j.ijmedinf.2016.09.014", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 19 - ] - ], - "date-time": "2017-09-19T19:22:13Z", - "timestamp": 1505848933378 - }, - "reference-count": 37, - "publisher": "Elsevier BV", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2017, - 1, - 1 - ] - ], - "date-time": "2017-01-01T00:00:00Z", - "timestamp": 1483228800000 - }, - "delay-in-days": 0, - "content-version": "tdm" - } - ], - "content-domain": { - "domain": [ - "clinicalkey.jp", - "clinicalkey.com", - "ijmijournal.com", - "clinicalkey.es", - "clinicalkey.com.au", - "clinicalkey.fr", - "elsevier.com", - "sciencedirect.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2017, - 1 - ] - ] - }, - "DOI": "10.1016/j.ijmedinf.2016.09.014", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 10, - 7 - ] - ], - "date-time": "2016-10-07T00:23:21Z", - "timestamp": 1475799801000 - }, - "page": "120-127", - "update-policy": "http://dx.doi.org/10.1016/elsevier_cm_policy", - "source": "Crossref", - "is-referenced-by-count": 7, - "title": "A machine learning-based framework to identify type 2 diabetes through electronic health records", - "prefix": "10.1016", - "volume": "97", - "author": [ - { - "given": "Tao", - "family": "Zheng", - "affiliation": [] - }, - { - "given": "Wei", - "family": "Xie", - "affiliation": [] - }, - { - "given": "Liling", - "family": "Xu", - "affiliation": [] - }, - { - "given": "Xiaoying", - "family": "He", - "affiliation": [] - }, - { - "given": "Ya", - "family": "Zhang", - "affiliation": [] - }, - { - "given": "Mingrong", - "family": "You", - "affiliation": [] - }, - { - "given": "Gong", - "family": "Yang", - "affiliation": [] - }, - { - "given": "You", - "family": "Chen", - "affiliation": [] - } - ], - "member": "78", - "container-title": "International Journal of Medical Informatics", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S1386505616302155?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S1386505616302155?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T00:32:57Z", - "timestamp": 1498350777000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 1 - ] - ] - }, - "references-count": 37, - "alternative-id": [ - "S1386505616302155" - ], - "URL": "https://doi.org/10.1016/j.ijmedinf.2016.09.014", - "relation": { - "has-preprint": [ - { - "id-type": "doi", - "id": "10.1101/078634", - "asserted-by": "object" - } - ] - }, - "subject": [ - "Health Informatics" - ], - "container-title-short": "International Journal of Medical Informatics", - "assertion": [ - { - "value": "Elsevier", - "name": "publisher", - "label": "This article is maintained by" - }, - { - "value": "A machine learning-based framework to identify type 2 diabetes through electronic health records", - "name": "articletitle", - "label": "Article Title" - }, - { - "value": "International Journal of Medical Informatics", - "name": "journaltitle", - "label": "Journal Title" - }, - { - "value": "http://dx.doi.org/10.1016/j.ijmedinf.2016.09.014", - "name": "articlelink", - "label": "CrossRef DOI link to publisher maintained version" - }, - { - "value": "article", - "name": "content_type", - "label": "Content Type" - }, - { - "value": "© 2016 Elsevier Ireland Ltd. All rights reserved.", - "name": "copyright", - "label": "Copyright" - } - ], - "id": "1Ar4f4vfR" - }, - "citation_id": "1Ar4f4vfR" - }, - "doi:10.1016/j.jbi.2014.03.016": { - "source": "doi", - "identifer": "10.1016/j.jbi.2014.03.016", - "standard_citation": "doi:10.1016/j.jbi.2014.03.016", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 11 - ] - ], - "date-time": "2017-08-11T00:25:04Z", - "timestamp": 1502411104777 - }, - "reference-count": 36, - "publisher": "Elsevier BV", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2014, - 10, - 1 - ] - ], - "date-time": "2014-10-01T00:00:00Z", - "timestamp": 1412121600000 - }, - "delay-in-days": 0, - "content-version": "tdm" - }, - { - "URL": "http://www.elsevier.com/open-access/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2016, - 3, - 23 - ] - ], - "date-time": "2016-03-23T00:00:00Z", - "timestamp": 1458691200000 - }, - "delay-in-days": 539, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "j-biomed-inform.com", - "elsevier.com", - "sciencedirect.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2014, - 10 - ] - ] - }, - "DOI": "10.1016/j.jbi.2014.03.016", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 4, - 13 - ] - ], - "date-time": "2014-04-13T08:45:11Z", - "timestamp": 1397378711000 - }, - "page": "24-34", - "update-policy": "http://dx.doi.org/10.1016/elsevier_cm_policy", - "source": "Crossref", - "is-referenced-by-count": 13, - "title": "Identifying and mitigating biases in EHR laboratory tests", - "prefix": "10.1016", - "volume": "51", - "author": [ - { - "given": "Rimma", - "family": "Pivovarov", - "affiliation": [] - }, - { - "given": "David J.", - "family": "Albers", - "affiliation": [] - }, - { - "given": "Jorge L.", - "family": "Sepulveda", - "affiliation": [] - }, - { - "given": "Noémie", - "family": "Elhadad", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Journal of Biomedical Informatics", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S1532046414000847?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S1532046414000847?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 22 - ] - ], - "date-time": "2017-06-22T11:29:43Z", - "timestamp": 1498130983000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 10 - ] - ] - }, - "references-count": 36, - "alternative-id": [ - "S1532046414000847" - ], - "URL": "https://doi.org/10.1016/j.jbi.2014.03.016", - "relation": {}, - "subject": [ - "Health Informatics", - "Computer Science Applications" - ], - "container-title-short": "Journal of Biomedical Informatics", - "assertion": [ - { - "value": "Elsevier", - "name": "publisher", - "label": "This article is maintained by" - }, - { - "value": "Identifying and mitigating biases in EHR laboratory tests", - "name": "articletitle", - "label": "Article Title" - }, - { - "value": "Journal of Biomedical Informatics", - "name": "journaltitle", - "label": "Journal Title" - }, - { - "value": "http://dx.doi.org/10.1016/j.jbi.2014.03.016", - "name": "articlelink", - "label": "CrossRef DOI link to publisher maintained version" - }, - { - "value": "article", - "name": "content_type", - "label": "Content Type" - }, - { - "value": "Copyright © 2014 Elsevier Inc. All rights reserved.", - "name": "copyright", - "label": "Copyright" - } - ], - "id": "1C97CTU9S" - }, - "citation_id": "1C97CTU9S" - }, - "doi:10.1016/j.jbi.2014.10.006": { - "source": "doi", - "identifer": "10.1016/j.jbi.2014.10.006", - "standard_citation": "doi:10.1016/j.jbi.2014.10.006", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 25 - ] - ], - "date-time": "2017-09-25T12:22:09Z", - "timestamp": 1506342129160 - }, - "reference-count": 43, - "publisher": "Elsevier BV", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2015, - 2, - 1 - ] - ], - "date-time": "2015-02-01T00:00:00Z", - "timestamp": 1422748800000 - }, - "delay-in-days": 0, - "content-version": "tdm" - }, - { - "URL": "http://www.elsevier.com/open-access/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2016, - 2, - 1 - ] - ], - "date-time": "2016-02-01T00:00:00Z", - "timestamp": 1454284800000 - }, - "delay-in-days": 365, - "content-version": "vor" - } - ], - "funder": [ - { - "name": "Innovative Medicines Initiative Joint Undertaking", - "award": [ - "115189" - ] - }, - { - "name": "European Union’s Seventh Framework Programme", - "award": [ - "FP7/2007–2013" - ] - }, - { - "name": "EFPIA", - "award": [] - } - ], - "content-domain": { - "domain": [ - "j-biomed-inform.com", - "elsevier.com", - "sciencedirect.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2015, - 2 - ] - ] - }, - "DOI": "10.1016/j.jbi.2014.10.006", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 10, - 18 - ] - ], - "date-time": "2014-10-18T09:38:16Z", - "timestamp": 1413625096000 - }, - "page": "162-173", - "update-policy": "http://dx.doi.org/10.1016/elsevier_cm_policy", - "source": "Crossref", - "is-referenced-by-count": 29, - "title": "Using electronic health records for clinical research: The case of the EHR4CR project", - "prefix": "10.1016", - "volume": "53", - "author": [ - { - "given": "Georges", - "family": "De Moor", - "affiliation": [] - }, - { - "given": "Mats", - "family": "Sundgren", - "affiliation": [] - }, - { - "given": "Dipak", - "family": "Kalra", - "affiliation": [] - }, - { - "given": "Andreas", - "family": "Schmidt", - "affiliation": [] - }, - { - "given": "Martin", - "family": "Dugas", - "affiliation": [] - }, - { - "given": "Brecht", - "family": "Claerhout", - "affiliation": [] - }, - { - "given": "Töresin", - "family": "Karakoyun", - "affiliation": [] - }, - { - "given": "Christian", - "family": "Ohmann", - "affiliation": [] - }, - { - "given": "Pierre-Yves", - "family": "Lastic", - "affiliation": [] - }, - { - "given": "Nadir", - "family": "Ammour", - "affiliation": [] - }, - { - "given": "Rebecca", - "family": "Kush", - "affiliation": [] - }, - { - "given": "Danielle", - "family": "Dupont", - "affiliation": [] - }, - { - "given": "Marc", - "family": "Cuggia", - "affiliation": [] - }, - { - "given": "Christel", - "family": "Daniel", - "affiliation": [] - }, - { - "given": "Geert", - "family": "Thienpont", - "affiliation": [] - }, - { - "given": "Pascal", - "family": "Coorevits", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Journal of Biomedical Informatics", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S1532046414002263?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S1532046414002263?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 22 - ] - ], - "date-time": "2017-06-22T22:38:08Z", - "timestamp": 1498171088000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 2 - ] - ] - }, - "references-count": 43, - "alternative-id": [ - "S1532046414002263" - ], - "URL": "https://doi.org/10.1016/j.jbi.2014.10.006", - "relation": {}, - "subject": [ - "Health Informatics", - "Computer Science Applications" - ], - "container-title-short": "Journal of Biomedical Informatics", - "assertion": [ - { - "value": "Elsevier", - "name": "publisher", - "label": "This article is maintained by" - }, - { - "value": "Using electronic health records for clinical research: The case of the EHR4CR project", - "name": "articletitle", - "label": "Article Title" - }, - { - "value": "Journal of Biomedical Informatics", - "name": "journaltitle", - "label": "Journal Title" - }, - { - "value": "http://dx.doi.org/10.1016/j.jbi.2014.10.006", - "name": "articlelink", - "label": "CrossRef DOI link to publisher maintained version" - }, - { - "value": "article", - "name": "content_type", - "label": "Content Type" - }, - { - "value": "Copyright © 2014 Elsevier Inc.", - "name": "copyright", - "label": "Copyright" - } - ], - "id": "13filvWwr" - }, - "citation_id": "13filvWwr" - }, - "doi:10.1016/j.jbi.2016.10.007": { - "source": "doi", - "identifer": "10.1016/j.jbi.2016.10.007", - "standard_citation": "doi:10.1016/j.jbi.2016.10.007", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 14 - ] - ], - "date-time": "2017-09-14T22:42:07Z", - "timestamp": 1505428927267 - }, - "reference-count": 46, - "publisher": "Elsevier BV", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2016, - 12, - 1 - ] - ], - "date-time": "2016-12-01T00:00:00Z", - "timestamp": 1480550400000 - }, - "delay-in-days": 0, - "content-version": "tdm" - }, - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2016, - 10, - 11 - ] - ], - "date-time": "2016-10-11T00:00:00Z", - "timestamp": 1476144000000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/100004897", - "name": "Pennsylvania Department of Health", - "doi-asserted-by": "publisher", - "award": [] - }, - { - "name": "Commonwealth Universal Research Enhancement", - "award": [] - }, - { - "name": "Gordon and Betty Moore Foundation’s Data-Driven Discovery Initiative", - "award": [ - "GBMF4552" - ] - } - ], - "content-domain": { - "domain": [ - "elsevier.com", - "sciencedirect.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2016, - 12 - ] - ] - }, - "DOI": "10.1016/j.jbi.2016.10.007", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 10, - 16 - ] - ], - "date-time": "2016-10-16T05:07:08Z", - "timestamp": 1476594428000 - }, - "page": "168-178", - "update-policy": "http://dx.doi.org/10.1016/elsevier_cm_policy", - "source": "Crossref", - "is-referenced-by-count": 3, - "title": "Semi-supervised learning of the electronic health record for phenotype stratification", - "prefix": "10.1016", - "volume": "64", - "author": [ - { - "given": "Brett K.", - "family": "Beaulieu-Jones", - "affiliation": [] - }, - { - "given": "Casey S.", - "family": "Greene", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Journal of Biomedical Informatics", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S153204641630140X?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S153204641630140X?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 9, - 14 - ] - ], - "date-time": "2017-09-14T13:26:58Z", - "timestamp": 1505395618000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 12 - ] - ] - }, - "references-count": 46, - "alternative-id": [ - "S153204641630140X" - ], - "URL": "https://doi.org/10.1016/j.jbi.2016.10.007", - "relation": {}, - "subject": [ - "Health Informatics", - "Computer Science Applications" - ], - "container-title-short": "Journal of Biomedical Informatics", - "assertion": [ - { - "value": "Elsevier", - "name": "publisher", - "label": "This article is maintained by" - }, - { - "value": "Semi-supervised learning of the electronic health record for phenotype stratification", - "name": "articletitle", - "label": "Article Title" - }, - { - "value": "Journal of Biomedical Informatics", - "name": "journaltitle", - "label": "Journal Title" - }, - { - "value": "http://dx.doi.org/10.1016/j.jbi.2016.10.007", - "name": "articlelink", - "label": "CrossRef DOI link to publisher maintained version" - }, - { - "value": "article", - "name": "content_type", - "label": "Content Type" - }, - { - "value": "© 2016 The Author(s). Published by Elsevier Inc.", - "name": "copyright", - "label": "Copyright" - } - ], - "id": "5x3uMSKi" - }, - "citation_id": "5x3uMSKi" - }, - "doi:10.1016/j.jhealeco.2016.01.012": { - "source": "doi", - "identifer": "10.1016/j.jhealeco.2016.01.012", - "standard_citation": "doi:10.1016/j.jhealeco.2016.01.012", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T04:02:15Z", - "timestamp": 1507089735902 - }, - "reference-count": 50, - "publisher": "Elsevier BV", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2016, - 5, - 1 - ] - ], - "date-time": "2016-05-01T00:00:00Z", - "timestamp": 1462060800000 - }, - "delay-in-days": 0, - "content-version": "tdm" - } - ], - "content-domain": { - "domain": [ - "elsevier.com", - "sciencedirect.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2016, - 5 - ] - ] - }, - "DOI": "10.1016/j.jhealeco.2016.01.012", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 2, - 13 - ] - ], - "date-time": "2016-02-13T22:55:49Z", - "timestamp": 1455404149000 - }, - "page": "20-33", - "update-policy": "http://dx.doi.org/10.1016/elsevier_cm_policy", - "source": "Crossref", - "is-referenced-by-count": 151, - "title": "Innovation in the pharmaceutical industry: New estimates of R&D costs", - "prefix": "10.1016", - "volume": "47", - "author": [ - { - "given": "Joseph A.", - "family": "DiMasi", - "affiliation": [] - }, - { - "given": "Henry G.", - "family": "Grabowski", - "affiliation": [] - }, - { - "given": "Ronald W.", - "family": "Hansen", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Journal of Health Economics", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S0167629616000291?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S0167629616000291?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T05:51:36Z", - "timestamp": 1498283496000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 5 - ] - ] - }, - "references-count": 50, - "alternative-id": [ - "S0167629616000291" - ], - "URL": "https://doi.org/10.1016/j.jhealeco.2016.01.012", - "relation": {}, - "subject": [ - "Public Health, Environmental and Occupational Health", - "Health Policy" - ], - "container-title-short": "Journal of Health Economics", - "assertion": [ - { - "value": "Elsevier", - "name": "publisher", - "label": "This article is maintained by" - }, - { - "value": "Innovation in the pharmaceutical industry: New estimates of R&D costs", - "name": "articletitle", - "label": "Article Title" - }, - { - "value": "Journal of Health Economics", - "name": "journaltitle", - "label": "Journal Title" - }, - { - "value": "http://dx.doi.org/10.1016/j.jhealeco.2016.01.012", - "name": "articlelink", - "label": "CrossRef DOI link to publisher maintained version" - }, - { - "value": "article", - "name": "content_type", - "label": "Content Type" - }, - { - "value": "Copyright © 2016 Elsevier B.V. All rights reserved.", - "name": "copyright", - "label": "Copyright" - } - ], - "id": "13c9OPizf" - }, - "citation_id": "13c9OPizf" - }, - "doi:10.1016/j.neubiorev.2010.07.002": { - "source": "doi", - "identifer": "10.1016/j.neubiorev.2010.07.002", - "standard_citation": "doi:10.1016/j.neubiorev.2010.07.002", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T07:02:17Z", - "timestamp": 1507100537885 - }, - "reference-count": 60, - "publisher": "Elsevier BV", - "issue": "3", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2011, - 1, - 1 - ] - ], - "date-time": "2011-01-01T00:00:00Z", - "timestamp": 1293840000000 - }, - "delay-in-days": 0, - "content-version": "tdm" - } - ], - "funder": [ - { - "name": "Robert Wood Johnson Foundation Health & Society Scholars program", - "award": [] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2011, - 1 - ] - ] - }, - "DOI": "10.1016/j.neubiorev.2010.07.002", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2010, - 7, - 9 - ] - ], - "date-time": "2010-07-09T08:18:55Z", - "timestamp": 1278663535000 - }, - "page": "565-572", - "source": "Crossref", - "is-referenced-by-count": 253, - "title": "Sex bias in neuroscience and biomedical research", - "prefix": "10.1016", - "volume": "35", - "author": [ - { - "given": "Annaliese K.", - "family": "Beery", - "affiliation": [] - }, - { - "given": "Irving", - "family": "Zucker", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Neuroscience & Biobehavioral Reviews", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S0149763410001156?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S0149763410001156?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 19 - ] - ], - "date-time": "2017-06-19T10:36:15Z", - "timestamp": 1497868575000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2011, - 1 - ] - ] - }, - "references-count": 60, - "alternative-id": [ - "S0149763410001156" - ], - "URL": "https://doi.org/10.1016/j.neubiorev.2010.07.002", - "relation": {}, - "subject": [ - "Cognitive Neuroscience", - "Behavioral Neuroscience", - "Neuropsychology and Physiological Psychology" - ], - "container-title-short": "Neuroscience & Biobehavioral Reviews", - "id": "sOBzMC57" - }, - "citation_id": "sOBzMC57" - }, - "doi:10.1016/j.tig.2015.05.007": { - "source": "doi", - "identifer": "10.1016/j.tig.2015.05.007", - "standard_citation": "doi:10.1016/j.tig.2015.05.007", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T03:43:50Z", - "timestamp": 1502336630450 - }, - "reference-count": 80, - "publisher": "Elsevier BV", - "issue": "8", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2015, - 8, - 1 - ] - ], - "date-time": "2015-08-01T00:00:00Z", - "timestamp": 1438387200000 - }, - "delay-in-days": 0, - "content-version": "tdm" - } - ], - "content-domain": { - "domain": [ - "cell.com", - "bmn.com", - "elsevier.com", - "sciencedirect.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2015, - 8 - ] - ] - }, - "DOI": "10.1016/j.tig.2015.05.007", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 6, - 13 - ] - ], - "date-time": "2015-06-13T21:11:12Z", - "timestamp": 1434229872000 - }, - "page": "426-433", - "update-policy": "http://dx.doi.org/10.1016/elsevier_cm_policy", - "source": "Crossref", - "is-referenced-by-count": 27, - "title": "A unified architecture of transcriptional regulatory elements", - "prefix": "10.1016", - "volume": "31", - "author": [ - { - "given": "Robin", - "family": "Andersson", - "affiliation": [] - }, - { - "given": "Albin", - "family": "Sandelin", - "affiliation": [] - }, - { - "given": "Charles G.", - "family": "Danko", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Trends in Genetics", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S0168952515001043?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S0168952515001043?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 23 - ] - ], - "date-time": "2017-06-23T13:39:51Z", - "timestamp": 1498225191000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 8 - ] - ] - }, - "references-count": 80, - "alternative-id": [ - "S0168952515001043" - ], - "URL": "https://doi.org/10.1016/j.tig.2015.05.007", - "relation": { - "has-preprint": [ - { - "id-type": "doi", - "id": "10.1101/019844", - "asserted-by": "object" - } - ] - }, - "subject": [ - "Genetics" - ], - "container-title-short": "Trends in Genetics", - "assertion": [ - { - "value": "Elsevier", - "name": "publisher", - "label": "This article is maintained by" - }, - { - "value": "A unified architecture of transcriptional regulatory elements", - "name": "articletitle", - "label": "Article Title" - }, - { - "value": "Trends in Genetics", - "name": "journaltitle", - "label": "Journal Title" - }, - { - "value": "http://dx.doi.org/10.1016/j.tig.2015.05.007", - "name": "articlelink", - "label": "CrossRef DOI link to publisher maintained version" - }, - { - "value": "article", - "name": "content_type", - "label": "Content Type" - }, - { - "value": "Copyright © 2015 Elsevier Ltd. All rights reserved.", - "name": "copyright", - "label": "Copyright" - } - ], - "id": "J0PJHcHK" - }, - "citation_id": "J0PJHcHK" - }, - "doi:10.1021/acs.jmedchem.5b01849": { - "source": "doi", - "identifer": "10.1021/acs.jmedchem.5b01849", - "standard_citation": "doi:10.1021/acs.jmedchem.5b01849", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 22 - ] - ], - "date-time": "2017-09-22T04:22:14Z", - "timestamp": 1506054134023 - }, - "reference-count": 138, - "publisher": "American Chemical Society (ACS)", - "issue": "9", - "funder": [ - { - "DOI": "10.13039/501100003006", - "name": "Eidgenössische Technische Hochschule Zürich", - "doi-asserted-by": "publisher", - "award": [] - }, - { - "name": "inSili.com LLC", - "award": [] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 5, - 12 - ] - ] - }, - "DOI": "10.1021/acs.jmedchem.5b01849", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 2, - 16 - ] - ], - "date-time": "2016-02-16T15:36:24Z", - "timestamp": 1455636984000 - }, - "page": "4077-4086", - "source": "Crossref", - "is-referenced-by-count": 11, - "title": "De Novo Design at the Edge of Chaos", - "prefix": "10.1021", - "volume": "59", - "author": [ - { - "given": "Petra", - "family": "Schneider", - "affiliation": [] - }, - { - "given": "Gisbert", - "family": "Schneider", - "affiliation": [] - } - ], - "member": "316", - "container-title": "Journal of Medicinal Chemistry", - "original-title": [], - "link": [ - { - "URL": "http://pubs.acs.org/doi/pdf/10.1021/acs.jmedchem.5b01849", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T06:03:14Z", - "timestamp": 1498284194000 - }, - "score": 1.0, - "subtitle": [ - "Miniperspective" - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 5, - 12 - ] - ] - }, - "references-count": 138, - "alternative-id": [ - "10.1021/acs.jmedchem.5b01849" - ], - "URL": "https://doi.org/10.1021/acs.jmedchem.5b01849", - "relation": {}, - "subject": [ - "Molecular Medicine", - "Drug Discovery" - ], - "container-title-short": "J. Med. Chem.", - "id": "omzv9ryI" - }, - "citation_id": "omzv9ryI" - }, - "doi:10.1021/acs.jproteome.6b00618": { - "source": "doi", - "identifer": "10.1021/acs.jproteome.6b00618", - "standard_citation": "doi:10.1021/acs.jproteome.6b00618", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 4 - ] - ], - "date-time": "2017-09-04T15:02:12Z", - "timestamp": 1504537332845 - }, - "reference-count": 35, - "publisher": "American Chemical Society (ACS)", - "issue": "4", - "funder": [ - { - "DOI": "10.13039/501100002338", - "name": "Ministry of Education of the People's Republic of China", - "doi-asserted-by": "publisher", - "award": [ - "2015zzts163" - ] - }, - { - "DOI": "10.13039/501100001809", - "name": "National Natural Science Foundation of China", - "doi-asserted-by": "publisher", - "award": [ - "21305163", - "21375151", - "81402853", - "21175157" - ] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2017, - 4, - 7 - ] - ] - }, - "DOI": "10.1021/acs.jproteome.6b00618", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2017, - 3, - 6 - ] - ], - "date-time": "2017-03-06T23:08:00Z", - "timestamp": 1488841680000 - }, - "page": "1401-1409", - "source": "Crossref", - "is-referenced-by-count": 3, - "title": "Deep-Learning-Based Drug–Target Interaction Prediction", - "prefix": "10.1021", - "volume": "16", - "author": [ - { - "given": "Ming", - "family": "Wen", - "affiliation": [] - }, - { - "given": "Zhimin", - "family": "Zhang", - "affiliation": [] - }, - { - "given": "Shaoyu", - "family": "Niu", - "affiliation": [] - }, - { - "given": "Haozhi", - "family": "Sha", - "affiliation": [] - }, - { - "given": "Ruihan", - "family": "Yang", - "affiliation": [] - }, - { - "given": "Yonghuan", - "family": "Yun", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-4686-4491", - "authenticated-orcid": true, - "given": "Hongmei", - "family": "Lu", - "affiliation": [] - } - ], - "member": "316", - "published-online": { - "date-parts": [ - [ - 2017, - 3, - 13 - ] - ] - }, - "container-title": "Journal of Proteome Research", - "original-title": [], - "link": [ - { - "URL": "http://pubs.acs.org/doi/pdf/10.1021/acs.jproteome.6b00618", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T12:00:49Z", - "timestamp": 1498392049000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 3, - 13 - ] - ] - }, - "references-count": 35, - "alternative-id": [ - "10.1021/acs.jproteome.6b00618" - ], - "URL": "https://doi.org/10.1021/acs.jproteome.6b00618", - "relation": {}, - "subject": [ - "Biochemistry", - "General Chemistry" - ], - "container-title-short": "J. Proteome Res.", - "id": "oTF8O79C" - }, - "citation_id": "oTF8O79C" - }, - "doi:10.1021/acs.molpharmaceut.5b00982": { - "source": "doi", - "identifer": "10.1021/acs.molpharmaceut.5b00982", - "standard_citation": "doi:10.1021/acs.molpharmaceut.5b00982", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 14 - ] - ], - "date-time": "2017-09-14T09:22:13Z", - "timestamp": 1505380933213 - }, - "reference-count": 53, - "publisher": "American Chemical Society (ACS)", - "issue": "5", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 5, - 2 - ] - ] - }, - "DOI": "10.1021/acs.molpharmaceut.5b00982", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 3, - 23 - ] - ], - "date-time": "2016-03-23T16:10:48Z", - "timestamp": 1458749448000 - }, - "page": "1445-1454", - "source": "Crossref", - "is-referenced-by-count": 29, - "title": "Applications of Deep Learning in Biomedicine", - "prefix": "10.1021", - "volume": "13", - "author": [ - { - "given": "Polina", - "family": "Mamoshina", - "affiliation": [] - }, - { - "given": "Armando", - "family": "Vieira", - "affiliation": [] - }, - { - "given": "Evgeny", - "family": "Putin", - "affiliation": [] - }, - { - "given": "Alex", - "family": "Zhavoronkov", - "affiliation": [] - } - ], - "member": "316", - "container-title": "Molecular Pharmaceutics", - "original-title": [], - "link": [ - { - "URL": "http://pubs.acs.org/doi/pdf/10.1021/acs.molpharmaceut.5b00982", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T09:09:34Z", - "timestamp": 1498295374000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 5, - 2 - ] - ] - }, - "references-count": 53, - "alternative-id": [ - "10.1021/acs.molpharmaceut.5b00982" - ], - "URL": "https://doi.org/10.1021/acs.molpharmaceut.5b00982", - "relation": {}, - "subject": [ - "Molecular Medicine", - "Drug Discovery", - "Pharmaceutical Science" - ], - "container-title-short": "Mol. Pharmaceutics", - "id": "1VZjheOA" - }, - "citation_id": "1VZjheOA" - }, - "doi:10.1021/acs.molpharmaceut.6b00248": { - "source": "doi", - "identifer": "10.1021/acs.molpharmaceut.6b00248", - "standard_citation": "doi:10.1021/acs.molpharmaceut.6b00248", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 4 - ] - ], - "date-time": "2017-09-04T15:02:12Z", - "timestamp": 1504537332840 - }, - "reference-count": 40, - "publisher": "American Chemical Society (ACS)", - "issue": "7", - "license": [ - { - "URL": "http://pubs.acs.org/page/policy/authorchoice_termsofuse.html", - "start": { - "date-parts": [ - [ - 2016, - 6, - 8 - ] - ], - "date-time": "2016-06-08T00:00:00Z", - "timestamp": 1465344000000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 7, - 5 - ] - ] - }, - "DOI": "10.1021/acs.molpharmaceut.6b00248", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 5, - 22 - ] - ], - "date-time": "2016-05-22T01:18:39Z", - "timestamp": 1463879919000 - }, - "page": "2524-2530", - "source": "Crossref", - "is-referenced-by-count": 16, - "title": "Deep Learning Applications for Predicting Pharmacological Properties of Drugs and Drug Repurposing Using Transcriptomic Data", - "prefix": "10.1021", - "volume": "13", - "author": [ - { - "given": "Alexander", - "family": "Aliper", - "affiliation": [] - }, - { - "given": "Sergey", - "family": "Plis", - "affiliation": [] - }, - { - "given": "Artem", - "family": "Artemov", - "affiliation": [] - }, - { - "given": "Alvaro", - "family": "Ulloa", - "affiliation": [] - }, - { - "given": "Polina", - "family": "Mamoshina", - "affiliation": [] - }, - { - "given": "Alex", - "family": "Zhavoronkov", - "affiliation": [] - } - ], - "member": "316", - "container-title": "Molecular Pharmaceutics", - "original-title": [], - "link": [ - { - "URL": "http://pubs.acs.org/doi/pdf/10.1021/acs.molpharmaceut.6b00248", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T14:22:31Z", - "timestamp": 1498314151000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 7, - 5 - ] - ] - }, - "references-count": 40, - "alternative-id": [ - "10.1021/acs.molpharmaceut.6b00248" - ], - "URL": "https://doi.org/10.1021/acs.molpharmaceut.6b00248", - "relation": {}, - "subject": [ - "Molecular Medicine", - "Drug Discovery", - "Pharmaceutical Science" - ], - "container-title-short": "Mol. Pharmaceutics", - "id": "EMDwvRGb" - }, - "citation_id": "EMDwvRGb" - }, - "doi:10.1021/ci400518g": { - "source": "doi", - "identifer": "10.1021/ci400518g", - "standard_citation": "doi:10.1021/ci400518g", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 21 - ] - ], - "date-time": "2017-09-21T19:02:12Z", - "timestamp": 1506020532483 - }, - "reference-count": 25, - "publisher": "American Chemical Society (ACS)", - "issue": "12", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2013, - 12, - 23 - ] - ] - }, - "DOI": "10.1021/ci400518g", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2013, - 11, - 13 - ] - ], - "date-time": "2013-11-13T21:06:40Z", - "timestamp": 1384376800000 - }, - "page": "3373-3383", - "source": "Crossref", - "is-referenced-by-count": 35, - "title": "XenoSite: Accurately Predicting CYP-Mediated Sites of Metabolism with Neural Networks", - "prefix": "10.1021", - "volume": "53", - "author": [ - { - "given": "Jed", - "family": "Zaretzki", - "affiliation": [] - }, - { - "given": "Matthew", - "family": "Matlock", - "affiliation": [] - }, - { - "given": "S. Joshua", - "family": "Swamidass", - "affiliation": [] - } - ], - "member": "316", - "container-title": "Journal of Chemical Information and Modeling", - "original-title": [], - "link": [ - { - "URL": "http://pubs.acs.org/doi/pdf/10.1021/ci400518g", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 22 - ] - ], - "date-time": "2017-06-22T01:54:47Z", - "timestamp": 1498096487000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2013, - 12, - 23 - ] - ] - }, - "references-count": 25, - "alternative-id": [ - "10.1021/ci400518g" - ], - "URL": "https://doi.org/10.1021/ci400518g", - "relation": {}, - "subject": [ - "General Chemistry", - "General Chemical Engineering", - "Library and Information Sciences", - "Computer Science Applications" - ], - "container-title-short": "J. Chem. Inf. Model.", - "id": "7QsMcDYy" - }, - "citation_id": "7QsMcDYy" - }, - "doi:10.1021/ci500340n": { - "source": "doi", - "identifer": "10.1021/ci500340n", - "standard_citation": "doi:10.1021/ci500340n", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 4 - ] - ], - "date-time": "2017-09-04T15:22:12Z", - "timestamp": 1504538532899 - }, - "reference-count": 29, - "publisher": "American Chemical Society (ACS)", - "issue": "9", - "funder": [ - { - "DOI": "10.13039/501100002855", - "name": "Ministry of Science and Technology of the People's Republic of China", - "doi-asserted-by": "publisher", - "award": [ - "2012ZX09503001_001" - ] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2014, - 9, - 22 - ] - ] - }, - "DOI": "10.1021/ci500340n", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 8, - 13 - ] - ], - "date-time": "2014-08-13T13:53:40Z", - "timestamp": 1407938020000 - }, - "page": "2562-2569", - "source": "Crossref", - "is-referenced-by-count": 10, - "title": "Drug–Disease Association and Drug-Repositioning Predictions in Complex Diseases Using Causal Inference–Probabilistic Matrix Factorization", - "prefix": "10.1021", - "volume": "54", - "author": [ - { - "given": "Jihong", - "family": "Yang", - "affiliation": [] - }, - { - "given": "Zheng", - "family": "Li", - "affiliation": [] - }, - { - "given": "Xiaohui", - "family": "Fan", - "affiliation": [] - }, - { - "given": "Yiyu", - "family": "Cheng", - "affiliation": [] - } - ], - "member": "316", - "container-title": "Journal of Chemical Information and Modeling", - "original-title": [], - "link": [ - { - "URL": "http://pubs.acs.org/doi/pdf/10.1021/ci500340n", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 22 - ] - ], - "date-time": "2017-06-22T18:23:50Z", - "timestamp": 1498155830000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 9, - 22 - ] - ] - }, - "references-count": 29, - "alternative-id": [ - "10.1021/ci500340n" - ], - "URL": "https://doi.org/10.1021/ci500340n", - "relation": {}, - "subject": [ - "General Chemistry", - "General Chemical Engineering", - "Library and Information Sciences", - "Computer Science Applications" - ], - "container-title-short": "J. Chem. Inf. Model.", - "id": "16FEYidu2" - }, - "citation_id": "16FEYidu2" - }, - "doi:10.1021/jm048957q": { - "source": "doi", - "identifer": "10.1021/jm048957q", - "standard_citation": "doi:10.1021/jm048957q", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 26 - ] - ], - "date-time": "2017-09-26T08:02:18Z", - "timestamp": 1506412938053 - }, - "reference-count": 0, - "publisher": "American Chemical Society (ACS)", - "issue": "12", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2005, - 6 - ] - ] - }, - "DOI": "10.1021/jm048957q", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2005, - 6, - 9 - ] - ], - "date-time": "2005-06-09T04:18:54Z", - "timestamp": 1118290734000 - }, - "page": "4111-4119", - "source": "Crossref", - "is-referenced-by-count": 233, - "title": "The PDBbind Database:  Methodologies and Updates", - "prefix": "10.1021", - "volume": "48", - "author": [ - { - "given": "Renxiao", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Xueliang", - "family": "Fang", - "affiliation": [] - }, - { - "given": "Yipin", - "family": "Lu", - "affiliation": [] - }, - { - "given": "Chao-Yie", - "family": "Yang", - "affiliation": [] - }, - { - "given": "Shaomeng", - "family": "Wang", - "affiliation": [] - } - ], - "member": "316", - "container-title": "Journal of Medicinal Chemistry", - "original-title": [], - "link": [ - { - "URL": "http://pubs.acs.org/doi/pdf/10.1021/jm048957q", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 9, - 2 - ] - ], - "date-time": "2016-09-02T02:05:06Z", - "timestamp": 1472781906000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2005, - 6 - ] - ] - }, - "references-count": 0, - "alternative-id": [ - "10.1021/jm048957q" - ], - "URL": "https://doi.org/10.1021/jm048957q", - "relation": {}, - "subject": [ - "Molecular Medicine", - "Drug Discovery" - ], - "container-title-short": "J. Med. Chem.", - "id": "YO41GAOP" - }, - "citation_id": "YO41GAOP" - }, - "doi:10.1037/h0037350": { - "source": "doi", - "identifer": "10.1037/h0037350", - "standard_citation": "doi:10.1037/h0037350", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T09:02:22Z", - "timestamp": 1507107742821 - }, - "reference-count": 0, - "publisher": "American Psychological Association (APA)", - "issue": "5", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 1974 - ] - ] - }, - "DOI": "10.1037/h0037350", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2006, - 6, - 8 - ] - ], - "date-time": "2006-06-08T01:08:10Z", - "timestamp": 1149728890000 - }, - "page": "688-701", - "source": "Crossref", - "is-referenced-by-count": 1766, - "title": "Estimating causal effects of treatments in randomized and nonrandomized studies.", - "prefix": "10.1037", - "volume": "66", - "author": [ - { - "given": "Donald B.", - "family": "Rubin", - "affiliation": [] - } - ], - "member": "15", - "container-title": "Journal of Educational Psychology", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2011, - 8, - 23 - ] - ], - "date-time": "2011-08-23T13:47:16Z", - "timestamp": 1314107236000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 1974 - ] - ] - }, - "references-count": 0, - "alternative-id": [ - "1975-06502-001" - ], - "URL": "https://doi.org/10.1037/h0037350", - "relation": {}, - "subject": [ - "Education", - "Developmental and Educational Psychology" - ], - "container-title-short": "Journal of Educational Psychology", - "id": "cpNVdlL7" - }, - "citation_id": "cpNVdlL7" - }, - "doi:10.1038/484155a": { - "source": "doi", - "identifer": "10.1038/484155a", - "standard_citation": "doi:10.1038/484155a", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T12:21:34Z", - "timestamp": 1502367694595 - }, - "reference-count": 0, - "publisher": "Springer Nature", - "issue": "7393", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/484155a", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2012, - 4, - 5 - ] - ], - "date-time": "2012-04-05T10:08:18Z", - "timestamp": 1333620498000 - }, - "page": "155-158", - "source": "Crossref", - "is-referenced-by-count": 8, - "title": "Children of the 90s: Coming of age", - "prefix": "10.1038", - "volume": "484", - "author": [ - { - "given": "Helen", - "family": "Pearson", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2012, - 4, - 11 - ] - ] - }, - "container-title": "Nature", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2012, - 4, - 30 - ] - ], - "date-time": "2012-04-30T14:27:38Z", - "timestamp": 1335796058000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2012, - 4, - 11 - ] - ] - }, - "references-count": 0, - "alternative-id": [ - "484155a" - ], - "URL": "https://doi.org/10.1038/484155a", - "relation": {}, - "subject": [ - "Multidisciplinary" - ], - "container-title-short": "Nature", - "id": "1FjSxrV1k" - }, - "citation_id": "1FjSxrV1k" - }, - "doi:10.1038/nature14539": { - "source": "doi", - "identifer": "10.1038/nature14539", - "standard_citation": "doi:10.1038/nature14539", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T11:02:15Z", - "timestamp": 1507114935016 - }, - "reference-count": 103, - "publisher": "Springer Nature", - "issue": "7553", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/nature14539", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 5, - 26 - ] - ], - "date-time": "2015-05-26T16:10:15Z", - "timestamp": 1432656615000 - }, - "page": "436-444", - "source": "Crossref", - "is-referenced-by-count": 1928, - "title": "Deep learning", - "prefix": "10.1038", - "volume": "521", - "author": [ - { - "given": "Yann", - "family": "LeCun", - "affiliation": [] - }, - { - "given": "Yoshua", - "family": "Bengio", - "affiliation": [] - }, - { - "given": "Geoffrey", - "family": "Hinton", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2015, - 5, - 27 - ] - ] - }, - "container-title": "Nature", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2015, - 5, - 28 - ] - ], - "date-time": "2015-05-28T09:57:36Z", - "timestamp": 1432807056000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 5, - 27 - ] - ] - }, - "references-count": 103, - "alternative-id": [ - "nature14539" - ], - "URL": "https://doi.org/10.1038/nature14539", - "relation": { - "cites": [] - }, - "subject": [ - "Multidisciplinary" - ], - "container-title-short": "Nature", - "id": "BeijBSRE" - }, - "citation_id": "BeijBSRE" - }, - "doi:10.1038/nature21056": { - "source": "doi", - "identifer": "10.1038/nature21056", - "standard_citation": "doi:10.1038/nature21056", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T04:02:42Z", - "timestamp": 1507003362746 - }, - "reference-count": 30, - "publisher": "Springer Nature", - "issue": "7639", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/nature21056", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2017, - 1, - 25 - ] - ], - "date-time": "2017-01-25T21:12:47Z", - "timestamp": 1485378767000 - }, - "page": "115-118", - "source": "Crossref", - "is-referenced-by-count": 74, - "title": "Dermatologist-level classification of skin cancer with deep neural networks", - "prefix": "10.1038", - "volume": "542", - "author": [ - { - "given": "Andre", - "family": "Esteva", - "affiliation": [] - }, - { - "given": "Brett", - "family": "Kuprel", - "affiliation": [] - }, - { - "given": "Roberto A.", - "family": "Novoa", - "affiliation": [] - }, - { - "given": "Justin", - "family": "Ko", - "affiliation": [] - }, - { - "given": "Susan M.", - "family": "Swetter", - "affiliation": [] - }, - { - "given": "Helen M.", - "family": "Blau", - "affiliation": [] - }, - { - "given": "Sebastian", - "family": "Thrun", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2017, - 1, - 25 - ] - ] - }, - "container-title": "Nature", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T09:03:36Z", - "timestamp": 1498381416000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 1, - 25 - ] - ] - }, - "references-count": 30, - "alternative-id": [ - "nature21056" - ], - "URL": "https://doi.org/10.1038/nature21056", - "relation": { - "cites": [] - }, - "subject": [ - "Multidisciplinary" - ], - "container-title-short": "Nature", - "id": "XnYNYoYB" - }, - "citation_id": "XnYNYoYB" - }, - "doi:10.1038/nbt.3313": { - "source": "doi", - "identifer": "10.1038/nbt.3313", - "standard_citation": "doi:10.1038/nbt.3313", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T02:42:17Z", - "timestamp": 1506998537347 - }, - "reference-count": 10, - "publisher": "Springer Nature", - "issue": "8", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/nbt.3313", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 8, - 7 - ] - ], - "date-time": "2015-08-07T16:59:17Z", - "timestamp": 1438966757000 - }, - "page": "825-826", - "source": "Crossref", - "is-referenced-by-count": 14, - "title": "Deep learning for regulatory genomics", - "prefix": "10.1038", - "volume": "33", - "author": [ - { - "given": "Yongjin", - "family": "Park", - "affiliation": [] - }, - { - "given": "Manolis", - "family": "Kellis", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2015, - 8, - 7 - ] - ] - }, - "container-title": "Nature Biotechnology", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 23 - ] - ], - "date-time": "2017-06-23T17:13:48Z", - "timestamp": 1498238028000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 8, - 7 - ] - ] - }, - "references-count": 10, - "alternative-id": [ - "nbt.3313" - ], - "URL": "https://doi.org/10.1038/nbt.3313", - "relation": { - "cites": [] - }, - "subject": [ - "Biotechnology", - "Molecular Medicine", - "Applied Microbiology and Biotechnology", - "Bioengineering", - "Biomedical Engineering" - ], - "container-title-short": "Nat Biotechnol", - "id": "yXqhuueV" - }, - "citation_id": "yXqhuueV" - }, - "doi:10.1038/nbt.3780": { - "source": "doi", - "identifer": "10.1038/nbt.3780", - "standard_citation": "doi:10.1038/nbt.3780", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 11 - ] - ], - "date-time": "2017-09-11T12:22:11Z", - "timestamp": 1505132531877 - }, - "reference-count": 33, - "publisher": "Springer Nature", - "issue": "4", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/nbt.3780", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2017, - 3, - 13 - ] - ], - "date-time": "2017-03-13T16:03:33Z", - "timestamp": 1489421013000 - }, - "page": "342-346", - "source": "Crossref", - "is-referenced-by-count": 3, - "title": "Reproducibility of computational workflows is automated using continuous analysis", - "prefix": "10.1038", - "volume": "35", - "author": [ - { - "given": "Brett K", - "family": "Beaulieu-Jones", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0001-8713-9213", - "authenticated-orcid": false, - "given": "Casey S", - "family": "Greene", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2017, - 3, - 13 - ] - ] - }, - "container-title": "Nature Biotechnology", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T12:32:14Z", - "timestamp": 1498393934000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 3, - 13 - ] - ] - }, - "references-count": 33, - "alternative-id": [ - "nbt.3780" - ], - "URL": "https://doi.org/10.1038/nbt.3780", - "relation": { - "has-preprint": [ - { - "id-type": "doi", - "id": "10.1101/056473", - "asserted-by": "object" - } - ], - "cites": [] - }, - "subject": [ - "Biotechnology", - "Molecular Medicine", - "Applied Microbiology and Biotechnology", - "Bioengineering", - "Biomedical Engineering" - ], - "container-title-short": "Nat Biotechnol", - "id": "Qh7xTLwz" - }, - "citation_id": "Qh7xTLwz" - }, - "doi:10.1038/ncomms13090": { - "source": "doi", - "identifer": "10.1038/ncomms13090", - "standard_citation": "doi:10.1038/ncomms13090", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 20 - ] - ], - "date-time": "2017-09-20T08:02:40Z", - "timestamp": 1505894560615 - }, - "reference-count": 70, - "publisher": "Springer Nature", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/ncomms13090", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 10, - 7 - ] - ], - "date-time": "2016-10-07T09:17:09Z", - "timestamp": 1475831829000 - }, - "page": "13090", - "source": "Crossref", - "is-referenced-by-count": 9, - "title": "Multi-omics integration accurately predicts cellular state in unexplored conditions for Escherichia coli", - "prefix": "10.1038", - "volume": "7", - "author": [ - { - "given": "Minseung", - "family": "Kim", - "affiliation": [] - }, - { - "given": "Navneet", - "family": "Rai", - "affiliation": [] - }, - { - "given": "Violeta", - "family": "Zorraquino", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0003-1104-7616", - "authenticated-orcid": false, - "given": "Ilias", - "family": "Tagkopoulos", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2016, - 10, - 7 - ] - ] - }, - "container-title": "Nature Communications", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T00:34:39Z", - "timestamp": 1498350879000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 10, - 7 - ] - ] - }, - "references-count": 70, - "alternative-id": [ - "ncomms13090" - ], - "URL": "https://doi.org/10.1038/ncomms13090", - "relation": { - "cites": [] - }, - "subject": [ - "General Biochemistry, Genetics and Molecular Biology", - "General Physics and Astronomy", - "General Chemistry" - ], - "container-title-short": "Nat Comms", - "id": "8SMDF816" - }, - "citation_id": "8SMDF816" - }, - "doi:10.1038/ncomms5022": { - "source": "doi", - "identifer": "10.1038/ncomms5022", - "standard_citation": "doi:10.1038/ncomms5022", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 14 - ] - ], - "date-time": "2017-09-14T09:02:14Z", - "timestamp": 1505379734199 - }, - "reference-count": 23, - "publisher": "Springer Nature", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/ncomms5022", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 6, - 24 - ] - ], - "date-time": "2014-06-24T14:59:47Z", - "timestamp": 1403621987000 - }, - "source": "Crossref", - "is-referenced-by-count": 20, - "title": "Temporal disease trajectories condensed from population-wide registry data covering 6.2 million patients", - "prefix": "10.1038", - "volume": "5", - "author": [ - { - "given": "Anders Boeck", - "family": "Jensen", - "affiliation": [] - }, - { - "given": "Pope L.", - "family": "Moseley", - "affiliation": [] - }, - { - "given": "Tudor I.", - "family": "Oprea", - "affiliation": [] - }, - { - "given": "Sabrina Gade", - "family": "Ellesøe", - "affiliation": [] - }, - { - "given": "Robert", - "family": "Eriksson", - "affiliation": [] - }, - { - "given": "Henriette", - "family": "Schmock", - "affiliation": [] - }, - { - "given": "Peter Bjødstrup", - "family": "Jensen", - "affiliation": [] - }, - { - "given": "Lars Juhl", - "family": "Jensen", - "affiliation": [] - }, - { - "given": "Søren", - "family": "Brunak", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2014, - 6, - 24 - ] - ] - }, - "container-title": "Nature Communications", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 22 - ] - ], - "date-time": "2017-06-22T15:27:12Z", - "timestamp": 1498145232000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 6, - 24 - ] - ] - }, - "references-count": 23, - "alternative-id": [ - "ncomms5022" - ], - "URL": "https://doi.org/10.1038/ncomms5022", - "relation": { - "cites": [] - }, - "subject": [ - "General Biochemistry, Genetics and Molecular Biology", - "General Physics and Astronomy", - "General Chemistry" - ], - "container-title-short": "Nat Comms", - "id": "ogs3PPp7" - }, - "citation_id": "ogs3PPp7" - }, - "doi:10.1038/ncomms5308": { - "source": "doi", - "identifer": "10.1038/ncomms5308", - "standard_citation": "doi:10.1038/ncomms5308", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 23 - ] - ], - "date-time": "2017-09-23T16:42:11Z", - "timestamp": 1506184931235 - }, - "reference-count": 25, - "publisher": "Springer Nature", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/ncomms5308", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 7, - 2 - ] - ], - "date-time": "2014-07-02T12:16:16Z", - "timestamp": 1404303376000 - }, - "source": "Crossref", - "is-referenced-by-count": 73, - "title": "Searching for exotic particles in high-energy physics with deep learning", - "prefix": "10.1038", - "volume": "5", - "author": [ - { - "given": "P.", - "family": "Baldi", - "affiliation": [] - }, - { - "given": "P.", - "family": "Sadowski", - "affiliation": [] - }, - { - "given": "D.", - "family": "Whiteson", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2014, - 7, - 2 - ] - ] - }, - "container-title": "Nature Communications", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2014, - 7, - 2 - ] - ], - "date-time": "2014-07-02T12:16:16Z", - "timestamp": 1404303376000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 7, - 2 - ] - ] - }, - "references-count": 25, - "alternative-id": [ - "ncomms5308" - ], - "URL": "https://doi.org/10.1038/ncomms5308", - "relation": { - "cites": [] - }, - "subject": [ - "General Biochemistry, Genetics and Molecular Biology", - "General Physics and Astronomy", - "General Chemistry" - ], - "container-title-short": "Nat Comms", - "id": "TDruxF1s" - }, - "citation_id": "TDruxF1s" - }, - "doi:10.1038/ng.2892": { - "source": "doi", - "identifer": "10.1038/ng.2892", - "standard_citation": "doi:10.1038/ng.2892", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T14:02:16Z", - "timestamp": 1507039336013 - }, - "reference-count": 59, - "publisher": "Springer Nature", - "issue": "3", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/ng.2892", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 2, - 2 - ] - ], - "date-time": "2014-02-02T21:46:05Z", - "timestamp": 1391377565000 - }, - "page": "310-315", - "source": "Crossref", - "is-referenced-by-count": 920, - "title": "A general framework for estimating the relative pathogenicity of human genetic variants", - "prefix": "10.1038", - "volume": "46", - "author": [ - { - "ORCID": "http://orcid.org/0000-0001-9278-5471", - "authenticated-orcid": false, - "given": "Martin", - "family": "Kircher", - "affiliation": [] - }, - { - "given": "Daniela M", - "family": "Witten", - "affiliation": [] - }, - { - "given": "Preti", - "family": "Jain", - "affiliation": [] - }, - { - "given": "Brian J", - "family": "O'Roak", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0001-5509-9923", - "authenticated-orcid": false, - "given": "Gregory M", - "family": "Cooper", - "affiliation": [] - }, - { - "given": "Jay", - "family": "Shendure", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2014, - 2, - 2 - ] - ] - }, - "container-title": "Nature Genetics", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2014, - 2, - 26 - ] - ], - "date-time": "2014-02-26T09:54:48Z", - "timestamp": 1393408488000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 2, - 2 - ] - ] - }, - "references-count": 59, - "alternative-id": [ - "ng.2892" - ], - "URL": "https://doi.org/10.1038/ng.2892", - "relation": { - "cites": [] - }, - "subject": [ - "Genetics" - ], - "container-title-short": "Nat Genet", - "id": "KxEzGxJ6" - }, - "citation_id": "KxEzGxJ6" - }, - "doi:10.1038/ng.3830": { - "source": "doi", - "identifer": "10.1038/ng.3830", - "standard_citation": "doi:10.1038/ng.3830", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 11 - ] - ], - "date-time": "2017-08-11T03:05:17Z", - "timestamp": 1502420717268 - }, - "reference-count": 10, - "publisher": "Springer Nature", - "issue": "4", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/ng.3830", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2017, - 3, - 30 - ] - ], - "date-time": "2017-03-30T09:59:04Z", - "timestamp": 1490867944000 - }, - "page": "483-484", - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Celebrating parasites", - "prefix": "10.1038", - "volume": "49", - "author": [ - { - "given": "Casey S", - "family": "Greene", - "affiliation": [] - }, - { - "given": "Lana X", - "family": "Garmire", - "affiliation": [] - }, - { - "given": "Jack A", - "family": "Gilbert", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-1208-1720", - "authenticated-orcid": false, - "given": "Marylyn D", - "family": "Ritchie", - "affiliation": [] - }, - { - "given": "Lawrence E", - "family": "Hunter", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2017, - 3, - 30 - ] - ] - }, - "container-title": "Nature Genetics", - "original-title": [], - "link": [ - { - "URL": "http://www.nature.com/doifinder/10.1038/ng.3830", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T13:49:38Z", - "timestamp": 1498398578000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 3, - 30 - ] - ] - }, - "references-count": 10, - "alternative-id": [ - "ng.3830" - ], - "URL": "https://doi.org/10.1038/ng.3830", - "relation": { - "cites": [] - }, - "subject": [ - "Genetics" - ], - "container-title-short": "Nat Genet", - "id": "194IoYUs3" - }, - "citation_id": "194IoYUs3" - }, - "doi:10.1038/nmeth.3945": { - "source": "doi", - "identifer": "10.1038/nmeth.3945", - "standard_citation": "doi:10.1038/nmeth.3945", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T16:15:21Z", - "timestamp": 1502295321053 - }, - "reference-count": 1, - "publisher": "Springer Nature", - "issue": "8", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/nmeth.3945", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 7, - 28 - ] - ], - "date-time": "2016-07-28T16:05:57Z", - "timestamp": 1469721957000 - }, - "page": "603-604", - "source": "Crossref", - "is-referenced-by-count": 11, - "title": "Points of Significance: Classification evaluation", - "prefix": "10.1038", - "volume": "13", - "author": [ - { - "given": "Jake", - "family": "Lever", - "affiliation": [] - }, - { - "given": "Martin", - "family": "Krzywinski", - "affiliation": [] - }, - { - "given": "Naomi", - "family": "Altman", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2016, - 7, - 28 - ] - ] - }, - "container-title": "Nature Methods", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2016, - 7, - 28 - ] - ], - "date-time": "2016-07-28T16:06:00Z", - "timestamp": 1469721960000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 7, - 28 - ] - ] - }, - "references-count": 1, - "alternative-id": [ - "nmeth.3945" - ], - "URL": "https://doi.org/10.1038/nmeth.3945", - "relation": { - "cites": [] - }, - "subject": [ - "Biotechnology", - "Cell Biology", - "Biochemistry", - "Molecular Biology" - ], - "container-title-short": "Nat Meth", - "id": "JmHFuXEM" - }, - "citation_id": "JmHFuXEM" - }, - "doi:10.1038/nmeth.4182": { - "source": "doi", - "identifer": "10.1038/nmeth.4182", - "standard_citation": "doi:10.1038/nmeth.4182", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T23:49:39Z", - "timestamp": 1502408979011 - }, - "reference-count": 44, - "publisher": "Springer Nature", - "issue": "4", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/nmeth.4182", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2017, - 2, - 21 - ] - ], - "date-time": "2017-02-21T15:31:03Z", - "timestamp": 1487691063000 - }, - "page": "403-406", - "source": "Crossref", - "is-referenced-by-count": 1, - "title": "Prospective identification of hematopoietic lineage choice by deep learning", - "prefix": "10.1038", - "volume": "14", - "author": [ - { - "given": "Felix", - "family": "Buggenthin", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0001-5587-6761", - "authenticated-orcid": false, - "given": "Florian", - "family": "Buettner", - "affiliation": [] - }, - { - "given": "Philipp S", - "family": "Hoppe", - "affiliation": [] - }, - { - "given": "Max", - "family": "Endele", - "affiliation": [] - }, - { - "given": "Manuel", - "family": "Kroiss", - "affiliation": [] - }, - { - "given": "Michael", - "family": "Strasser", - "affiliation": [] - }, - { - "given": "Michael", - "family": "Schwarzfischer", - "affiliation": [] - }, - { - "given": "Dirk", - "family": "Loeffler", - "affiliation": [] - }, - { - "given": "Konstantinos D", - "family": "Kokkaliaris", - "affiliation": [] - }, - { - "given": "Oliver", - "family": "Hilsenbeck", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-9271-8815", - "authenticated-orcid": false, - "given": "Timm", - "family": "Schroeder", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-2419-1943", - "authenticated-orcid": false, - "given": "Fabian J", - "family": "Theis", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0003-2154-4552", - "authenticated-orcid": false, - "given": "Carsten", - "family": "Marr", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2017, - 2, - 20 - ] - ] - }, - "container-title": "Nature Methods", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T11:02:25Z", - "timestamp": 1498388545000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 2, - 20 - ] - ] - }, - "references-count": 44, - "alternative-id": [ - "nmeth.4182" - ], - "URL": "https://doi.org/10.1038/nmeth.4182", - "relation": { - "cites": [] - }, - "subject": [ - "Biotechnology", - "Cell Biology", - "Biochemistry", - "Molecular Biology" - ], - "container-title-short": "Nat Meth", - "id": "On4vW5aU" - }, - "citation_id": "On4vW5aU" - }, - "doi:10.1038/npjsba.2016.15": { - "source": "doi", - "identifer": "10.1038/npjsba.2016.15", - "standard_citation": "doi:10.1038/npjsba.2016.15", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T16:49:09Z", - "timestamp": 1502297349351 - }, - "reference-count": 34, - "publisher": "Springer Nature", - "issue": "1", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2016, - 8, - 4 - ] - ], - "date-time": "2016-08-04T00:00:00Z", - "timestamp": 1470268800000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "link.springer.com" - ], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 12 - ] - ] - }, - "DOI": "10.1038/npjsba.2016.15", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 8, - 4 - ] - ], - "date-time": "2016-08-04T11:08:37Z", - "timestamp": 1470308917000 - }, - "update-policy": "http://dx.doi.org/10.1007/springer_crossmark_policy", - "source": "Crossref", - "is-referenced-by-count": 7, - "title": "L1000CDS2: LINCS L1000 characteristic direction signatures search engine", - "prefix": "10.1038", - "volume": "2", - "author": [ - { - "given": "Qiaonan", - "family": "Duan", - "affiliation": [] - }, - { - "given": "St Patrick", - "family": "Reid", - "affiliation": [] - }, - { - "given": "Neil R", - "family": "Clark", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-1415-1286", - "authenticated-orcid": false, - "given": "Zichen", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Nicolas F", - "family": "Fernandez", - "affiliation": [] - }, - { - "given": "Andrew D", - "family": "Rouillard", - "affiliation": [] - }, - { - "given": "Ben", - "family": "Readhead", - "affiliation": [] - }, - { - "given": "Sarah R", - "family": "Tritsch", - "affiliation": [] - }, - { - "given": "Rachel", - "family": "Hodos", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0003-1337-7598", - "authenticated-orcid": false, - "given": "Marc", - "family": "Hafner", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0003-1415-6295", - "authenticated-orcid": false, - "given": "Mario", - "family": "Niepel", - "affiliation": [] - }, - { - "given": "Peter K", - "family": "Sorger", - "affiliation": [] - }, - { - "given": "Joel T", - "family": "Dudley", - "affiliation": [] - }, - { - "given": "Sina", - "family": "Bavari", - "affiliation": [] - }, - { - "given": "Rekha G", - "family": "Panchal", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-6904-1017", - "authenticated-orcid": false, - "given": "Avi", - "family": "Ma’ayan", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2016, - 8, - 4 - ] - ] - }, - "container-title": "npj Systems Biology and Applications", - "original-title": [], - "link": [ - { - "URL": "http://www.nature.com/articles/npjsba201615.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://www.nature.com/articles/npjsba201615", - "content-type": "text/html", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://www.nature.com/articles/npjsba201615.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T19:56:35Z", - "timestamp": 1498334195000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 8, - 4 - ] - ] - }, - "references-count": 34, - "alternative-id": [ - "BFnpjsba201615" - ], - "URL": "https://doi.org/10.1038/npjsba.2016.15", - "relation": { - "cites": [] - }, - "container-title-short": "npj Syst Biol Appl", - "article-number": "16015", - "id": "cQAldRdg" - }, - "citation_id": "cQAldRdg" - }, - "doi:10.1038/nrd1086": { - "source": "doi", - "identifer": "10.1038/nrd1086", - "standard_citation": "doi:10.1038/nrd1086", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 30 - ] - ], - "date-time": "2017-09-30T20:42:14Z", - "timestamp": 1506804134753 - }, - "reference-count": 0, - "publisher": "Springer Nature", - "issue": "5", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2003, - 5 - ] - ] - }, - "DOI": "10.1038/nrd1086", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2003, - 5, - 15 - ] - ], - "date-time": "2003-05-15T17:19:25Z", - "timestamp": 1053019165000 - }, - "page": "369-378", - "source": "Crossref", - "is-referenced-by-count": 449, - "title": "A guide to drug discovery: Hit and lead generation: beyond high-throughput screening", - "prefix": "10.1038", - "volume": "2", - "author": [ - { - "given": "Konrad H.", - "family": "Bleicher", - "affiliation": [] - }, - { - "given": "Hans-Joachim", - "family": "Böhm", - "affiliation": [] - }, - { - "given": "Klaus", - "family": "Müller", - "affiliation": [] - }, - { - "given": "Alexander I.", - "family": "Alanine", - "affiliation": [] - } - ], - "member": "339", - "container-title": "Nature Reviews Drug Discovery", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2011, - 8, - 22 - ] - ], - "date-time": "2011-08-22T04:24:18Z", - "timestamp": 1313987058000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2003, - 5 - ] - ] - }, - "references-count": 0, - "alternative-id": [ - "nrd1086" - ], - "URL": "https://doi.org/10.1038/nrd1086", - "relation": {}, - "subject": [ - "Pharmacology", - "Drug Discovery", - "General Medicine" - ], - "container-title-short": "Nat Rev Drug Discov", - "id": "RAadmvJN" - }, - "citation_id": "RAadmvJN" - }, - "doi:10.1038/nrd4609": { - "source": "doi", - "identifer": "10.1038/nrd4609", - "standard_citation": "doi:10.1038/nrd4609", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 2 - ] - ], - "date-time": "2017-10-02T22:02:11Z", - "timestamp": 1506981731966 - }, - "reference-count": 26, - "publisher": "Springer Nature", - "issue": "7", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/nrd4609", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 6, - 19 - ] - ], - "date-time": "2015-06-19T16:10:19Z", - "timestamp": 1434730219000 - }, - "page": "475-486", - "source": "Crossref", - "is-referenced-by-count": 160, - "title": "An analysis of the attrition of drug candidates from four major pharmaceutical companies", - "prefix": "10.1038", - "volume": "14", - "author": [ - { - "given": "Michael J.", - "family": "Waring", - "affiliation": [] - }, - { - "given": "John", - "family": "Arrowsmith", - "affiliation": [] - }, - { - "given": "Andrew R.", - "family": "Leach", - "affiliation": [] - }, - { - "given": "Paul D.", - "family": "Leeson", - "affiliation": [] - }, - { - "given": "Sam", - "family": "Mandrell", - "affiliation": [] - }, - { - "given": "Robert M.", - "family": "Owen", - "affiliation": [] - }, - { - "given": "Garry", - "family": "Pairaudeau", - "affiliation": [] - }, - { - "given": "William D.", - "family": "Pennie", - "affiliation": [] - }, - { - "given": "Stephen D.", - "family": "Pickett", - "affiliation": [] - }, - { - "given": "Jibo", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Owen", - "family": "Wallace", - "affiliation": [] - }, - { - "given": "Alex", - "family": "Weir", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2015, - 6, - 19 - ] - ] - }, - "container-title": "Nature Reviews Drug Discovery", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2015, - 7, - 1 - ] - ], - "date-time": "2015-07-01T07:41:29Z", - "timestamp": 1435736489000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 6, - 19 - ] - ] - }, - "references-count": 26, - "alternative-id": [ - "nrd4609" - ], - "URL": "https://doi.org/10.1038/nrd4609", - "relation": { - "cites": [] - }, - "subject": [ - "Pharmacology", - "Drug Discovery", - "General Medicine" - ], - "container-title-short": "Nat Rev Drug Discov", - "id": "79Ktl2" - }, - "citation_id": "79Ktl2" - }, - "doi:10.1038/nrg2813": { - "source": "doi", - "identifer": "10.1038/nrg2813", - "standard_citation": "doi:10.1038/nrg2813", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T08:42:17Z", - "timestamp": 1507106537600 - }, - "reference-count": 44, - "publisher": "Springer Nature", - "issue": "7", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/nrg2813", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2010, - 6, - 15 - ] - ], - "date-time": "2010-06-15T04:03:11Z", - "timestamp": 1276574591000 - }, - "page": "459-463", - "source": "Crossref", - "is-referenced-by-count": 309, - "title": "New approaches to population stratification in genome-wide association studies", - "prefix": "10.1038", - "volume": "11", - "author": [ - { - "given": "Alkes L.", - "family": "Price", - "affiliation": [] - }, - { - "given": "Noah A.", - "family": "Zaitlen", - "affiliation": [] - }, - { - "given": "David", - "family": "Reich", - "affiliation": [] - }, - { - "given": "Nick", - "family": "Patterson", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2010, - 6, - 15 - ] - ] - }, - "container-title": "Nature Reviews Genetics", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2011, - 8, - 10 - ] - ], - "date-time": "2011-08-10T10:27:49Z", - "timestamp": 1312972069000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2010, - 6, - 15 - ] - ] - }, - "references-count": 44, - "alternative-id": [ - "nrg2813" - ], - "URL": "https://doi.org/10.1038/nrg2813", - "relation": { - "cites": [] - }, - "subject": [ - "Genetics(clinical)", - "Genetics", - "Molecular Biology" - ], - "container-title-short": "Nat Rev Genet", - "id": "T3GG8iJN" - }, - "citation_id": "T3GG8iJN" - }, - "doi:10.1038/nrg3208": { - "source": "doi", - "identifer": "10.1038/nrg3208", - "standard_citation": "doi:10.1038/nrg3208", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T02:22:22Z", - "timestamp": 1506997342171 - }, - "reference-count": 112, - "publisher": "Springer Nature", - "issue": "6", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/nrg3208", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2012, - 5, - 2 - ] - ], - "date-time": "2012-05-02T05:52:24Z", - "timestamp": 1335937944000 - }, - "page": "395-405", - "source": "Crossref", - "is-referenced-by-count": 294, - "title": "Mining electronic health records: towards better research applications and clinical care", - "prefix": "10.1038", - "volume": "13", - "author": [ - { - "given": "Peter B.", - "family": "Jensen", - "affiliation": [] - }, - { - "given": "Lars J.", - "family": "Jensen", - "affiliation": [] - }, - { - "given": "Søren", - "family": "Brunak", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2012, - 5, - 2 - ] - ] - }, - "container-title": "Nature Reviews Genetics", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 20 - ] - ], - "date-time": "2017-06-20T18:13:32Z", - "timestamp": 1497982412000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2012, - 5, - 2 - ] - ] - }, - "references-count": 112, - "alternative-id": [ - "nrg3208" - ], - "URL": "https://doi.org/10.1038/nrg3208", - "relation": { - "cites": [] - }, - "subject": [ - "Genetics(clinical)", - "Genetics", - "Molecular Biology" - ], - "container-title-short": "Nat Rev Genet", - "id": "FkSZ1qmz" - }, - "citation_id": "FkSZ1qmz" - }, - "doi:10.1038/nrg3458": { - "source": "doi", - "identifer": "10.1038/nrg3458", - "standard_citation": "doi:10.1038/nrg3458", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T14:02:16Z", - "timestamp": 1507039336024 - }, - "reference-count": 75, - "publisher": "Springer Nature", - "issue": "4", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/nrg3458", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2013, - 3, - 18 - ] - ], - "date-time": "2013-03-18T08:46:41Z", - "timestamp": 1363596401000 - }, - "page": "288-295", - "source": "Crossref", - "is-referenced-by-count": 79, - "title": "Enhancers: five essential questions", - "prefix": "10.1038", - "volume": "14", - "author": [ - { - "given": "Len A.", - "family": "Pennacchio", - "affiliation": [] - }, - { - "given": "Wendy", - "family": "Bickmore", - "affiliation": [] - }, - { - "given": "Ann", - "family": "Dean", - "affiliation": [] - }, - { - "given": "Marcelo A.", - "family": "Nobrega", - "affiliation": [] - }, - { - "given": "Gill", - "family": "Bejerano", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2013, - 3, - 18 - ] - ] - }, - "container-title": "Nature Reviews Genetics", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 21 - ] - ], - "date-time": "2017-06-21T09:33:07Z", - "timestamp": 1498037587000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2013, - 3, - 18 - ] - ] - }, - "references-count": 75, - "alternative-id": [ - "nrg3458" - ], - "URL": "https://doi.org/10.1038/nrg3458", - "relation": { - "cites": [] - }, - "subject": [ - "Genetics(clinical)", - "Genetics", - "Molecular Biology" - ], - "container-title-short": "Nat Rev Genet", - "id": "8yA3foA6" - }, - "citation_id": "8yA3foA6" - }, - "doi:10.1038/nrg3920": { - "source": "doi", - "identifer": "10.1038/nrg3920", - "standard_citation": "doi:10.1038/nrg3920", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T07:22:12Z", - "timestamp": 1507015332875 - }, - "reference-count": 68, - "publisher": "Springer Nature", - "issue": "6", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/nrg3920", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 5, - 7 - ] - ], - "date-time": "2015-05-07T07:07:44Z", - "timestamp": 1430982464000 - }, - "page": "321-332", - "source": "Crossref", - "is-referenced-by-count": 91, - "title": "Machine learning applications in genetics and genomics", - "prefix": "10.1038", - "volume": "16", - "author": [ - { - "given": "Maxwell W.", - "family": "Libbrecht", - "affiliation": [] - }, - { - "given": "William Stafford", - "family": "Noble", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2015, - 5, - 7 - ] - ] - }, - "container-title": "Nature Reviews Genetics", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 23 - ] - ], - "date-time": "2017-06-23T11:06:24Z", - "timestamp": 1498215984000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 5, - 7 - ] - ] - }, - "references-count": 68, - "alternative-id": [ - "nrg3920" - ], - "URL": "https://doi.org/10.1038/nrg3920", - "relation": { - "cites": [] - }, - "subject": [ - "Genetics(clinical)", - "Genetics", - "Molecular Biology" - ], - "container-title-short": "Nat Rev Genet", - "id": "TNPWa7Br" - }, - "citation_id": "TNPWa7Br" - }, - "doi:10.1038/srep11476": { - "source": "doi", - "identifer": "10.1038/srep11476", - "standard_citation": "doi:10.1038/srep11476", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 22 - ] - ], - "date-time": "2017-09-22T06:42:12Z", - "timestamp": 1506062532567 - }, - "reference-count": 48, - "publisher": "Springer Nature", - "issue": "1", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2015, - 6, - 22 - ] - ], - "date-time": "2015-06-22T00:00:00Z", - "timestamp": 1434931200000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "link.springer.com" - ], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015, - 9 - ] - ] - }, - "DOI": "10.1038/srep11476", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 6, - 22 - ] - ], - "date-time": "2015-06-22T15:18:26Z", - "timestamp": 1434986306000 - }, - "update-policy": "http://dx.doi.org/10.1007/springer_crossmark_policy", - "source": "Crossref", - "is-referenced-by-count": 45, - "title": "Improving prediction of secondary structure, local backbone angles and solvent accessible surface area of proteins by iterative deep learning", - "prefix": "10.1038", - "volume": "5", - "author": [ - { - "given": "Rhys", - "family": "Heffernan", - "affiliation": [] - }, - { - "given": "Kuldip", - "family": "Paliwal", - "affiliation": [] - }, - { - "given": "James", - "family": "Lyons", - "affiliation": [] - }, - { - "given": "Abdollah", - "family": "Dehzangi", - "affiliation": [] - }, - { - "given": "Alok", - "family": "Sharma", - "affiliation": [] - }, - { - "given": "Jihua", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Abdul", - "family": "Sattar", - "affiliation": [] - }, - { - "given": "Yuedong", - "family": "Yang", - "affiliation": [] - }, - { - "given": "Yaoqi", - "family": "Zhou", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2015, - 6, - 22 - ] - ] - }, - "container-title": "Scientific Reports", - "original-title": [], - "link": [ - { - "URL": "http://www.nature.com/articles/srep11476", - "content-type": "text/html", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://www.nature.com/articles/srep11476.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://www.nature.com/articles/srep11476.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 1 - ] - ], - "date-time": "2017-08-01T14:33:04Z", - "timestamp": 1501597984000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 6, - 22 - ] - ] - }, - "references-count": 48, - "alternative-id": [ - "BFsrep11476" - ], - "URL": "https://doi.org/10.1038/srep11476", - "relation": { - "cites": [] - }, - "subject": [ - "Multidisciplinary" - ], - "container-title-short": "Sci Rep", - "article-number": "11476", - "id": "UpFrhdJf" - }, - "citation_id": "UpFrhdJf" - }, - "doi:10.1038/srep18962": { - "source": "doi", - "identifer": "10.1038/srep18962", - "standard_citation": "doi:10.1038/srep18962", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 25 - ] - ], - "date-time": "2017-08-25T09:03:12Z", - "timestamp": 1503651792041 - }, - "reference-count": 79, - "publisher": "Springer Nature", - "issue": "1", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2016, - 1, - 11 - ] - ], - "date-time": "2016-01-11T00:00:00Z", - "timestamp": 1452470400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "link.springer.com" - ], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 5 - ] - ] - }, - "DOI": "10.1038/srep18962", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 1, - 11 - ] - ], - "date-time": "2016-01-11T09:48:56Z", - "timestamp": 1452505736000 - }, - "update-policy": "http://dx.doi.org/10.1007/springer_crossmark_policy", - "source": "Crossref", - "is-referenced-by-count": 24, - "title": "Protein Secondary Structure Prediction Using Deep Convolutional Neural Fields", - "prefix": "10.1038", - "volume": "6", - "author": [ - { - "given": "Sheng", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Jian", - "family": "Peng", - "affiliation": [] - }, - { - "given": "Jianzhu", - "family": "Ma", - "affiliation": [] - }, - { - "given": "Jinbo", - "family": "Xu", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2016, - 1, - 11 - ] - ] - }, - "container-title": "Scientific Reports", - "original-title": [], - "link": [ - { - "URL": "http://www.nature.com/articles/srep18962", - "content-type": "text/html", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://www.nature.com/articles/srep18962.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://www.nature.com/articles/srep18962.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T21:06:15Z", - "timestamp": 1502312775000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 1, - 11 - ] - ] - }, - "references-count": 79, - "alternative-id": [ - "BFsrep18962" - ], - "URL": "https://doi.org/10.1038/srep18962", - "relation": { - "cites": [] - }, - "subject": [ - "Multidisciplinary" - ], - "container-title-short": "Sci Rep", - "article-number": "18962", - "id": "UO8L6nd" - }, - "citation_id": "UO8L6nd" - }, - "doi:10.1038/srep26094": { - "source": "doi", - "identifer": "10.1038/srep26094", - "standard_citation": "doi:10.1038/srep26094", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T10:22:20Z", - "timestamp": 1507112540995 - }, - "reference-count": 48, - "publisher": "Springer Nature", - "issue": "1", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2016, - 5, - 17 - ] - ], - "date-time": "2016-05-17T00:00:00Z", - "timestamp": 1463443200000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "link.springer.com" - ], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 9 - ] - ] - }, - "DOI": "10.1038/srep26094", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 5, - 17 - ] - ], - "date-time": "2016-05-17T09:38:50Z", - "timestamp": 1463477930000 - }, - "update-policy": "http://dx.doi.org/10.1007/springer_crossmark_policy", - "source": "Crossref", - "is-referenced-by-count": 32, - "title": "Deep Patient: An Unsupervised Representation to Predict the Future of Patients from the Electronic Health Records", - "prefix": "10.1038", - "volume": "6", - "author": [ - { - "given": "Riccardo", - "family": "Miotto", - "affiliation": [] - }, - { - "given": "Li", - "family": "Li", - "affiliation": [] - }, - { - "given": "Brian A.", - "family": "Kidd", - "affiliation": [] - }, - { - "given": "Joel T.", - "family": "Dudley", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2016, - 5, - 17 - ] - ] - }, - "container-title": "Scientific Reports", - "original-title": [], - "link": [ - { - "URL": "http://www.nature.com/articles/srep26094.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://www.nature.com/articles/srep26094", - "content-type": "text/html", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://www.nature.com/articles/srep26094.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T13:47:27Z", - "timestamp": 1498312047000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 5, - 17 - ] - ] - }, - "references-count": 48, - "alternative-id": [ - "BFsrep26094" - ], - "URL": "https://doi.org/10.1038/srep26094", - "relation": { - "cites": [] - }, - "subject": [ - "Multidisciplinary" - ], - "container-title-short": "Sci Rep", - "article-number": "26094", - "id": "WrNCJ9sO" - }, - "citation_id": "WrNCJ9sO" - }, - "doi:10.1038/srep32404": { - "source": "doi", - "identifer": "10.1038/srep32404", - "standard_citation": "doi:10.1038/srep32404", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 14 - ] - ], - "date-time": "2017-08-14T20:40:52Z", - "timestamp": 1502743252821 - }, - "reference-count": 58, - "publisher": "Springer Nature", - "issue": "1", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2016, - 8, - 31 - ] - ], - "date-time": "2016-08-31T00:00:00Z", - "timestamp": 1472601600000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "link.springer.com" - ], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 10 - ] - ] - }, - "DOI": "10.1038/srep32404", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 8, - 31 - ] - ], - "date-time": "2016-08-31T16:36:25Z", - "timestamp": 1472661385000 - }, - "update-policy": "http://dx.doi.org/10.1007/springer_crossmark_policy", - "source": "Crossref", - "is-referenced-by-count": 1, - "title": "Large-Scale Discovery of Disease-Disease and Disease-Gene Associations", - "prefix": "10.1038", - "volume": "6", - "author": [ - { - "given": "Djordje", - "family": "Gligorijevic", - "affiliation": [] - }, - { - "given": "Jelena", - "family": "Stojanovic", - "affiliation": [] - }, - { - "given": "Nemanja", - "family": "Djuric", - "affiliation": [] - }, - { - "given": "Vladan", - "family": "Radosavljevic", - "affiliation": [] - }, - { - "given": "Mihajlo", - "family": "Grbovic", - "affiliation": [] - }, - { - "given": "Rob J.", - "family": "Kulathinal", - "affiliation": [] - }, - { - "given": "Zoran", - "family": "Obradovic", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2016, - 8, - 31 - ] - ] - }, - "container-title": "Scientific Reports", - "original-title": [], - "link": [ - { - "URL": "http://www.nature.com/articles/srep32404.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://www.nature.com/articles/srep32404", - "content-type": "text/html", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://www.nature.com/articles/srep32404.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T21:43:40Z", - "timestamp": 1498340620000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 8, - 31 - ] - ] - }, - "references-count": 58, - "alternative-id": [ - "BFsrep32404" - ], - "URL": "https://doi.org/10.1038/srep32404", - "relation": { - "cites": [] - }, - "subject": [ - "Multidisciplinary" - ], - "container-title-short": "Sci Rep", - "article-number": "32404", - "id": "1G2xP5yOM" - }, - "citation_id": "1G2xP5yOM" - }, - "doi:10.1056/nejme1516564": { - "source": "doi", - "identifer": "10.1056/nejme1516564", - "standard_citation": "doi:10.1056/nejme1516564", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 23 - ] - ], - "date-time": "2017-09-23T13:22:16Z", - "timestamp": 1506172936885 - }, - "reference-count": 1, - "publisher": "New England Journal of Medicine (NEJM/MMS)", - "issue": "3", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 1, - 21 - ] - ] - }, - "DOI": "10.1056/nejme1516564", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 1, - 20 - ] - ], - "date-time": "2016-01-20T22:05:44Z", - "timestamp": 1453327544000 - }, - "page": "276-277", - "source": "Crossref", - "is-referenced-by-count": 79, - "title": "Data Sharing", - "prefix": "10.1056", - "volume": "374", - "author": [ - { - "given": "Dan L.", - "family": "Longo", - "affiliation": [] - }, - { - "given": "Jeffrey M.", - "family": "Drazen", - "affiliation": [] - } - ], - "member": "150", - "container-title": "New England Journal of Medicine", - "original-title": [], - "link": [ - { - "URL": "http://www.nejm.org/doi/pdf/10.1056/NEJMe1516564", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 5, - 8 - ] - ], - "date-time": "2017-05-08T20:26:11Z", - "timestamp": 1494275171000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 1, - 21 - ] - ] - }, - "references-count": 1, - "alternative-id": [ - "10.1056/NEJMe1516564" - ], - "URL": "https://doi.org/10.1056/nejme1516564", - "relation": {}, - "subject": [ - "General Medicine" - ], - "container-title-short": "N Engl J Med", - "id": "o0F1MXBC" - }, - "citation_id": "o0F1MXBC" - }, - "doi:10.1056/nejmp1006304": { - "source": "doi", - "identifer": "10.1056/nejmp1006304", - "standard_citation": "doi:10.1056/nejmp1006304", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 28 - ] - ], - "date-time": "2017-09-28T16:02:20Z", - "timestamp": 1506614540795 - }, - "reference-count": 1, - "publisher": "New England Journal of Medicine (NEJM/MMS)", - "issue": "4", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2010, - 7, - 22 - ] - ] - }, - "DOI": "10.1056/nejmp1006304", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2010, - 6, - 16 - ] - ], - "date-time": "2010-06-16T01:56:14Z", - "timestamp": 1276653374000 - }, - "page": "301-304", - "source": "Crossref", - "is-referenced-by-count": 646, - "title": "The Path to Personalized Medicine", - "prefix": "10.1056", - "volume": "363", - "author": [ - { - "given": "Margaret A.", - "family": "Hamburg", - "affiliation": [] - }, - { - "given": "Francis S.", - "family": "Collins", - "affiliation": [] - } - ], - "member": "150", - "container-title": "New England Journal of Medicine", - "original-title": [], - "link": [ - { - "URL": "http://www.nejm.org/doi/pdf/10.1056/NEJMp1006304", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 11, - 2 - ] - ], - "date-time": "2016-11-02T19:51:34Z", - "timestamp": 1478116294000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2010, - 7, - 22 - ] - ] - }, - "references-count": 1, - "alternative-id": [ - "10.1056/NEJMp1006304" - ], - "URL": "https://doi.org/10.1056/nejmp1006304", - "relation": {}, - "subject": [ - "General Medicine" - ], - "container-title-short": "N Engl J Med", - "id": "VOQtVhWs" - }, - "citation_id": "VOQtVhWs" - }, - "doi:10.1073/pnas.0805923106": { - "source": "doi", - "identifer": "10.1073/pnas.0805923106", - "standard_citation": "doi:10.1073/pnas.0805923106", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 9 - ] - ], - "date-time": "2017-09-09T00:42:10Z", - "timestamp": 1504917730084 - }, - "reference-count": 35, - "publisher": "Proceedings of the National Academy of Sciences", - "issue": "1", - "content-domain": { - "domain": [ - "pnas.org" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2009, - 1, - 6 - ] - ] - }, - "DOI": "10.1073/pnas.0805923106", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2008, - 12, - 31 - ] - ], - "date-time": "2008-12-31T01:26:09Z", - "timestamp": 1230686769000 - }, - "page": "67-72", - "update-policy": "http://dx.doi.org/10.1073/pnas.cm10313", - "source": "Crossref", - "is-referenced-by-count": 302, - "title": "Identification of direct residue contacts in protein-protein interaction by message passing", - "prefix": "10.1073", - "volume": "106", - "author": [ - { - "given": "M.", - "family": "Weigt", - "affiliation": [] - }, - { - "given": "R. A.", - "family": "White", - "affiliation": [] - }, - { - "given": "H.", - "family": "Szurmant", - "affiliation": [] - }, - { - "given": "J. A.", - "family": "Hoch", - "affiliation": [] - }, - { - "given": "T.", - "family": "Hwa", - "affiliation": [] - } - ], - "member": "341", - "published-online": { - "date-parts": [ - [ - 2008, - 12, - 30 - ] - ] - }, - "container-title": "Proceedings of the National Academy of Sciences", - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1073/pnas.0805923106", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 18 - ] - ], - "date-time": "2017-06-18T15:12:42Z", - "timestamp": 1497798762000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2008, - 12, - 30 - ] - ] - }, - "references-count": 35, - "URL": "https://doi.org/10.1073/pnas.0805923106", - "relation": { - "cites": [] - }, - "subject": [ - "Multidisciplinary" - ], - "container-title-short": "Proceedings of the National Academy of Sciences", - "id": "kboAopkh" - }, - "citation_id": "kboAopkh" - }, - "doi:10.1073/pnas.2136655100": { - "source": "doi", - "identifer": "10.1073/pnas.2136655100", - "standard_citation": "doi:10.1073/pnas.2136655100", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 11 - ] - ], - "date-time": "2017-09-11T15:22:12Z", - "timestamp": 1505143332932 - }, - "reference-count": 20, - "publisher": "Proceedings of the National Academy of Sciences", - "issue": "26", - "content-domain": { - "domain": [ - "pnas.org" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2003, - 12, - 23 - ] - ] - }, - "DOI": "10.1073/pnas.2136655100", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2003, - 12, - 23 - ] - ], - "date-time": "2003-12-23T17:58:11Z", - "timestamp": 1072202291000 - }, - "page": "15776-15781", - "update-policy": "http://dx.doi.org/10.1073/pnas.cm10313", - "source": "Crossref", - "is-referenced-by-count": 256, - "title": "Cap analysis gene expression for high-throughput analysis of transcriptional starting point and identification of promoter usage", - "prefix": "10.1073", - "volume": "100", - "author": [ - { - "given": "T.", - "family": "Shiraki", - "affiliation": [] - }, - { - "given": "S.", - "family": "Kondo", - "affiliation": [] - }, - { - "given": "S.", - "family": "Katayama", - "affiliation": [] - }, - { - "given": "K.", - "family": "Waki", - "affiliation": [] - }, - { - "given": "T.", - "family": "Kasukawa", - "affiliation": [] - }, - { - "given": "H.", - "family": "Kawaji", - "affiliation": [] - }, - { - "given": "R.", - "family": "Kodzius", - "affiliation": [] - }, - { - "given": "A.", - "family": "Watahiki", - "affiliation": [] - }, - { - "given": "M.", - "family": "Nakamura", - "affiliation": [] - }, - { - "given": "T.", - "family": "Arakawa", - "affiliation": [] - }, - { - "given": "S.", - "family": "Fukuda", - "affiliation": [] - }, - { - "given": "D.", - "family": "Sasaki", - "affiliation": [] - }, - { - "given": "A.", - "family": "Podhajska", - "affiliation": [] - }, - { - "given": "M.", - "family": "Harbers", - "affiliation": [] - }, - { - "given": "J.", - "family": "Kawai", - "affiliation": [] - }, - { - "given": "P.", - "family": "Carninci", - "affiliation": [] - }, - { - "given": "Y.", - "family": "Hayashizaki", - "affiliation": [] - } - ], - "member": "341", - "published-online": { - "date-parts": [ - [ - 2003, - 12, - 8 - ] - ] - }, - "container-title": "Proceedings of the National Academy of Sciences", - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1073/pnas.2136655100", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 16 - ] - ], - "date-time": "2017-06-16T03:00:41Z", - "timestamp": 1497582041000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2003, - 12, - 8 - ] - ] - }, - "references-count": 20, - "URL": "https://doi.org/10.1073/pnas.2136655100", - "relation": { - "cites": [] - }, - "subject": [ - "Multidisciplinary" - ], - "container-title-short": "Proceedings of the National Academy of Sciences", - "id": "9XBPQ8b" - }, - "citation_id": "9XBPQ8b" - }, - "doi:10.1080/17460441.2016.1201262": { - "source": "doi", - "identifer": "10.1080/17460441.2016.1201262", - "standard_citation": "doi:10.1080/17460441.2016.1201262", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 21 - ] - ], - "date-time": "2017-09-21T20:42:15Z", - "timestamp": 1506026535010 - }, - "reference-count": 105, - "publisher": "Informa UK Limited", - "issue": "8", - "content-domain": { - "domain": [ - "www.tandfonline.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2016, - 8, - 2 - ] - ] - }, - "DOI": "10.1080/17460441.2016.1201262", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 6, - 13 - ] - ], - "date-time": "2016-06-13T12:53:50Z", - "timestamp": 1465822430000 - }, - "page": "785-795", - "update-policy": "http://dx.doi.org/10.1080/tandf_crossmark_01", - "source": "Crossref", - "is-referenced-by-count": 12, - "title": "A renaissance of neural networks in drug discovery", - "prefix": "10.1080", - "volume": "11", - "author": [ - { - "given": "Igor I.", - "family": "Baskin", - "affiliation": [] - }, - { - "given": "David", - "family": "Winkler", - "affiliation": [] - }, - { - "given": "Igor V.", - "family": "Tetko", - "affiliation": [] - } - ], - "member": "301", - "published-online": { - "date-parts": [ - [ - 2016, - 7, - 4 - ] - ] - }, - "container-title": "Expert Opinion on Drug Discovery", - "original-title": [], - "link": [ - { - "URL": "http://www.tandfonline.com/doi/pdf/10.1080/17460441.2016.1201262", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 12, - 13 - ] - ], - "date-time": "2016-12-13T13:31:46Z", - "timestamp": 1481635906000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 7, - 4 - ] - ] - }, - "references-count": 105, - "alternative-id": [ - "10.1080/17460441.2016.1201262" - ], - "URL": "https://doi.org/10.1080/17460441.2016.1201262", - "relation": { - "cites": [] - }, - "subject": [ - "Drug Discovery" - ], - "container-title-short": "Expert Opinion on Drug Discovery", - "assertion": [ - { - "value": "The publishing and review policy for this title is described in its Aims & Scope.", - "order": 1, - "name": "peerreview_statement", - "label": "Peer Review Statement" - }, - { - "value": "http://www.tandfonline.com/action/journalInformation?show=aimsScope&journalCode=iedc20", - "URL": "http://www.tandfonline.com/action/journalInformation?show=aimsScope&journalCode=iedc20", - "order": 2, - "name": "aims_and_scope_url", - "label": "Aim & Scope" - } - ], - "id": "xPkT1z7D" - }, - "citation_id": "xPkT1z7D" - }, - "doi:10.1083/jcb.201610026": { - "source": "doi", - "identifer": "10.1083/jcb.201610026", - "standard_citation": "doi:10.1083/jcb.201610026", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T11:02:15Z", - "timestamp": 1507114935025 - }, - "reference-count": 74, - "publisher": "Rockefeller University Press", - "issue": "1", - "license": [ - { - "URL": "http://www.rupress.org/terms/", - "start": { - "date-parts": [ - [ - 2016, - 12, - 9 - ] - ], - "date-time": "2016-12-09T00:00:00Z", - "timestamp": 1481241600000 - }, - "delay-in-days": 0, - "content-version": "vor" - }, - { - "URL": "https://creativecommons.org/licenses/by-nc-sa/4.0/", - "start": { - "date-parts": [ - [ - 2017, - 7, - 3 - ] - ], - "date-time": "2017-07-03T00:00:00Z", - "timestamp": 1499040000000 - }, - "delay-in-days": 206, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/https://doi.org/10.13039/501100000024", - "name": "Canadian Institutes of Health Research", - "doi-asserted-by": "publisher", - "award": [] - } - ], - "content-domain": { - "domain": [ - "rupress.org" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2017, - 1, - 2 - ] - ] - }, - "abstract": "With recent advances in high-throughput, automated microscopy, there has been an increased demand for effective computational strategies to analyze large-scale, image-based data. To this end, computer vision approaches have been applied to cell segmentation and feature extraction, whereas machine-learning approaches have been developed to aid in phenotypic classification and clustering of data acquired from biological images. Here, we provide an overview of the commonly used computer vision and machine-learning methods for generating and categorizing phenotypic profiles, highlighting the general biological utility of each approach.", - "DOI": "10.1083/jcb.201610026", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 12, - 9 - ] - ], - "date-time": "2016-12-09T15:16:33Z", - "timestamp": 1481296593000 - }, - "page": "65-71", - "update-policy": "http://dx.doi.org/10.1083/jcb.crossmarkpolicy", - "source": "Crossref", - "is-referenced-by-count": 8, - "title": "Machine learning and computer vision approaches for phenotypic profiling", - "prefix": "10.1083", - "volume": "216", - "author": [ - { - "ORCID": "http://orcid.org/0000-0002-4888-8729", - "authenticated-orcid": false, - "given": "Ben T.", - "family": "Grys", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0001-9206-8384", - "authenticated-orcid": false, - "given": "Dara S.", - "family": "Lo", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0001-6285-5854", - "authenticated-orcid": false, - "given": "Nil", - "family": "Sahin", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-6328-9492", - "authenticated-orcid": false, - "given": "Oren Z.", - "family": "Kraus", - "affiliation": [] - }, - { - "given": "Quaid", - "family": "Morris", - "affiliation": [] - }, - { - "given": "Charles", - "family": "Boone", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0001-6427-6493", - "authenticated-orcid": false, - "given": "Brenda J.", - "family": "Andrews", - "affiliation": [] - } - ], - "member": "291", - "published-online": { - "date-parts": [ - [ - 2016, - 12, - 9 - ] - ] - }, - "container-title": "The Journal of Cell Biology", - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1083/jcb.201610026", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "https://syndication.highwire.org/content/doi/10.1083/jcb.201610026", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 4, - 19 - ] - ], - "date-time": "2017-04-19T23:47:43Z", - "timestamp": 1492645663000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 12, - 9 - ] - ] - }, - "references-count": 74, - "alternative-id": [ - "10.1083/jcb.201610026" - ], - "URL": "https://doi.org/10.1083/jcb.201610026", - "relation": { - "cites": [] - }, - "subject": [ - "Cell Biology" - ], - "container-title-short": "J. Cell Biol.", - "id": "McjXFLLq" - }, - "citation_id": "McjXFLLq" - }, - "doi:10.1086/512821": { - "source": "doi", - "identifer": "10.1086/512821", - "standard_citation": "doi:10.1086/512821", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T09:02:22Z", - "timestamp": 1507107742662 - }, - "reference-count": 21, - "publisher": "Elsevier BV", - "issue": "4", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2007, - 4, - 1 - ] - ], - "date-time": "2007-04-01T00:00:00Z", - "timestamp": 1175385600000 - }, - "delay-in-days": 0, - "content-version": "tdm" - }, - { - "URL": "http://www.elsevier.com/open-access/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2013, - 7, - 19 - ] - ], - "date-time": "2013-07-19T00:00:00Z", - "timestamp": 1374192000000 - }, - "delay-in-days": 2301, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2007, - 4 - ] - ] - }, - "DOI": "10.1086/512821", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2007, - 3, - 13 - ] - ], - "date-time": "2007-03-13T21:22:28Z", - "timestamp": 1173820948000 - }, - "page": "605-615", - "source": "Crossref", - "is-referenced-by-count": 197, - "title": "Overcoming the Winner’s Curse: Estimating Penetrance Parameters from Case-Control Data", - "prefix": "10.1016", - "volume": "80", - "author": [ - { - "given": "Sebastian", - "family": "Zöllner", - "affiliation": [] - }, - { - "given": "Jonathan K.", - "family": "Pritchard", - "affiliation": [] - } - ], - "member": "78", - "container-title": "The American Journal of Human Genetics", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S0002929707610970?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S0002929707610970?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 17 - ] - ], - "date-time": "2017-06-17T15:21:12Z", - "timestamp": 1497712872000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2007, - 4 - ] - ] - }, - "references-count": 21, - "alternative-id": [ - "S0002929707610970" - ], - "URL": "https://doi.org/10.1086/512821", - "relation": {}, - "subject": [ - "Genetics(clinical)", - "Genetics" - ], - "container-title-short": "The American Journal of Human Genetics", - "id": "10shRODux" - }, - "citation_id": "10shRODux" - }, - "doi:10.1093/bib/4.1.22": { - "source": "doi", - "identifer": "10.1093/bib/4.1.22", - "standard_citation": "doi:10.1093/bib/4.1.22", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 23 - ] - ], - "date-time": "2017-08-23T00:40:21Z", - "timestamp": 1503448821085 - }, - "reference-count": 0, - "publisher": "Oxford University Press (OUP)", - "issue": "1", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2003, - 1, - 1 - ] - ] - }, - "DOI": "10.1093/bib/4.1.22", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2006, - 1, - 18 - ] - ], - "date-time": "2006-01-18T19:42:59Z", - "timestamp": 1137613379000 - }, - "page": "22-30", - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "The state of the art of mammalian promoter recognition", - "prefix": "10.1093", - "volume": "4", - "author": [ - { - "given": "T.", - "family": "Werner", - "affiliation": [] - } - ], - "member": "286", - "container-title": "Briefings in Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/bib/article-pdf/4/1/22/829436/22.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 22 - ] - ], - "date-time": "2017-08-22T23:53:21Z", - "timestamp": 1503446001000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2003, - 1, - 1 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1093/bib/4.1.22", - "relation": {}, - "subject": [ - "Molecular Biology", - "Information Systems" - ], - "container-title-short": "Briefings in Bioinformatics", - "id": "19jjiGHWc" - }, - "citation_id": "19jjiGHWc" - }, - "doi:10.1093/bib/bbv020": { - "source": "doi", - "identifer": "10.1093/bib/bbv020", - "standard_citation": "doi:10.1093/bib/bbv020", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 2 - ] - ], - "date-time": "2017-10-02T15:22:13Z", - "timestamp": 1506957733511 - }, - "reference-count": 101, - "publisher": "Oxford University Press (OUP)", - "issue": "1", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 1 - ] - ] - }, - "DOI": "10.1093/bib/bbv020", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 4, - 2 - ] - ], - "date-time": "2015-04-02T16:51:54Z", - "timestamp": 1427993514000 - }, - "page": "2-12", - "source": "Crossref", - "is-referenced-by-count": 41, - "title": "A survey of current trends in computational drug repositioning", - "prefix": "10.1093", - "volume": "17", - "author": [ - { - "given": "Jiao", - "family": "Li", - "affiliation": [] - }, - { - "given": "Si", - "family": "Zheng", - "affiliation": [] - }, - { - "given": "Bin", - "family": "Chen", - "affiliation": [] - }, - { - "given": "Atul J.", - "family": "Butte", - "affiliation": [] - }, - { - "given": "S. Joshua", - "family": "Swamidass", - "affiliation": [] - }, - { - "given": "Zhiyong", - "family": "Lu", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2015, - 3, - 31 - ] - ] - }, - "container-title": "Briefings in Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/bib/article-pdf/17/1/2/6684680/bbv020.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 24 - ] - ], - "date-time": "2017-08-24T09:49:59Z", - "timestamp": 1503568199000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 3, - 31 - ] - ] - }, - "references-count": 101, - "alternative-id": [ - "10.1093/bib/bbv020" - ], - "URL": "https://doi.org/10.1093/bib/bbv020", - "relation": {}, - "subject": [ - "Molecular Biology", - "Information Systems" - ], - "container-title-short": "Brief Bioinform", - "id": "gTwjIQqB" - }, - "citation_id": "gTwjIQqB" - }, - "doi:10.1093/bib/bbw068": { - "source": "doi", - "identifer": "10.1093/bib/bbw068", - "standard_citation": "doi:10.1093/bib/bbw068", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T04:22:16Z", - "timestamp": 1507004536562 - }, - "reference-count": 225, - "publisher": "Oxford University Press (OUP)", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1093/bib/bbw068", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 7, - 30 - ] - ], - "date-time": "2016-07-30T01:10:49Z", - "timestamp": 1469841049000 - }, - "page": "bbw068", - "source": "Crossref", - "is-referenced-by-count": 14, - "title": "Deep learning in bioinformatics", - "prefix": "10.1093", - "author": [ - { - "given": "Seonwoo", - "family": "Min", - "affiliation": [] - }, - { - "given": "Byunghan", - "family": "Lee", - "affiliation": [] - }, - { - "given": "Sungroh", - "family": "Yoon", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2016, - 7, - 29 - ] - ] - }, - "container-title": "Briefings in Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/bib/article-pdf/doi/10.1093/bib/bbw068/9522577/bbw068.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 25 - ] - ], - "date-time": "2017-08-25T02:30:29Z", - "timestamp": 1503628229000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 7, - 29 - ] - ] - }, - "references-count": 225, - "alternative-id": [ - "10.1093/bib/bbw068" - ], - "URL": "https://doi.org/10.1093/bib/bbw068", - "relation": {}, - "subject": [ - "Molecular Biology", - "Information Systems" - ], - "container-title-short": "Brief Bioinform", - "id": "G00xvi94" - }, - "citation_id": "G00xvi94" - }, - "doi:10.1093/bib/bbw110": { - "source": "doi", - "identifer": "10.1093/bib/bbw110", - "standard_citation": "doi:10.1093/bib/bbw110", - "citeproc": { - "type": "article-journal", - "title": "A review of validation strategies for computational drug repositioning", - "container-title": "Briefings in Bioinformatics", - "source": "academic.oup.com", - "URL": "https://academic.oup.com/bib/article/doi/10.1093/bib/bbw110/2562646/A-review-of-validation-strategies-for", - "DOI": "10.1093/bib/bbw110", - "journalAbbreviation": "Brief Bioinform", - "author": [ - { - "family": "Brown", - "given": "Adam S." - }, - { - "family": "Patel", - "given": "Chirag J." - } - ], - "issued": { - "date-parts": [ - [ - "2016", - 11, - 22 - ] - ] - }, - "accessed": { - "date-parts": [ - [ - "2017", - 5, - 23 - ] - ] - }, - "id": "ir7ElHha" - }, - "citation_id": "ir7ElHha" - }, - "doi:10.1093/bib/bbw112": { - "source": "doi", - "identifer": "10.1093/bib/bbw112", - "standard_citation": "doi:10.1093/bib/bbw112", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 18 - ] - ], - "date-time": "2017-09-18T12:22:59Z", - "timestamp": 1505737379662 - }, - "reference-count": 109, - "publisher": "Oxford University Press (OUP)", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1093/bib/bbw112", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2017, - 1, - 3 - ] - ], - "date-time": "2017-01-03T20:05:50Z", - "timestamp": 1483473950000 - }, - "page": "bbw112", - "source": "Crossref", - "is-referenced-by-count": 1, - "title": "A review of connectivity map and computational approaches in pharmacogenomics", - "prefix": "10.1093", - "author": [ - { - "given": "Aliyu", - "family": "Musa", - "affiliation": [] - }, - { - "given": "Laleh Soltan", - "family": "Ghoraie", - "affiliation": [] - }, - { - "given": "Shu-Dong", - "family": "Zhang", - "affiliation": [] - }, - { - "given": "Galina", - "family": "Galzko", - "affiliation": [] - }, - { - "given": "Olli", - "family": "Yli-Harja", - "affiliation": [] - }, - { - "given": "Matthias", - "family": "Dehmer", - "affiliation": [] - }, - { - "given": "Benjamin", - "family": "Haibe-Kains", - "affiliation": [] - }, - { - "given": "Frank", - "family": "Emmert-Streib", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2017, - 1, - 9 - ] - ] - }, - "container-title": "Briefings in Bioinformatics", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T07:11:02Z", - "timestamp": 1498374662000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 1, - 9 - ] - ] - }, - "references-count": 109, - "alternative-id": [ - "10.1093/bib/bbw112" - ], - "URL": "https://doi.org/10.1093/bib/bbw112", - "relation": {}, - "subject": [ - "Molecular Biology", - "Information Systems" - ], - "container-title-short": "Brief Bioinform", - "id": "1BkEtNVsj" - }, - "citation_id": "1BkEtNVsj" - }, - "doi:10.1093/bib/bbx044": { - "source": "doi", - "identifer": "10.1093/bib/bbx044", - "standard_citation": "doi:10.1093/bib/bbx044", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 25 - ] - ], - "date-time": "2017-09-25T08:42:18Z", - "timestamp": 1506328938402 - }, - "reference-count": 119, - "publisher": "Oxford University Press (OUP)", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1093/bib/bbx044", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2017, - 4, - 5 - ] - ], - "date-time": "2017-04-05T19:38:50Z", - "timestamp": 1491421130000 - }, - "source": "Crossref", - "is-referenced-by-count": 3, - "title": "Deep learning for healthcare: review, opportunities and challenges", - "prefix": "10.1093", - "author": [ - { - "given": "Riccardo", - "family": "Miotto", - "affiliation": [] - }, - { - "given": "Fei", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Shuang", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Xiaoqian", - "family": "Jiang", - "affiliation": [] - }, - { - "given": "Joel T.", - "family": "Dudley", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2017, - 5, - 6 - ] - ] - }, - "container-title": "Briefings in Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/bib/article-pdf/doi/10.1093/bib/bbx044/16650575/bbx044.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 24 - ] - ], - "date-time": "2017-08-24T19:32:53Z", - "timestamp": 1503603173000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 5, - 6 - ] - ] - }, - "references-count": 119, - "URL": "https://doi.org/10.1093/bib/bbx044", - "relation": {}, - "subject": [ - "Molecular Biology", - "Information Systems" - ], - "id": "11I7bLcP3" - }, - "citation_id": "11I7bLcP3" - }, - "doi:10.1093/bioinformatics/bts475": { - "source": "doi", - "identifer": "10.1093/bioinformatics/bts475", - "standard_citation": "doi:10.1093/bioinformatics/bts475", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 27 - ] - ], - "date-time": "2017-09-27T20:42:17Z", - "timestamp": 1506544937251 - }, - "reference-count": 32, - "publisher": "Oxford University Press (OUP)", - "issue": "19", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2012, - 10, - 1 - ] - ] - }, - "DOI": "10.1093/bioinformatics/bts475", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2012, - 7, - 31 - ] - ], - "date-time": "2012-07-31T06:55:55Z", - "timestamp": 1343717755000 - }, - "page": "2449-2457", - "source": "Crossref", - "is-referenced-by-count": 76, - "title": "Deep architectures for protein contact map prediction", - "prefix": "10.1093", - "volume": "28", - "author": [ - { - "given": "Pietro", - "family": "Di Lena", - "affiliation": [] - }, - { - "given": "Ken", - "family": "Nagata", - "affiliation": [] - }, - { - "given": "Pierre", - "family": "Baldi", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2012, - 7, - 30 - ] - ] - }, - "container-title": "Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/bioinformatics/article-pdf/28/19/2449/13834227/bts475.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 22 - ] - ], - "date-time": "2017-08-22T20:50:31Z", - "timestamp": 1503435031000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2012, - 7, - 30 - ] - ] - }, - "references-count": 32, - "URL": "https://doi.org/10.1093/bioinformatics/bts475", - "relation": {}, - "subject": [ - "Statistics and Probability", - "Computational Theory and Mathematics", - "Biochemistry", - "Molecular Biology", - "Computational Mathematics", - "Computer Science Applications" - ], - "id": "xdoT1yUx" - }, - "citation_id": "xdoT1yUx" - }, - "doi:10.1093/bioinformatics/bts598": { - "source": "doi", - "identifer": "10.1093/bioinformatics/bts598", - "standard_citation": "doi:10.1093/bioinformatics/bts598", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 28 - ] - ], - "date-time": "2017-09-28T06:42:11Z", - "timestamp": 1506580931740 - }, - "reference-count": 43, - "publisher": "Oxford University Press (OUP)", - "issue": "23", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2012, - 12 - ] - ] - }, - "DOI": "10.1093/bioinformatics/bts598", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2012, - 10, - 10 - ] - ], - "date-time": "2012-10-10T08:45:27Z", - "timestamp": 1349858727000 - }, - "page": "3066-3072", - "source": "Crossref", - "is-referenced-by-count": 49, - "title": "Predicting protein residue–residue contacts using deep networks and boosting", - "prefix": "10.1093", - "volume": "28", - "author": [ - { - "given": "Jesse", - "family": "Eickholt", - "affiliation": [] - }, - { - "given": "Jianlin", - "family": "Cheng", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2012, - 10, - 9 - ] - ] - }, - "container-title": "Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/bioinformatics/article-pdf/28/23/3066/18530766/bts598.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 22 - ] - ], - "date-time": "2017-08-22T15:50:14Z", - "timestamp": 1503417014000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2012, - 10, - 9 - ] - ] - }, - "references-count": 43, - "URL": "https://doi.org/10.1093/bioinformatics/bts598", - "relation": {}, - "subject": [ - "Statistics and Probability", - "Computational Theory and Mathematics", - "Biochemistry", - "Molecular Biology", - "Computational Mathematics", - "Computer Science Applications" - ], - "id": "18bNbDNlc" - }, - "citation_id": "18bNbDNlc" - }, - "doi:10.1093/bioinformatics/btt234": { - "source": "doi", - "identifer": "10.1093/bioinformatics/btt234", - "standard_citation": "doi:10.1093/bioinformatics/btt234", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 22 - ] - ], - "date-time": "2017-08-22T16:41:49Z", - "timestamp": 1503420109300 - }, - "reference-count": 53, - "publisher": "Oxford University Press (OUP)", - "issue": "13", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2013, - 7 - ] - ] - }, - "DOI": "10.1093/bioinformatics/btt234", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2013, - 6, - 27 - ] - ], - "date-time": "2013-06-27T05:33:26Z", - "timestamp": 1372311206000 - }, - "page": "i126-i134", - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Predicting drug-target interactions using restricted Boltzmann machines", - "prefix": "10.1093", - "volume": "29", - "author": [ - { - "given": "Yuhao", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Jianyang", - "family": "Zeng", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2013, - 6, - 19 - ] - ] - }, - "container-title": "Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/bioinformatics/article-pdf/29/13/i126/18535773/btt234.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 22 - ] - ], - "date-time": "2017-08-22T15:58:57Z", - "timestamp": 1503417537000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2013, - 6, - 19 - ] - ] - }, - "references-count": 53, - "URL": "https://doi.org/10.1093/bioinformatics/btt234", - "relation": {}, - "subject": [ - "Statistics and Probability", - "Computational Theory and Mathematics", - "Biochemistry", - "Molecular Biology", - "Computational Mathematics", - "Computer Science Applications" - ], - "id": "1AU7wzPqa" - }, - "citation_id": "1AU7wzPqa" - }, - "doi:10.1093/bioinformatics/btu703": { - "source": "doi", - "identifer": "10.1093/bioinformatics/btu703", - "standard_citation": "doi:10.1093/bioinformatics/btu703", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 2 - ] - ], - "date-time": "2017-10-02T05:02:31Z", - "timestamp": 1506920551584 - }, - "reference-count": 7, - "publisher": "Oxford University Press (OUP)", - "issue": "5", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015, - 3, - 1 - ] - ] - }, - "DOI": "10.1093/bioinformatics/btu703", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 10, - 23 - ] - ], - "date-time": "2014-10-23T00:08:46Z", - "timestamp": 1414022926000 - }, - "page": "761-763", - "source": "Crossref", - "is-referenced-by-count": 47, - "title": "DANN: a deep learning approach for annotating the pathogenicity of genetic variants", - "prefix": "10.1093", - "volume": "31", - "author": [ - { - "given": "Daniel", - "family": "Quang", - "affiliation": [] - }, - { - "given": "Yifei", - "family": "Chen", - "affiliation": [] - }, - { - "given": "Xiaohui", - "family": "Xie", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2014, - 10, - 22 - ] - ] - }, - "container-title": "Bioinformatics", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 22 - ] - ], - "date-time": "2017-06-22T22:54:07Z", - "timestamp": 1498172047000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 10, - 22 - ] - ] - }, - "references-count": 7, - "URL": "https://doi.org/10.1093/bioinformatics/btu703", - "relation": {}, - "subject": [ - "Statistics and Probability", - "Computational Theory and Mathematics", - "Biochemistry", - "Molecular Biology", - "Computational Mathematics", - "Computer Science Applications" - ], - "id": "15E5yG1Ho" - }, - "citation_id": "15E5yG1Ho" - }, - "doi:10.1093/bioinformatics/btu791": { - "source": "doi", - "identifer": "10.1093/bioinformatics/btu791", - "standard_citation": "doi:10.1093/bioinformatics/btu791", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 29 - ] - ], - "date-time": "2017-09-29T17:02:14Z", - "timestamp": 1506704534019 - }, - "reference-count": 29, - "publisher": "Oxford University Press (OUP)", - "issue": "7", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015, - 4, - 1 - ] - ] - }, - "DOI": "10.1093/bioinformatics/btu791", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 11, - 28 - ] - ], - "date-time": "2014-11-28T15:35:07Z", - "timestamp": 1417188907000 - }, - "page": "999-1006", - "source": "Crossref", - "is-referenced-by-count": 62, - "title": "MetaPSICOV: combining coevolution methods for accurate prediction of contacts and long range hydrogen bonding in proteins", - "prefix": "10.1093", - "volume": "31", - "author": [ - { - "given": "David T.", - "family": "Jones", - "affiliation": [] - }, - { - "given": "Tanya", - "family": "Singh", - "affiliation": [] - }, - { - "given": "Tomasz", - "family": "Kosciolek", - "affiliation": [] - }, - { - "given": "Stuart", - "family": "Tetchner", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2014, - 11, - 26 - ] - ] - }, - "container-title": "Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/bioinformatics/article-pdf/31/7/999/17125423/btu791.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 22 - ] - ], - "date-time": "2017-08-22T16:23:54Z", - "timestamp": 1503419034000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 11, - 26 - ] - ] - }, - "references-count": 29, - "URL": "https://doi.org/10.1093/bioinformatics/btu791", - "relation": {}, - "subject": [ - "Statistics and Probability", - "Computational Theory and Mathematics", - "Biochemistry", - "Molecular Biology", - "Computational Mathematics", - "Computer Science Applications" - ], - "id": "7atXz0r" - }, - "citation_id": "7atXz0r" - }, - "doi:10.1093/bioinformatics/btv472": { - "source": "doi", - "identifer": "10.1093/bioinformatics/btv472", - "standard_citation": "doi:10.1093/bioinformatics/btv472", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 27 - ] - ], - "date-time": "2017-09-27T20:42:17Z", - "timestamp": 1506544937251 - }, - "reference-count": 44, - "publisher": "Oxford University Press (OUP)", - "issue": "21", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015, - 11, - 1 - ] - ] - }, - "DOI": "10.1093/bioinformatics/btv472", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 8, - 15 - ] - ], - "date-time": "2015-08-15T01:29:49Z", - "timestamp": 1439602189000 - }, - "page": "3506-3513", - "source": "Crossref", - "is-referenced-by-count": 18, - "title": "Protein contact prediction by integrating joint evolutionary coupling analysis and supervised learning", - "prefix": "10.1093", - "volume": "31", - "author": [ - { - "given": "Jianzhu", - "family": "Ma", - "affiliation": [] - }, - { - "given": "Sheng", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Zhiyong", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Jinbo", - "family": "Xu", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2015, - 8, - 14 - ] - ] - }, - "container-title": "Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/bioinformatics/article-pdf/31/21/3506/17122523/btv472.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 22 - ] - ], - "date-time": "2017-08-22T15:50:13Z", - "timestamp": 1503417013000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 8, - 14 - ] - ] - }, - "references-count": 44, - "alternative-id": [ - "10.1093/bioinformatics/btv472" - ], - "URL": "https://doi.org/10.1093/bioinformatics/btv472", - "relation": {}, - "subject": [ - "Statistics and Probability", - "Computational Theory and Mathematics", - "Biochemistry", - "Molecular Biology", - "Computational Mathematics", - "Computer Science Applications" - ], - "container-title-short": "Bioinformatics", - "id": "kqjqFesT" - }, - "citation_id": "kqjqFesT" - }, - "doi:10.1093/ije/dyn022": { - "source": "doi", - "identifer": "10.1093/ije/dyn022", - "standard_citation": "doi:10.1093/ije/dyn022", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 23 - ] - ], - "date-time": "2017-08-23T04:12:07Z", - "timestamp": 1503461527681 - }, - "reference-count": 22, - "publisher": "Oxford University Press (OUP)", - "issue": "3", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2008, - 6, - 1 - ] - ] - }, - "DOI": "10.1093/ije/dyn022", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2008, - 2, - 28 - ] - ], - "date-time": "2008-02-28T01:24:04Z", - "timestamp": 1204161844000 - }, - "page": "678-682", - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Epidemiological research labelled as a violation of privacy: the case of Estonia", - "prefix": "10.1093", - "volume": "37", - "author": [ - { - "given": "M.", - "family": "Rahu", - "affiliation": [] - }, - { - "given": "M.", - "family": "McKee", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2008, - 2, - 26 - ] - ] - }, - "container-title": "International Journal of Epidemiology", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/ije/article-pdf/37/3/678/2230349/dyn022.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 23 - ] - ], - "date-time": "2017-08-23T03:34:23Z", - "timestamp": 1503459263000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2008, - 2, - 26 - ] - ] - }, - "references-count": 22, - "URL": "https://doi.org/10.1093/ije/dyn022", - "relation": {}, - "subject": [ - "Epidemiology", - "General Medicine" - ], - "container-title-short": "International Journal of Epidemiology", - "id": "CVnO5njl" - }, - "citation_id": "CVnO5njl" - }, - "doi:10.1093/ije/dyu188": { - "source": "doi", - "identifer": "10.1093/ije/dyu188", - "standard_citation": "doi:10.1093/ije/dyu188", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 31 - ] - ], - "date-time": "2017-08-31T15:44:02Z", - "timestamp": 1504194242850 - }, - "reference-count": 56, - "publisher": "Oxford University Press (OUP)", - "issue": "6", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2014, - 12 - ] - ] - }, - "DOI": "10.1093/ije/dyu188", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 9, - 28 - ] - ], - "date-time": "2014-09-28T00:27:43Z", - "timestamp": 1411864063000 - }, - "page": "1929-1944", - "source": "Crossref", - "is-referenced-by-count": 24, - "title": "DataSHIELD: taking the analysis to the data, not the data to the analysis", - "prefix": "10.1093", - "volume": "43", - "author": [ - { - "given": "Amadou", - "family": "Gaye", - "affiliation": [] - }, - { - "given": "Yannick", - "family": "Marcon", - "affiliation": [] - }, - { - "given": "Julia", - "family": "Isaeva", - "affiliation": [] - }, - { - "given": "Philippe", - "family": "LaFlamme", - "affiliation": [] - }, - { - "given": "Andrew", - "family": "Turner", - "affiliation": [] - }, - { - "given": "Elinor M", - "family": "Jones", - "affiliation": [] - }, - { - "given": "Joel", - "family": "Minion", - "affiliation": [] - }, - { - "given": "Andrew W", - "family": "Boyd", - "affiliation": [] - }, - { - "given": "Christopher J", - "family": "Newby", - "affiliation": [] - }, - { - "given": "Marja-Liisa", - "family": "Nuotio", - "affiliation": [] - }, - { - "given": "Rebecca", - "family": "Wilson", - "affiliation": [] - }, - { - "given": "Oliver", - "family": "Butters", - "affiliation": [] - }, - { - "given": "Barnaby", - "family": "Murtagh", - "affiliation": [] - }, - { - "given": "Ipek", - "family": "Demir", - "affiliation": [] - }, - { - "given": "Dany", - "family": "Doiron", - "affiliation": [] - }, - { - "given": "Lisette", - "family": "Giepmans", - "affiliation": [] - }, - { - "given": "Susan E", - "family": "Wallace", - "affiliation": [] - }, - { - "given": "Isabelle", - "family": "Budin-Ljøsne", - "affiliation": [] - }, - { - "given": "Carsten", - "family": "Oliver Schmidt", - "affiliation": [] - }, - { - "given": "Paolo", - "family": "Boffetta", - "affiliation": [] - }, - { - "given": "Mathieu", - "family": "Boniol", - "affiliation": [] - }, - { - "given": "Maria", - "family": "Bota", - "affiliation": [] - }, - { - "given": "Kim W", - "family": "Carter", - "affiliation": [] - }, - { - "given": "Nick", - "family": "deKlerk", - "affiliation": [] - }, - { - "given": "Chris", - "family": "Dibben", - "affiliation": [] - }, - { - "given": "Richard W", - "family": "Francis", - "affiliation": [] - }, - { - "given": "Tero", - "family": "Hiekkalinna", - "affiliation": [] - }, - { - "given": "Kristian", - "family": "Hveem", - "affiliation": [] - }, - { - "given": "Kirsti", - "family": "Kvaløy", - "affiliation": [] - }, - { - "given": "Sean", - "family": "Millar", - "affiliation": [] - }, - { - "given": "Ivan J", - "family": "Perry", - "affiliation": [] - }, - { - "given": "Annette", - "family": "Peters", - "affiliation": [] - }, - { - "given": "Catherine M", - "family": "Phillips", - "affiliation": [] - }, - { - "given": "Frank", - "family": "Popham", - "affiliation": [] - }, - { - "given": "Gillian", - "family": "Raab", - "affiliation": [] - }, - { - "given": "Eva", - "family": "Reischl", - "affiliation": [] - }, - { - "given": "Nuala", - "family": "Sheehan", - "affiliation": [] - }, - { - "given": "Melanie", - "family": "Waldenberger", - "affiliation": [] - }, - { - "given": "Markus", - "family": "Perola", - "affiliation": [] - }, - { - "given": "Edwin", - "family": "van den Heuvel", - "affiliation": [] - }, - { - "given": "John", - "family": "Macleod", - "affiliation": [] - }, - { - "given": "Bartha M", - "family": "Knoppers", - "affiliation": [] - }, - { - "given": "Ronald P", - "family": "Stolk", - "affiliation": [] - }, - { - "given": "Isabel", - "family": "Fortier", - "affiliation": [] - }, - { - "given": "Jennifer R", - "family": "Harris", - "affiliation": [] - }, - { - "given": "Bruce HR", - "family": "Woffenbuttel", - "affiliation": [] - }, - { - "given": "Madeleine J", - "family": "Murtagh", - "affiliation": [] - }, - { - "given": "Vincent", - "family": "Ferretti", - "affiliation": [] - }, - { - "given": "Paul R", - "family": "Burton", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2014, - 9, - 27 - ] - ] - }, - "container-title": "International Journal of Epidemiology", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/ije/article-pdf/43/6/1929/18482399/dyu188.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 23 - ] - ], - "date-time": "2017-08-23T10:31:23Z", - "timestamp": 1503484283000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 9, - 27 - ] - ] - }, - "references-count": 56, - "URL": "https://doi.org/10.1093/ije/dyu188", - "relation": {}, - "subject": [ - "Epidemiology", - "General Medicine" - ], - "id": "SfxIiPJ1" - }, - "citation_id": "SfxIiPJ1" - }, - "doi:10.1093/ije/dyv193": { - "source": "doi", - "identifer": "10.1093/ije/dyv193", - "standard_citation": "doi:10.1093/ije/dyv193", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 28 - ] - ], - "date-time": "2017-08-28T00:11:21Z", - "timestamp": 1503879081951 - }, - "reference-count": 15, - "publisher": "Oxford University Press (OUP)", - "issue": "2", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 4 - ] - ] - }, - "DOI": "10.1093/ije/dyv193", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 5, - 12 - ] - ], - "date-time": "2016-05-12T14:58:17Z", - "timestamp": 1463065097000 - }, - "page": "408-416", - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "ViPAR: a software platform for the Virtual Pooling and Analysis of Research Data", - "prefix": "10.1093", - "volume": "45", - "author": [ - { - "given": "Kim W", - "family": "Carter", - "affiliation": [] - }, - { - "given": "KW", - "family": "Carter", - "affiliation": [] - }, - { - "given": "RW", - "family": "Francis", - "affiliation": [] - }, - { - "given": "M", - "family": "Bresnahan", - "affiliation": [] - }, - { - "given": "M", - "family": "Gissler", - "affiliation": [] - }, - { - "given": "TK", - "family": "Grønborg", - "affiliation": [] - }, - { - "given": "R", - "family": "Gross", - "affiliation": [] - }, - { - "given": "N", - "family": "Gunnes", - "affiliation": [] - }, - { - "given": "G", - "family": "Hammond", - "affiliation": [] - }, - { - "given": "M", - "family": "Hornig", - "affiliation": [] - }, - { - "given": "CM", - "family": "Hultman", - "affiliation": [] - }, - { - "given": "J", - "family": "Huttunen", - "affiliation": [] - }, - { - "given": "A", - "family": "Langridge", - "affiliation": [] - }, - { - "given": "H", - "family": "Leonard", - "affiliation": [] - }, - { - "given": "S", - "family": "Newman", - "affiliation": [] - }, - { - "given": "ET", - "family": "Parner", - "affiliation": [] - }, - { - "given": "G", - "family": "Petersson", - "affiliation": [] - }, - { - "given": "A", - "family": "Reichenberg", - "affiliation": [] - }, - { - "given": "S", - "family": "Sandin", - "affiliation": [] - }, - { - "given": "DE", - "family": "Schendel", - "affiliation": [] - }, - { - "given": "L", - "family": "Schalkwyk", - "affiliation": [] - }, - { - "given": "A", - "family": "Sourander", - "affiliation": [] - }, - { - "given": "C", - "family": "Steadman", - "affiliation": [] - }, - { - "given": "C", - "family": "Stoltenberg", - "affiliation": [] - }, - { - "given": "A", - "family": "Suominen", - "affiliation": [] - }, - { - "given": "P", - "family": "Surén", - "affiliation": [] - }, - { - "given": "E", - "family": "Susser", - "affiliation": [] - }, - { - "given": "A", - "family": "Sylvester Vethanayagam", - "affiliation": [] - }, - { - "given": "Z", - "family": "Yusof", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2015, - 10, - 8 - ] - ] - }, - "container-title": "International Journal of Epidemiology", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/ije/article-pdf/45/2/408/14113104/dyv193.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 25 - ] - ], - "date-time": "2017-08-25T04:41:09Z", - "timestamp": 1503636069000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 10, - 8 - ] - ] - }, - "references-count": 15, - "alternative-id": [ - "10.1093/ije/dyv193" - ], - "URL": "https://doi.org/10.1093/ije/dyv193", - "relation": {}, - "subject": [ - "Epidemiology", - "General Medicine" - ], - "container-title-short": "Int. J. Epidemiol.", - "id": "1D6b3tMu9" - }, - "citation_id": "1D6b3tMu9" - }, - "doi:10.1093/jamia/ocv111": { - "source": "doi", - "identifer": "10.1093/jamia/ocv111", - "standard_citation": "doi:10.1093/jamia/ocv111", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 28 - ] - ], - "date-time": "2017-08-28T00:11:18Z", - "timestamp": 1503879078401 - }, - "reference-count": 28, - "publisher": "Oxford University Press (OUP)", - "issue": "2", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 3 - ] - ] - }, - "DOI": "10.1093/jamia/ocv111", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 2, - 6 - ] - ], - "date-time": "2016-02-06T03:04:19Z", - "timestamp": 1454727859000 - }, - "page": "413-419", - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Harnessing next-generation informatics for personalizing medicine: a report from AMIA’s 2014 Health Policy Invitational Meeting", - "prefix": "10.1093", - "volume": "23", - "author": [ - { - "given": "Laura K", - "family": "Wiley", - "affiliation": [] - }, - { - "given": "Peter", - "family": "Tarczy-Hornoch", - "affiliation": [] - }, - { - "given": "Joshua C", - "family": "Denny", - "affiliation": [] - }, - { - "given": "Robert R", - "family": "Freimuth", - "affiliation": [] - }, - { - "given": "Casey L", - "family": "Overby", - "affiliation": [] - }, - { - "given": "Nigam", - "family": "Shah", - "affiliation": [] - }, - { - "given": "Ross D", - "family": "Martin", - "affiliation": [] - }, - { - "given": "Indra Neil", - "family": "Sarkar", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2016, - 2, - 5 - ] - ] - }, - "container-title": "Journal of the American Medical Informatics Association", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/jamia/article-pdf/23/2/413/7920593/ocv111.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 25 - ] - ], - "date-time": "2017-08-25T04:40:49Z", - "timestamp": 1503636049000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 2, - 5 - ] - ] - }, - "references-count": 28, - "alternative-id": [ - "10.1093/jamia/ocv111" - ], - "URL": "https://doi.org/10.1093/jamia/ocv111", - "relation": {}, - "subject": [ - "Health Informatics" - ], - "container-title-short": "J Am Med Inform Assoc", - "id": "2cYqPKf1" - }, - "citation_id": "2cYqPKf1" - }, - "doi:10.1093/jamia/ocw011": { - "source": "doi", - "identifer": "10.1093/jamia/ocw011", - "standard_citation": "doi:10.1093/jamia/ocw011", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 19 - ] - ], - "date-time": "2017-09-19T21:22:13Z", - "timestamp": 1505856133612 - }, - "reference-count": 31, - "publisher": "Oxford University Press (OUP)", - "issue": "4", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 7 - ] - ] - }, - "DOI": "10.1093/jamia/ocw011", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 4, - 24 - ] - ], - "date-time": "2016-04-24T00:18:19Z", - "timestamp": 1461457099000 - }, - "page": "731-740", - "source": "Crossref", - "is-referenced-by-count": 5, - "title": "Electronic medical record phenotyping using the anchor and learn framework", - "prefix": "10.1093", - "volume": "23", - "author": [ - { - "given": "Yoni", - "family": "Halpern", - "affiliation": [] - }, - { - "given": "Steven", - "family": "Horng", - "affiliation": [] - }, - { - "given": "Youngduck", - "family": "Choi", - "affiliation": [] - }, - { - "given": "David", - "family": "Sontag", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2016, - 4, - 23 - ] - ] - }, - "container-title": "Journal of the American Medical Informatics Association", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/jamia/article-pdf/23/4/731/13708086/ocw011.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 24 - ] - ], - "date-time": "2017-08-24T09:15:42Z", - "timestamp": 1503566142000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 4, - 23 - ] - ] - }, - "references-count": 31, - "alternative-id": [ - "10.1093/jamia/ocw011" - ], - "URL": "https://doi.org/10.1093/jamia/ocw011", - "relation": {}, - "subject": [ - "Health Informatics" - ], - "container-title-short": "J Am Med Inform Assoc", - "id": "A9JeoGV8" - }, - "citation_id": "A9JeoGV8" - }, - "doi:10.1093/nar/gkq747": { - "source": "doi", - "identifer": "10.1093/nar/gkq747", - "standard_citation": "doi:10.1093/nar/gkq747", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 26 - ] - ], - "date-time": "2017-09-26T12:42:15Z", - "timestamp": 1506429735706 - }, - "reference-count": 44, - "publisher": "Oxford University Press (OUP)", - "issue": "20", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2010, - 11 - ] - ] - }, - "DOI": "10.1093/nar/gkq747", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2010, - 8, - 31 - ] - ], - "date-time": "2010-08-31T00:46:17Z", - "timestamp": 1283215577000 - }, - "page": "e191-e191", - "source": "Crossref", - "is-referenced-by-count": 211, - "title": "FragGeneScan: predicting genes in short and error-prone reads", - "prefix": "10.1093", - "volume": "38", - "author": [ - { - "given": "Mina", - "family": "Rho", - "affiliation": [] - }, - { - "given": "Haixu", - "family": "Tang", - "affiliation": [] - }, - { - "given": "Yuzhen", - "family": "Ye", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2010, - 8, - 28 - ] - ] - }, - "container-title": "Nucleic Acids Research", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/nar/article-pdf/38/20/e191/16772703/gkq747.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 23 - ] - ], - "date-time": "2017-08-23T21:48:30Z", - "timestamp": 1503524910000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2010, - 8, - 28 - ] - ] - }, - "references-count": 44, - "URL": "https://doi.org/10.1093/nar/gkq747", - "relation": {}, - "subject": [ - "Genetics" - ], - "id": "QlbXLqH" - }, - "citation_id": "QlbXLqH" - }, - "doi:10.1093/nar/gkr777": { - "source": "doi", - "identifer": "10.1093/nar/gkr777", - "standard_citation": "doi:10.1093/nar/gkr777", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 27 - ] - ], - "date-time": "2017-09-27T11:42:20Z", - "timestamp": 1506512540064 - }, - "reference-count": 32, - "publisher": "Oxford University Press (OUP)", - "issue": "D1", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2012, - 1, - 1 - ] - ] - }, - "DOI": "10.1093/nar/gkr777", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2011, - 9, - 24 - ] - ], - "date-time": "2011-09-24T00:21:22Z", - "timestamp": 1316823682000 - }, - "page": "D1100-D1107", - "source": "Crossref", - "is-referenced-by-count": 1021, - "title": "ChEMBL: a large-scale bioactivity database for drug discovery", - "prefix": "10.1093", - "volume": "40", - "author": [ - { - "given": "A.", - "family": "Gaulton", - "affiliation": [] - }, - { - "given": "L. J.", - "family": "Bellis", - "affiliation": [] - }, - { - "given": "A. P.", - "family": "Bento", - "affiliation": [] - }, - { - "given": "J.", - "family": "Chambers", - "affiliation": [] - }, - { - "given": "M.", - "family": "Davies", - "affiliation": [] - }, - { - "given": "A.", - "family": "Hersey", - "affiliation": [] - }, - { - "given": "Y.", - "family": "Light", - "affiliation": [] - }, - { - "given": "S.", - "family": "McGlinchey", - "affiliation": [] - }, - { - "given": "D.", - "family": "Michalovich", - "affiliation": [] - }, - { - "given": "B.", - "family": "Al-Lazikani", - "affiliation": [] - }, - { - "given": "J. P.", - "family": "Overington", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2011, - 9, - 23 - ] - ] - }, - "container-title": "Nucleic Acids Research", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 20 - ] - ], - "date-time": "2017-06-20T06:52:05Z", - "timestamp": 1497941525000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2011, - 9, - 23 - ] - ] - }, - "references-count": 32, - "URL": "https://doi.org/10.1093/nar/gkr777", - "relation": {}, - "subject": [ - "Genetics" - ], - "container-title-short": "Nucleic Acids Research", - "id": "x1nE5icc" - }, - "citation_id": "x1nE5icc" - }, - "doi:10.1093/nar/gku1058": { - "source": "doi", - "identifer": "10.1093/nar/gku1058", - "standard_citation": "doi:10.1093/nar/gku1058", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 21 - ] - ], - "date-time": "2017-08-21T17:42:51Z", - "timestamp": 1503337371381 - }, - "reference-count": 38, - "publisher": "Oxford University Press (OUP)", - "issue": "1", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015, - 1, - 9 - ] - ] - }, - "DOI": "10.1093/nar/gku1058", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 11, - 7 - ] - ], - "date-time": "2014-11-07T05:22:41Z", - "timestamp": 1415337761000 - }, - "page": "e6-e6", - "source": "Crossref", - "is-referenced-by-count": 20, - "title": "DEEP: a general computational framework for predicting enhancers", - "prefix": "10.1093", - "volume": "43", - "author": [ - { - "given": "D.", - "family": "Kleftogiannis", - "affiliation": [] - }, - { - "given": "P.", - "family": "Kalnis", - "affiliation": [] - }, - { - "given": "V. B.", - "family": "Bajic", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2014, - 11, - 5 - ] - ] - }, - "container-title": "Nucleic Acids Research", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 22 - ] - ], - "date-time": "2017-06-22T23:45:21Z", - "timestamp": 1498175121000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 11, - 5 - ] - ] - }, - "references-count": 38, - "URL": "https://doi.org/10.1093/nar/gku1058", - "relation": {}, - "subject": [ - "Genetics" - ], - "container-title-short": "Nucleic Acids Research", - "id": "12aqvAgz6" - }, - "citation_id": "12aqvAgz6" - }, - "doi:10.1101/036129": { - "source": "doi", - "identifer": "10.1101/036129", - "standard_citation": "doi:10.1101/036129", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 11 - ] - ], - "date-time": "2017-08-11T08:06:56Z", - "timestamp": 1502438816944 - }, - "posted": { - "date-parts": [ - [ - 2016, - 1, - 7 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2016, - 5, - 18 - ] - ] - }, - "abstract": "Transcriptional enhancers are non-coding segments of DNA that play a central role in the spatiotemporal regulation of gene expression programs. However, systematically and precisely predicting enhancers on a genome-wide scale remain a major challenge. Although existing methods have achieved some success in enhancer prediction, they still suffer from a limited number of training samples, a simplicity of features, class-imbalanced data, and inconsistent performance across diverse cell types/tissues. Here, we developed a deep learning-based algorithmic framework named PEDLA (https://github.com/wenjiegroup/PEDLA), which can directly learn an enhancer predictor from massively heterogeneous data and generalize in ways that are mostly consistent across various cell types/tissues. We first trained PEDLA with 1,114-dimensional heterogeneous features in H1 cells, and we demonstrated that our PEDLA framework integrates diverse heterogeneous features and gives state-of-the-art performance relative to five existing methods for enhancer prediction. We further extended PEDLA to continuously learn from 22 training cell types/tissues, and the results showed that PEDLA manifested superior performance consistency in both training and independent test sets. On average, PEDLA achieved 95.0% accuracy and a 96.8% geometric mean (GM) across 22 training cell types/tissues, as well as 95.7% accuracy and a 96.8% GM across 20 independent test cell types/tissues. Together, our work illustrates the power of harnessing state-of-the-art deep learning techniques to consistently identify regulatory elements at a genome-wide scale from massively heterogeneous data across diverse cell types/tissues.", - "DOI": "10.1101/036129", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2016, - 1, - 14 - ] - ], - "date-time": "2016-01-14T00:28:21Z", - "timestamp": 1452731301000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "PEDLA: predicting enhancers with a deep learning-based algorithmic framework", - "prefix": "10.1101", - "author": [ - { - "given": "Feng", - "family": "Liu", - "affiliation": [] - }, - { - "given": "Hao", - "family": "Li", - "affiliation": [] - }, - { - "given": "Chao", - "family": "Ren", - "affiliation": [] - }, - { - "given": "Xiaochen", - "family": "Bo", - "affiliation": [] - }, - { - "given": "Wenjie", - "family": "Shu", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/036129", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 4, - 12 - ] - ], - "date-time": "2017-04-12T17:56:26Z", - "timestamp": 1492019786000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 1, - 7 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/036129", - "relation": { - "is-preprint-of": [ - { - "id-type": "doi", - "id": "10.1038/srep28517", - "asserted-by": "subject" - } - ] - }, - "id": "s5sy4AOi" - }, - "citation_id": "s5sy4AOi" - }, - "doi:10.1101/041616": { - "source": "doi", - "identifer": "10.1101/041616", - "standard_citation": "doi:10.1101/041616", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T03:22:32Z", - "timestamp": 1502248952344 - }, - "posted": { - "date-parts": [ - [ - 2016, - 2, - 28 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2016, - 2, - 28 - ] - ] - }, - "abstract": "Identifying active cis-regulatory regions in the human genome is critical for understanding gene regulation and assessing the impact of genetic variation on phenotype. Based on rich data resources such as the Encyclopedia of DNA Elements (ENCODE) and the Functional Annotation of the Mammalian Genome (FANTOM) projects, we introduce DECRES, the first supervised deep learning approach for the identification of enhancer and promoter regions in the human genome. Due to their ability to discover patterns in large and complex data, the introduction of deep learning methods enables a significant advance in our knowledge of the genomic locations of cis-regulatory regions. Using models for well-characterized cell lines, we identify key experimental features that contribute to the predictive performance. Applying DECRES, we delineate locations of 300,000 candidate enhancers genome wide (6.8% of the genome, of which 40,000 are supported by bidirectional transcription data) and 26,000 candidate promoters (0.6% of the genome).", - "DOI": "10.1101/041616", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2016, - 2, - 29 - ] - ], - "date-time": "2016-02-29T06:07:41Z", - "timestamp": 1456726061000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Genome-Wide Prediction of cis-Regulatory Regions Using Supervised Deep Learning Methods", - "prefix": "10.1101", - "author": [ - { - "given": "Yifeng", - "family": "Li", - "affiliation": [] - }, - { - "given": "Wenqiang", - "family": "Shi", - "affiliation": [] - }, - { - "given": "Wyeth W", - "family": "Wasserman", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/041616", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 1, - 4 - ] - ], - "date-time": "2017-01-04T06:12:58Z", - "timestamp": 1483510378000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 2, - 28 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/041616", - "relation": {}, - "id": "1HbQQcY2q" - }, - "citation_id": "1HbQQcY2q" - }, - "doi:10.1101/054775": { - "source": "doi", - "identifer": "10.1101/054775", - "standard_citation": "doi:10.1101/054775", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T10:50:22Z", - "timestamp": 1502275822366 - }, - "posted": { - "date-parts": [ - [ - 2016, - 5, - 22 - ] - ] - }, - "group-title": "Immunology", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2016, - 6, - 7 - ] - ] - }, - "abstract": "Predicting the binding affinity between MHC proteins and their peptide ligands is a key problem in computational immunology. State of the art performance is currently achieved by the allele-specific predictor NetMHC and the pan-allele predictor NetMHCpan, both of which are ensembles of shallow neural networks. We explore an intermediate between allele-specific and pan-allele prediction: training allele-specific predictors with synthetic samples generated by imputation of the peptide-MHC affinity matrix. We find that the imputation strategy is useful on alleles with very little training data. We have implemented our predictor as an open-source software package called MHCflurry and show that MHCflurry achieves competitive performance to NetMHC and NetMHCpan.", - "DOI": "10.1101/054775", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2016, - 5, - 23 - ] - ], - "date-time": "2016-05-23T17:23:01Z", - "timestamp": 1464024181000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Predicting Peptide-MHC Binding Affinities With Imputed Training Data", - "prefix": "10.1101", - "author": [ - { - "ORCID": "http://orcid.org/0000-0003-2839-2870", - "authenticated-orcid": false, - "given": "Alex", - "family": "Rubinsteyn", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-9949-069X", - "authenticated-orcid": false, - "given": "Timothy", - "family": "O'Donnell", - "affiliation": [] - }, - { - "given": "Nandita", - "family": "Damaraju", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0001-6596-8563", - "authenticated-orcid": false, - "given": "Jeffrey", - "family": "Hammerbacher", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/054775", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 1, - 4 - ] - ], - "date-time": "2017-01-04T06:15:43Z", - "timestamp": 1483510543000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 5, - 22 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/054775", - "relation": {}, - "id": "1Hk3NTSn2" - }, - "citation_id": "1Hk3NTSn2" - }, - "doi:10.1101/070441": { - "source": "doi", - "identifer": "10.1101/070441", - "standard_citation": "doi:10.1101/070441", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 21 - ] - ], - "date-time": "2017-08-21T15:22:51Z", - "timestamp": 1503328971114 - }, - "posted": { - "date-parts": [ - [ - 2016, - 8, - 21 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2017, - 1, - 14 - ] - ] - }, - "abstract": "To extract patterns from neuroimaging data, various statistical methods and machine learning algorithms have been explored for the diagnosis of Alzheimer′s disease among older adults in both clinical and research applications; however, distinguishing between Alzheimer′s and healthy brain data has been challenging in older adults (age > 75) due to highly similar patterns of brain atrophy and image intensities. Recently, cutting-edge deep learning technologies have rapidly expanded into numerous fields, including medical image analysis. This paper outlines state-of-the-art deep learning-based pipelines employed to distinguish Alzheimer′s magnetic resonance imaging (MRI) and functional MRI (fMRI) from normal healthy control data for a given age group. Using these pipelines, which were executed on a GPU-based high-performance computing platform, the data were strictly and carefully preprocessed. Next, scale- and shift-invariant low- to high-level features were obtained from a high volume of training images using convolutional neural network (CNN) architecture. In this study, fMRI data were used for the first time in deep learning applications for the purposes of medical image analysis and Alzheimer′s disease prediction. These proposed and implemented pipelines, which demonstrate a significant improvement in classification output over other studies, resulted in high and reproducible accuracy rates of 99.9% and 98.84% for the fMRI and MRI pipelines, respectively. Additionally, for clinical purposes, subject-level classification was performed, resulting in an average accuracy rate of 94.32% and 97.88% for the fMRI and MRI pipelines, respectively. Finally, a decision making algorithm designed for the subject-level classification improved the rate to 97.77% for fMRI and 100% for MRI pipelines.", - "DOI": "10.1101/070441", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2016, - 8, - 22 - ] - ], - "date-time": "2016-08-22T05:07:28Z", - "timestamp": 1471842448000 - }, - "source": "Crossref", - "is-referenced-by-count": 1, - "title": "DeepAD: Alzheimer′s Disease Classification via Deep Convolutional Neural Networks using MRI and fMRI", - "prefix": "10.1101", - "author": [ - { - "given": "Saman", - "family": "Sarraf", - "affiliation": [] - }, - { - "given": "Danielle D.", - "family": "DeSouza", - "affiliation": [] - }, - { - "given": "John", - "family": "Anderson", - "affiliation": [] - }, - { - "given": "Ghassem", - "family": "Tofighi", - "affiliation": [] - }, - { - "name": "for the Alzheimer's Disease Neuroimaging Initiativ", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/070441", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 1, - 15 - ] - ], - "date-time": "2017-01-15T06:10:13Z", - "timestamp": 1484460613000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 8, - 21 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/070441", - "relation": {}, - "id": "11NHbWB1V" - }, - "citation_id": "11NHbWB1V" - }, - "doi:10.1101/081364": { - "source": "doi", - "identifer": "10.1101/081364", - "standard_citation": "doi:10.1101/081364", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T01:35:48Z", - "timestamp": 1502328948014 - }, - "posted": { - "date-parts": [ - [ - 2016, - 10, - 17 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2017, - 6, - 5 - ] - ] - }, - "abstract": "We show that deep convolutional neural networks combined with non-linear dimension reduction enable reconstructing biological processes based on raw image data. We demonstrate this by reconstructing the cell cycle of Jurkat cells and disease progression in diabetic retinopathy. In further analysis of Jurkat cells, we detect and separate a subpopulation of dead cells in an unsupervised manner and, in classifying discrete cell cycle stages, we reach a 6-fold reduction in error rate compared to a recent approach based on boosting on image features. In contrast to previous methods, deep learning based predictions are fast enough for on-the-fly analysis in an imaging flow cytometer.", - "DOI": "10.1101/081364", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2016, - 10, - 19 - ] - ], - "date-time": "2016-10-19T05:13:03Z", - "timestamp": 1476853983000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Reconstructing cell cycle and disease progression using deep learning", - "prefix": "10.1101", - "author": [ - { - "given": "Philipp", - "family": "Eulenberg", - "affiliation": [] - }, - { - "given": "Niklas", - "family": "Koehler", - "affiliation": [] - }, - { - "given": "Thomas", - "family": "Blasi", - "affiliation": [] - }, - { - "given": "Andrew", - "family": "Filby", - "affiliation": [] - }, - { - "given": "Anne E.", - "family": "Carpenter", - "affiliation": [] - }, - { - "given": "Paul", - "family": "Rees", - "affiliation": [] - }, - { - "given": "Fabian J.", - "family": "Theis", - "affiliation": [] - }, - { - "given": "F. Alexander", - "family": "Wolf", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/081364", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 6 - ] - ], - "date-time": "2017-06-06T05:10:14Z", - "timestamp": 1496725814000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 10, - 17 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/081364", - "relation": {}, - "id": "gllSeTW" - }, - "citation_id": "gllSeTW" - }, - "doi:10.1101/085118": { - "source": "doi", - "identifer": "10.1101/085118", - "standard_citation": "doi:10.1101/085118", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T02:33:00Z", - "timestamp": 1502332380501 - }, - "posted": { - "date-parts": [ - [ - 2016, - 11, - 2 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2016, - 11, - 2 - ] - ] - }, - "abstract": "Morphological profiling aims to create signatures of genes, chemicals and diseases from microscopy images. Current approaches use classical computer vision-based segmentation and feature extraction. Deep learning models achieve state-of-the-art performance in many computer vision tasks such as classification and segmentation. We propose to transfer activation features of generic deep convolutional networks to extract features for morphological profiling. Our approach surpasses currently used methods in terms of accuracy and processing speed. Furthermore, it enables fully automated processing of microscopy images without need for single cell identification.", - "DOI": "10.1101/085118", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2016, - 11, - 3 - ] - ], - "date-time": "2016-11-03T05:12:05Z", - "timestamp": 1478149925000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Automating Morphological Profiling with Generic Deep Convolutional Networks", - "prefix": "10.1101", - "author": [ - { - "given": "Nick", - "family": "Pawlowski", - "affiliation": [] - }, - { - "given": "Juan C", - "family": "Caicedo", - "affiliation": [] - }, - { - "given": "Shantanu", - "family": "Singh", - "affiliation": [] - }, - { - "given": "Anne E", - "family": "Carpenter", - "affiliation": [] - }, - { - "given": "Amos", - "family": "Storkey", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/085118", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 1, - 4 - ] - ], - "date-time": "2017-01-04T06:17:11Z", - "timestamp": 1483510631000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 11, - 2 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/085118", - "relation": {}, - "id": "BMg062hc" - }, - "citation_id": "BMg062hc" - }, - "doi:10.1101/085241": { - "source": "doi", - "identifer": "10.1101/085241", - "standard_citation": "doi:10.1101/085241", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T02:33:00Z", - "timestamp": 1502332380487 - }, - "posted": { - "date-parts": [ - [ - 2016, - 11, - 2 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2016, - 11, - 2 - ] - ] - }, - "abstract": "In the human genome, distal enhancers are involved in regulating target genes through proximal promoters by forming enhancer-promoter interactions. However, although recently developed high-throughput experimental approaches have allowed us to recognize potential enhancer-promoter interactions genome-wide, it is still largely unknown whether there are sequence-level instructions encoded in our genome that help govern such interactions. Here we report a new computational method (named \"SPEID\") using deep learning models to predict enhancer-promoter interactions based on sequence-based features only, when the locations of putative enhancers and promoters in a particular cell type are given. Our results across six different cell types demonstrate that SPEID is effective in predicting enhancer-promoter interactions as compared to state-of-the-art methods that use non-sequence features from functional genomic signals. This work shows for the first time that sequence-based features alone can reliably predict enhancer-promoter interactions genome-wide, which provides important insights into the sequence determinants for long-range gene regulation.", - "DOI": "10.1101/085241", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2016, - 11, - 3 - ] - ], - "date-time": "2016-11-03T05:12:05Z", - "timestamp": 1478149925000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Predicting Enhancer-Promoter Interaction from Genomic Sequence with Deep Neural Networks", - "prefix": "10.1101", - "author": [ - { - "given": "Shashank", - "family": "Singh", - "affiliation": [] - }, - { - "given": "Yang", - "family": "Yang", - "affiliation": [] - }, - { - "given": "Barnabas", - "family": "Poczos", - "affiliation": [] - }, - { - "given": "Jian", - "family": "Ma", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/085241", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 1, - 4 - ] - ], - "date-time": "2017-01-04T06:17:14Z", - "timestamp": 1483510634000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 11, - 2 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/085241", - "relation": {}, - "id": "14TqLB9iZ" - }, - "citation_id": "14TqLB9iZ" - }, - "doi:10.1101/095786": { - "source": "doi", - "identifer": "10.1101/095786", - "standard_citation": "doi:10.1101/095786", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T05:44:11Z", - "timestamp": 1502343851195 - }, - "posted": { - "date-parts": [ - [ - 2016, - 12, - 20 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2016, - 12, - 20 - ] - ] - }, - "abstract": "Mass segmentation is an important task in mammogram analysis, providing effective morphological features and regions of interest (ROI) for mass detection and classification. Inspired by the success of using deep convolutional features for natural image analysis and conditional random fields (CRF) for structural learning, we propose an end-to-end network for mammographic mass segmentation. The network employs a fully convolutional network (FCN) to model potential function, followed by a CRF to perform structural learning. Because the mass distribution varies greatly with pixel position, the FCN is combined with position priori for the task. Due to the small size of mammogram datasets, we use adversarial training to control over-fitting. Four models with different convolutional kernels are further fused to improve the segmentation results. Experimental results on two public datasets, INbreast and DDSM-BCRP, show that our end-to-end network combined with adversarial training achieves the-state-of-the-art results.", - "DOI": "10.1101/095786", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2016, - 12, - 21 - ] - ], - "date-time": "2016-12-21T06:10:11Z", - "timestamp": 1482300611000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Adversarial Deep Structural Networks for Mammographic Mass Segmentation", - "prefix": "10.1101", - "author": [ - { - "given": "Wentao", - "family": "Zhu", - "affiliation": [] - }, - { - "given": "Xiaohui", - "family": "Xie", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/095786", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 1, - 4 - ] - ], - "date-time": "2017-01-04T06:18:45Z", - "timestamp": 1483510725000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 12, - 20 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/095786", - "relation": {}, - "id": "Xxb4t3zO" - }, - "citation_id": "Xxb4t3zO" - }, - "doi:10.1101/095794": { - "source": "doi", - "identifer": "10.1101/095794", - "standard_citation": "doi:10.1101/095794", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T05:44:11Z", - "timestamp": 1502343851150 - }, - "posted": { - "date-parts": [ - [ - 2016, - 12, - 20 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2016, - 12, - 20 - ] - ] - }, - "abstract": "Mammogram classification is directly related to computer-aided diagnosis of breast cancer. Traditional methods requires great effort to annotate the training data by costly manual labeling and specialized computational models to detect these annotations during test. Inspired by the success of using deep convolutional features for natural image analysis and multi-instance learning for labeling a set of instances/patches, we propose end-to-end trained deep multi-instance networks for mass classification based on whole mammogram without the aforementioned costly need to annotate the training data. We explore three different schemes to construct deep multi-instance networks for whole mammogram classification. Experimental results on the INbreast dataset demonstrate the robustness of proposed deep networks compared to previous work using segmentation and detection annotations in the training.", - "DOI": "10.1101/095794", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2016, - 12, - 21 - ] - ], - "date-time": "2016-12-21T06:10:11Z", - "timestamp": 1482300611000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Deep Multi-instance Networks with Sparse Label Assignment for Whole Mammogram Classification", - "prefix": "10.1101", - "author": [ - { - "given": "Wentao", - "family": "Zhu", - "affiliation": [] - }, - { - "given": "Qi", - "family": "Lou", - "affiliation": [] - }, - { - "given": "Yeeleng Scott", - "family": "Vang", - "affiliation": [] - }, - { - "given": "Xiaohui", - "family": "Xie", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/095794", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 1, - 4 - ] - ], - "date-time": "2017-01-04T06:18:51Z", - "timestamp": 1483510731000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 12, - 20 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/095794", - "relation": {}, - "id": "9G9Hv1Pp" - }, - "citation_id": "9G9Hv1Pp" - }, - "doi:10.1101/125229": { - "source": "doi", - "identifer": "10.1101/125229", - "standard_citation": "doi:10.1101/125229", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 11 - ] - ], - "date-time": "2017-08-11T03:50:28Z", - "timestamp": 1502423428882 - }, - "posted": { - "date-parts": [ - [ - 2017, - 4, - 7 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2017, - 4, - 7 - ] - ] - }, - "abstract": "Motivation: Reconstructing the full-length expressed transcripts (a.k.a. the transcript assembly problem) from the short sequencing reads produced by RNA-seq protocol plays a central role in identifying novel genes and transcripts as well as in studying gene expressions and gene functions. A crucial step in transcript assembly is to accurately determine the splicing junctions and boundaries of the expressed transcripts from the reads alignment. In contrast to the splicing junctions that can be efficiently detected from spliced reads, the problem of identifying boundaries remains open and challenging, due to the fact that the signal related to boundaries is noisy and weak.\n\nResults: We present DeepBound, an effective approach to identify boundaries of expressed transcripts from RNA-seq reads alignment. In its core DeepBound employs deep convolutional neural fields to learn the hidden distributions and patterns of boundaries. To accurately model the transition probabilities and to solve the label-imbalance problem, we novelly incorporate the AUC (area under the curve) score into the optimizing objective function. To address the issue that deep probabilistic graphical models requires large number of labeled training samples, we propose to use simulated RNA-seq datasets to train our model. Through extensive experimental studies on both simulation datasets of two species and biological datasets, we show that DeepBound consistently and significantly outperforms the two existing methods. \n\nAvailability: DeepBound is freely available at https://github.com/realbigws/DeepBound. \n\nContact: mingfu.shao@cs.cmu.edu, realbigws@gmail.com", - "DOI": "10.1101/125229", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2017, - 4, - 8 - ] - ], - "date-time": "2017-04-08T05:10:17Z", - "timestamp": 1491628217000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "DeepBound: Accurate Identification of Transcript Boundaries via Deep Convolutional Neural Fields", - "prefix": "10.1101", - "author": [ - { - "given": "Mingfu", - "family": "Shao", - "affiliation": [] - }, - { - "given": "Jianzhu", - "family": "Ma", - "affiliation": [] - }, - { - "given": "Sheng", - "family": "Wang", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/125229", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 4, - 8 - ] - ], - "date-time": "2017-04-08T05:10:25Z", - "timestamp": 1491628225000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 4, - 7 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/125229", - "relation": {}, - "id": "2M3zXijc" - }, - "citation_id": "2M3zXijc" - }, - "doi:10.1101/138685": { - "source": "doi", - "identifer": "10.1101/138685", - "standard_citation": "doi:10.1101/138685", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 11 - ] - ], - "date-time": "2017-08-11T14:49:10Z", - "timestamp": 1502462950421 - }, - "posted": { - "date-parts": [ - [ - 2017, - 5, - 17 - ] - ] - }, - "group-title": "Genomics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2017, - 5, - 17 - ] - ] - }, - "abstract": "Parallel single-cell sequencing protocols represent powerful methods for investigating regulatory relationships, including epigenome-transcriptome interactions. Here, we report the first single-cell method for parallel chromatin accessibility, DNA methylation and transcriptome profiling. scNMT-seq (single-cell nucleosome, methylation and transcription sequencing) uses a GpC methyltransferase to label open chromatin followed by bisulfite and RNA sequencing. We validate scNMT-seq by applying it to mouse embryonic stem cells, finding links between all three molecular layers and revealing strong and widespread associations between chromatin accessibility and DNA methylation.", - "DOI": "10.1101/138685", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2017, - 5, - 18 - ] - ], - "date-time": "2017-05-18T05:10:13Z", - "timestamp": 1495084213000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Joint Profiling Of Chromatin Accessibility, DNA Methylation And Transcription In Single Cells", - "prefix": "10.1101", - "author": [ - { - "given": "Stephen J.", - "family": "Clark", - "affiliation": [] - }, - { - "given": "Ricard", - "family": "Argelaguet", - "affiliation": [] - }, - { - "given": "Chantriolnt-Andreas", - "family": "Kapourani", - "affiliation": [] - }, - { - "given": "Thomas M.", - "family": "Stubbs", - "affiliation": [] - }, - { - "given": "Heather J.", - "family": "Lee", - "affiliation": [] - }, - { - "given": "Felix", - "family": "Krueger", - "affiliation": [] - }, - { - "given": "Guido", - "family": "Sanguinetti", - "affiliation": [] - }, - { - "given": "Gavin", - "family": "Kelsey", - "affiliation": [] - }, - { - "given": "John C.", - "family": "Marioni", - "affiliation": [] - }, - { - "given": "Oliver", - "family": "Stegle", - "affiliation": [] - }, - { - "given": "Wolf", - "family": "Reik", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/138685", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 5, - 18 - ] - ], - "date-time": "2017-05-18T05:10:14Z", - "timestamp": 1495084214000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 5, - 17 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/138685", - "relation": {}, - "id": "1CAw3FaPI" - }, - "citation_id": "1CAw3FaPI" - }, - "doi:10.1101/gr.110254.110": { - "source": "doi", - "identifer": "10.1101/gr.110254.110", - "standard_citation": "doi:10.1101/gr.110254.110", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 11 - ] - ], - "date-time": "2017-08-11T02:44:48Z", - "timestamp": 1502419488000 - }, - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "issue": "5", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2011, - 5, - 1 - ] - ] - }, - "DOI": "10.1101/gr.110254.110", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2011, - 3, - 4 - ] - ], - "date-time": "2011-03-04T04:00:40Z", - "timestamp": 1299211240000 - }, - "page": "775-789", - "source": "Crossref", - "is-referenced-by-count": 55, - "title": "Genome-wide characterization of transcriptional start sites in humans by integrative transcriptome analysis", - "prefix": "10.1101", - "volume": "21", - "author": [ - { - "given": "R.", - "family": "Yamashita", - "affiliation": [] - }, - { - "given": "N. P.", - "family": "Sathira", - "affiliation": [] - }, - { - "given": "A.", - "family": "Kanai", - "affiliation": [] - }, - { - "given": "K.", - "family": "Tanimoto", - "affiliation": [] - }, - { - "given": "T.", - "family": "Arauchi", - "affiliation": [] - }, - { - "given": "Y.", - "family": "Tanaka", - "affiliation": [] - }, - { - "given": "S.-i.", - "family": "Hashimoto", - "affiliation": [] - }, - { - "given": "S.", - "family": "Sugano", - "affiliation": [] - }, - { - "given": "K.", - "family": "Nakai", - "affiliation": [] - }, - { - "given": "Y.", - "family": "Suzuki", - "affiliation": [] - } - ], - "member": "246", - "published-online": { - "date-parts": [ - [ - 2011, - 3, - 3 - ] - ] - }, - "container-title": "Genome Research", - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/gr.110254.110", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 2, - 15 - ] - ], - "date-time": "2017-02-15T17:42:07Z", - "timestamp": 1487180527000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2011, - 3, - 3 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/gr.110254.110", - "relation": {}, - "subject": [ - "Genetics(clinical)", - "Genetics" - ], - "container-title-short": "Genome Research", - "id": "G0J9P3Ln" - }, - "citation_id": "G0J9P3Ln" - }, - "doi:10.1101/gr.200535.115": { - "source": "doi", - "identifer": "10.1101/gr.200535.115", - "standard_citation": "doi:10.1101/gr.200535.115", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T14:02:16Z", - "timestamp": 1507039336011 - }, - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "issue": "7", - "funder": [ - { - "DOI": "10.13039/100000066", - "name": "National Institute of Environmental Health Sciences", - "doi-asserted-by": "publisher", - "award": [ - "ES022984-02" - ] - }, - { - "DOI": "10.13039/100000025", - "name": "National Institute of Mental Health", - "doi-asserted-by": "publisher", - "award": [ - "R01MH102416-03" - ] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 7 - ] - ] - }, - "DOI": "10.1101/gr.200535.115", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 5, - 4 - ] - ], - "date-time": "2016-05-04T02:43:56Z", - "timestamp": 1462329836000 - }, - "page": "990-999", - "source": "Crossref", - "is-referenced-by-count": 23, - "title": "Basset: learning the regulatory code of the accessible genome with deep convolutional neural networks", - "prefix": "10.1101", - "volume": "26", - "author": [ - { - "given": "David R.", - "family": "Kelley", - "affiliation": [] - }, - { - "given": "Jasper", - "family": "Snoek", - "affiliation": [] - }, - { - "given": "John L.", - "family": "Rinn", - "affiliation": [] - } - ], - "member": "246", - "published-online": { - "date-parts": [ - [ - 2016, - 5, - 3 - ] - ] - }, - "container-title": "Genome Research", - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/gr.200535.115", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 2, - 15 - ] - ], - "date-time": "2017-02-15T17:46:41Z", - "timestamp": 1487180801000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 5, - 3 - ] - ] - }, - "references-count": 0, - "alternative-id": [ - "10.1101/gr.200535.115" - ], - "URL": "https://doi.org/10.1101/gr.200535.115", - "relation": { - "has-preprint": [ - { - "id-type": "doi", - "id": "10.1101/028399", - "asserted-by": "object" - } - ] - }, - "subject": [ - "Genetics(clinical)", - "Genetics" - ], - "container-title-short": "Genome Res.", - "id": "2CbHXoFn" - }, - "citation_id": "2CbHXoFn" - }, - "doi:10.1103/revmodphys.34.135": { - "source": "doi", - "identifer": "10.1103/revmodphys.34.135", - "standard_citation": "doi:10.1103/revmodphys.34.135", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 8 - ] - ], - "date-time": "2017-08-08T12:32:46Z", - "timestamp": 1502195566636 - }, - "reference-count": 4, - "publisher": "American Physical Society (APS)", - "issue": "1", - "license": [ - { - "URL": "http://link.aps.org/licenses/aps-default-license", - "start": { - "date-parts": [ - [ - 1962, - 1, - 1 - ] - ], - "date-time": "1962-01-01T00:00:00Z", - "timestamp": -252460800000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1103/revmodphys.34.135", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2002, - 7, - 27 - ] - ], - "date-time": "2002-07-27T04:44:24Z", - "timestamp": 1027745064000 - }, - "page": "135-142", - "source": "Crossref", - "is-referenced-by-count": 35, - "title": "Analysis of a Four-Layer Series-Coupled Perceptron. II", - "prefix": "10.1103", - "volume": "34", - "author": [ - { - "given": "H. D.", - "family": "Block", - "affiliation": [] - }, - { - "given": "B. W.", - "family": "Knight", - "affiliation": [] - }, - { - "given": "F.", - "family": "Rosenblatt", - "affiliation": [] - } - ], - "member": "16", - "published-online": { - "date-parts": [ - [ - 1962, - 1, - 1 - ] - ] - }, - "container-title": "Reviews of Modern Physics", - "original-title": [], - "link": [ - { - "URL": "http://link.aps.org/article/10.1103/RevModPhys.34.135", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "syndication" - }, - { - "URL": "http://harvest.aps.org/v2/journals/articles/10.1103/RevModPhys.34.135/fulltext", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 15 - ] - ], - "date-time": "2017-06-15T09:48:11Z", - "timestamp": 1497520091000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 1962, - 1, - 1 - ] - ] - }, - "references-count": 4, - "URL": "https://doi.org/10.1103/revmodphys.34.135", - "relation": { - "cites": [] - }, - "subject": [ - "General Physics and Astronomy" - ], - "container-title-short": "Rev. Mod. Phys.", - "id": "1G5eCiq4d" - }, - "citation_id": "1G5eCiq4d" - }, - "doi:10.1109/bibm.2014.6999129": { - "source": "doi", - "identifer": "10.1109/bibm.2014.6999129", - "standard_citation": "doi:10.1109/bibm.2014.6999129", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T19:09:06Z", - "timestamp": 1502305746802 - }, - "reference-count": 19, - "publisher": "IEEE", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2014, - 11 - ] - ] - }, - "DOI": "10.1109/bibm.2014.6999129", - "type": "paper-conference", - "created": { - "date-parts": [ - [ - 2015, - 1, - 20 - ] - ], - "date-time": "2015-01-20T02:42:30Z", - "timestamp": 1421721750000 - }, - "source": "Crossref", - "is-referenced-by-count": 3, - "title": "Pairwise input neural network for target-ligand interaction prediction", - "prefix": "10.1109", - "author": [ - { - "given": "Caihua", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Juan", - "family": "Liu", - "affiliation": [] - }, - { - "given": "Fei", - "family": "Luo", - "affiliation": [] - }, - { - "given": "Yafang", - "family": "Tan", - "affiliation": [] - }, - { - "given": "Zixin", - "family": "Deng", - "affiliation": [] - }, - { - "given": "Qian-Nan", - "family": "Hu", - "affiliation": [] - } - ], - "member": "263", - "container-title": "2014 IEEE International Conference on Bioinformatics and Biomedicine (BIBM)", - "original-title": [], - "link": [ - { - "URL": "http://xplorestaging.ieee.org/ielx7/6979782/6999111/06999129.pdf?arnumber=6999129", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 23 - ] - ], - "date-time": "2017-06-23T03:54:45Z", - "timestamp": 1498190085000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 11 - ] - ] - }, - "references-count": 19, - "URL": "https://doi.org/10.1109/bibm.2014.6999129", - "relation": {}, - "id": "TeIxEjqm" - }, - "citation_id": "TeIxEjqm" - }, - "doi:10.1109/tcbb.2014.2343960": { - "source": "doi", - "identifer": "10.1109/tcbb.2014.2343960", - "standard_citation": "doi:10.1109/tcbb.2014.2343960", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T04:22:16Z", - "timestamp": 1507004536713 - }, - "reference-count": 52, - "publisher": "Institute of Electrical and Electronics Engineers (IEEE)", - "issue": "1", - "funder": [ - { - "DOI": "10.13039/100000002", - "name": "National Institutes of Health", - "doi-asserted-by": "publisher", - "award": [ - "R01GM093123" - ] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015, - 1, - 1 - ] - ] - }, - "DOI": "10.1109/tcbb.2014.2343960", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 8, - 7 - ] - ], - "date-time": "2014-08-07T18:30:15Z", - "timestamp": 1407436215000 - }, - "page": "103-112", - "source": "Crossref", - "is-referenced-by-count": 31, - "title": "A Deep Learning Network Approach to ab initio Protein Secondary Structure Prediction", - "prefix": "10.1109", - "volume": "12", - "author": [ - { - "given": "Matt", - "family": "Spencer", - "affiliation": [] - }, - { - "given": "Jesse", - "family": "Eickholt", - "affiliation": [] - }, - { - "given": "Jianlin", - "family": "Cheng", - "affiliation": [] - } - ], - "member": "263", - "container-title": "IEEE/ACM Transactions on Computational Biology and Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://xplorestaging.ieee.org/ielx7/8857/7035191/06872810.pdf?arnumber=6872810", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 3, - 23 - ] - ], - "date-time": "2017-03-23T17:05:44Z", - "timestamp": 1490288744000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 1, - 1 - ] - ] - }, - "references-count": 52, - "URL": "https://doi.org/10.1109/tcbb.2014.2343960", - "relation": {}, - "subject": [ - "Biotechnology", - "Genetics", - "Applied Mathematics" - ], - "container-title-short": "IEEE/ACM Trans. Comput. Biol. and Bioinf.", - "id": "ZzaRyGuJ" - }, - "citation_id": "ZzaRyGuJ" - }, - "doi:10.1109/icnn.1994.374637": { - "source": "doi", - "identifer": "10.1109/icnn.1994.374637", - "standard_citation": "doi:10.1109/icnn.1994.374637", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 5, - 21 - ] - ], - "date-time": "2017-05-21T01:42:31Z", - "timestamp": 1495330951449 - }, - "issued": { - "date-parts": [ - [ - 1994 - ] - ] - }, - "reference-count": 7, - "publisher": "IEEE", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "short-container-title": [], - "DOI": "10.1109/icnn.1994.374637", - "type": "paper-conference", - "created": { - "date-parts": [ - [ - 2002, - 12, - 17 - ] - ], - "date-time": "2002-12-17T16:08:05Z", - "timestamp": 1040141285000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "End effector target position learning using feedforward with error back-propagation and recurrent neural networks", - "prefix": "10.1109", - "author": [ - { - "given": "J.", - "family": "Zurada", - "affiliation": [] - } - ], - "member": "263", - "container-title": "Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 3, - 9 - ] - ], - "date-time": "2017-03-09T07:36:33Z", - "timestamp": 1489044993000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "references-count": 7, - "URL": "https://doi.org/10.1109/icnn.1994.374637", - "relation": {}, - "id": "YUms527e" - }, - "citation_id": "YUms527e" - }, - "doi:10.1111/j.1740-9713.2016.00960.x": { - "source": "doi", - "identifer": "10.1111/j.1740-9713.2016.00960.x", - "standard_citation": "doi:10.1111/j.1740-9713.2016.00960.x", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 22 - ] - ], - "date-time": "2017-08-22T19:02:54Z", - "timestamp": 1503428574241 - }, - "reference-count": 13, - "publisher": "Wiley-Blackwell", - "issue": "5", - "license": [ - { - "URL": "http://doi.wiley.com/10.1002/tdm_license_1.1", - "start": { - "date-parts": [ - [ - 2015, - 9, - 1 - ] - ], - "date-time": "2015-09-01T00:00:00Z", - "timestamp": 1441065600000 - }, - "delay-in-days": 0, - "content-version": "tdm" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 10 - ] - ] - }, - "DOI": "10.1111/j.1740-9713.2016.00960.x", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 10, - 13 - ] - ], - "date-time": "2016-10-13T05:39:18Z", - "timestamp": 1476337158000 - }, - "page": "14-19", - "source": "Crossref", - "is-referenced-by-count": 4, - "title": "To predict and serve?", - "prefix": "10.1111", - "volume": "13", - "author": [ - { - "given": "Kristian", - "family": "Lum", - "affiliation": [ - { - "name": "Lead statistician at the Human Rights Data Analysis Group" - } - ] - }, - { - "given": "William", - "family": "Isaac", - "affiliation": [ - { - "name": "Doctoral candidate in the Department of Political Science at Michigan State University" - } - ] - } - ], - "member": "311", - "published-online": { - "date-parts": [ - [ - 2016, - 10, - 7 - ] - ] - }, - "container-title": "Significance", - "original-title": [], - "link": [ - { - "URL": "https://api.wiley.com/onlinelibrary/tdm/v1/articles/10.1111%2Fj.1740-9713.2016.00960.x", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "https://api.wiley.com/onlinelibrary/tdm/v1/articles/10.1111%2Fj.1740-9713.2016.00960.x", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://onlinelibrary.wiley.com/wol1/doi/10.1111/j.1740-9713.2016.00960.x/fullpdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T01:00:33Z", - "timestamp": 1498352433000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 10 - ] - ] - }, - "references-count": 13, - "URL": "https://doi.org/10.1111/j.1740-9713.2016.00960.x", - "relation": { - "cites": [] - }, - "subject": [ - "Statistics and Probability" - ], - "container-title-short": "Significance", - "id": "6co0adq" - }, - "citation_id": "6co0adq" - }, - "doi:10.1126/science.1132939": { - "source": "doi", - "identifer": "10.1126/science.1132939", - "standard_citation": "doi:10.1126/science.1132939", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T04:02:15Z", - "timestamp": 1507089735888 - }, - "reference-count": 0, - "publisher": "American Association for the Advancement of Science (AAAS)", - "issue": "5795", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2006, - 9, - 29 - ] - ] - }, - "DOI": "10.1126/science.1132939", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2006, - 9, - 28 - ] - ], - "date-time": "2006-09-28T21:35:12Z", - "timestamp": 1159479312000 - }, - "page": "1929-1935", - "source": "Crossref", - "is-referenced-by-count": 1475, - "title": "The Connectivity Map: Using Gene-Expression Signatures to Connect Small Molecules, Genes, and Disease", - "prefix": "10.1126", - "volume": "313", - "author": [ - { - "given": "J.", - "family": "Lamb", - "affiliation": [] - } - ], - "member": "221", - "container-title": "Science", - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1126/science.1132939", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 12, - 23 - ] - ], - "date-time": "2016-12-23T04:11:45Z", - "timestamp": 1482466305000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2006, - 9, - 29 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1126/science.1132939", - "relation": {}, - "subject": [ - "Multidisciplinary" - ], - "container-title-short": "Science", - "id": "Ot5bUkmI" - }, - "citation_id": "Ot5bUkmI" - }, - "doi:10.1126/science.333.6041.404-a": { - "source": "doi", - "identifer": "10.1126/science.333.6041.404-a", - "standard_citation": "doi:10.1126/science.333.6041.404-a", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 11 - ] - ], - "date-time": "2017-08-11T10:58:26Z", - "timestamp": 1502449106504 - }, - "reference-count": 1, - "publisher": "American Association for the Advancement of Science (AAAS)", - "issue": "6041", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2011, - 7, - 22 - ] - ] - }, - "DOI": "10.1126/science.333.6041.404-a", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2011, - 7, - 21 - ] - ], - "date-time": "2011-07-21T18:49:01Z", - "timestamp": 1311274141000 - }, - "page": "404-404", - "source": "Crossref", - "is-referenced-by-count": 20, - "title": "Retraction", - "prefix": "10.1126", - "volume": "333", - "author": [ - { - "given": "P.", - "family": "Sebastiani", - "affiliation": [] - }, - { - "given": "N.", - "family": "Solovieff", - "affiliation": [] - }, - { - "given": "A.", - "family": "Puca", - "affiliation": [] - }, - { - "given": "S. W.", - "family": "Hartley", - "affiliation": [] - }, - { - "given": "E.", - "family": "Melista", - "affiliation": [] - }, - { - "given": "S.", - "family": "Andersen", - "affiliation": [] - }, - { - "given": "D. A.", - "family": "Dworkis", - "affiliation": [] - }, - { - "given": "J. B.", - "family": "Wilk", - "affiliation": [] - }, - { - "given": "R. H.", - "family": "Myers", - "affiliation": [] - }, - { - "given": "M. H.", - "family": "Steinberg", - "affiliation": [] - }, - { - "given": "M.", - "family": "Montano", - "affiliation": [] - }, - { - "given": "C. T.", - "family": "Baldwin", - "affiliation": [] - }, - { - "given": "T. T.", - "family": "Perls", - "affiliation": [] - } - ], - "member": "221", - "published-online": { - "date-parts": [ - [ - 2011, - 7, - 21 - ] - ] - }, - "container-title": "Science", - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1126/science.333.6041.404-a", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 12, - 23 - ] - ], - "date-time": "2016-12-23T17:48:17Z", - "timestamp": 1482515297000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2011, - 7, - 21 - ] - ] - }, - "references-count": 1, - "URL": "https://doi.org/10.1126/science.333.6041.404-a", - "relation": { - "cites": [] - }, - "subject": [ - "Multidisciplinary" - ], - "container-title-short": "Science", - "id": "DmQPI43R" - }, - "citation_id": "DmQPI43R" - }, - "doi:10.1126/science.aah6168": { - "source": "doi", - "identifer": "10.1126/science.aah6168", - "standard_citation": "doi:10.1126/science.aah6168", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T20:22:11Z", - "timestamp": 1507062131194 - }, - "reference-count": 11, - "publisher": "American Association for the Advancement of Science (AAAS)", - "issue": "6317", - "license": [ - { - "URL": "http://www.sciencemag.org/about/science-licenses-journal-article-reuse", - "start": { - "date-parts": [ - [ - 2017, - 12, - 9 - ] - ], - "date-time": "2017-12-09T00:00:00Z", - "timestamp": 1512777600000 - }, - "delay-in-days": 366, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/100009827", - "name": "Laura and John Arnold Foundation", - "doi-asserted-by": "publisher", - "award": [] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 12, - 9 - ] - ] - }, - "DOI": "10.1126/science.aah6168", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 12, - 8 - ] - ], - "date-time": "2016-12-08T19:10:19Z", - "timestamp": 1481224219000 - }, - "page": "1240-1241", - "source": "Crossref", - "is-referenced-by-count": 16, - "title": "Enhancing reproducibility for computational methods", - "prefix": "10.1126", - "volume": "354", - "author": [ - { - "given": "V.", - "family": "Stodden", - "affiliation": [] - }, - { - "given": "M.", - "family": "McNutt", - "affiliation": [] - }, - { - "given": "D. H.", - "family": "Bailey", - "affiliation": [] - }, - { - "given": "E.", - "family": "Deelman", - "affiliation": [] - }, - { - "given": "Y.", - "family": "Gil", - "affiliation": [] - }, - { - "given": "B.", - "family": "Hanson", - "affiliation": [] - }, - { - "given": "M. A.", - "family": "Heroux", - "affiliation": [] - }, - { - "given": "J. P. A.", - "family": "Ioannidis", - "affiliation": [] - }, - { - "given": "M.", - "family": "Taufer", - "affiliation": [] - } - ], - "member": "221", - "published-online": { - "date-parts": [ - [ - 2016, - 12, - 8 - ] - ] - }, - "container-title": "Science", - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1126/science.aah6168", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T05:32:15Z", - "timestamp": 1498368735000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 12, - 8 - ] - ] - }, - "references-count": 11, - "URL": "https://doi.org/10.1126/science.aah6168", - "relation": { - "cites": [] - }, - "subject": [ - "Multidisciplinary" - ], - "container-title-short": "Science", - "id": "gvyja7v1" - }, - "citation_id": "gvyja7v1" - }, - "doi:10.1136/amiajnl-2011-000150": { - "source": "doi", - "identifer": "10.1136/amiajnl-2011-000150", - "standard_citation": "doi:10.1136/amiajnl-2011-000150", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 19 - ] - ], - "date-time": "2017-09-19T04:02:24Z", - "timestamp": 1505793744482 - }, - "reference-count": 17, - "publisher": "Oxford University Press (OUP)", - "issue": "5", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2011, - 9 - ] - ] - }, - "DOI": "10.1136/amiajnl-2011-000150", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2011, - 5, - 13 - ] - ], - "date-time": "2011-05-13T00:39:49Z", - "timestamp": 1305247189000 - }, - "page": "557-562", - "source": "Crossref", - "is-referenced-by-count": 54, - "title": "Machine-learned solutions for three stages of clinical information extraction: the state of the art at i2b2 2010", - "prefix": "10.1093", - "volume": "18", - "author": [ - { - "given": "Berry", - "family": "de Bruijn", - "affiliation": [] - }, - { - "given": "Colin", - "family": "Cherry", - "affiliation": [] - }, - { - "given": "Svetlana", - "family": "Kiritchenko", - "affiliation": [] - }, - { - "given": "Joel", - "family": "Martin", - "affiliation": [] - }, - { - "given": "Xiaodan", - "family": "Zhu", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2011, - 9, - 1 - ] - ] - }, - "container-title": "Journal of the American Medical Informatics Association", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/jamia/article-pdf/18/5/557/5963587/18-5-557.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 23 - ] - ], - "date-time": "2017-08-23T12:36:35Z", - "timestamp": 1503491795000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2011, - 9 - ] - ] - }, - "references-count": 17, - "alternative-id": [ - "10.1136/amiajnl-2011-000150" - ], - "URL": "https://doi.org/10.1136/amiajnl-2011-000150", - "relation": {}, - "subject": [ - "Health Informatics" - ], - "container-title-short": "J Am Med Inform Assoc", - "id": "sG3iVOTS" - }, - "citation_id": "sG3iVOTS" - }, - "doi:10.1136/amiajnl-2011-000501": { - "source": "doi", - "identifer": "10.1136/amiajnl-2011-000501", - "standard_citation": "doi:10.1136/amiajnl-2011-000501", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 14 - ] - ], - "date-time": "2017-09-14T09:22:13Z", - "timestamp": 1505380933216 - }, - "reference-count": 0, - "publisher": "Oxford University Press (OUP)", - "issue": "5", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2011, - 9 - ] - ] - }, - "DOI": "10.1136/amiajnl-2011-000501", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2011, - 8, - 16 - ] - ], - "date-time": "2011-08-16T20:50:05Z", - "timestamp": 1313527805000 - }, - "page": "539-539", - "source": "Crossref", - "is-referenced-by-count": 15, - "title": "Realizing the full potential of electronic health records: the role of natural language processing", - "prefix": "10.1093", - "volume": "18", - "author": [ - { - "given": "Lucila", - "family": "Ohno-Machado", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2011, - 9, - 1 - ] - ] - }, - "container-title": "Journal of the American Medical Informatics Association", - "original-title": [], - "link": [ - { - "URL": "https://academic.oup.com/jamia/article/18/5/539/829383/Realizing-the-full-potential-of-electronic-health", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 1, - 23 - ] - ], - "date-time": "2017-01-23T22:52:29Z", - "timestamp": 1485211949000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2011, - 9 - ] - ] - }, - "references-count": 0, - "alternative-id": [ - "10.1136/amiajnl-2011-000501" - ], - "URL": "https://doi.org/10.1136/amiajnl-2011-000501", - "relation": {}, - "subject": [ - "Health Informatics" - ], - "container-title-short": "J Am Med Inform Assoc", - "id": "uDaRUyh9" - }, - "citation_id": "uDaRUyh9" - }, - "doi:10.1136/amiajnl-2011-000681": { - "source": "doi", - "identifer": "10.1136/amiajnl-2011-000681", - "standard_citation": "doi:10.1136/amiajnl-2011-000681", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 14 - ] - ], - "date-time": "2017-09-14T02:22:12Z", - "timestamp": 1505355732785 - }, - "reference-count": 100, - "publisher": "Oxford University Press (OUP)", - "issue": "1", - "content-domain": { - "domain": [ - "bmj.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2013, - 1, - 1 - ] - ] - }, - "DOI": "10.1136/amiajnl-2011-000681", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2012, - 6, - 27 - ] - ], - "date-time": "2012-06-27T07:32:22Z", - "timestamp": 1340782342000 - }, - "page": "144-151", - "update-policy": "http://dx.doi.org/10.1136/crossmarkpolicy", - "source": "Crossref", - "is-referenced-by-count": 149, - "title": "Methods and dimensions of electronic health record data quality assessment: enabling reuse for clinical research", - "prefix": "10.1093", - "volume": "20", - "author": [ - { - "given": "N. G.", - "family": "Weiskopf", - "affiliation": [] - }, - { - "given": "C.", - "family": "Weng", - "affiliation": [] - } - ], - "member": "286", - "container-title": "Journal of the American Medical Informatics Association", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 20 - ] - ], - "date-time": "2017-06-20T20:34:41Z", - "timestamp": 1497990881000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2013, - 1, - 1 - ] - ] - }, - "references-count": 100, - "URL": "https://doi.org/10.1136/amiajnl-2011-000681", - "relation": {}, - "subject": [ - "Health Informatics" - ], - "container-title-short": "Journal of the American Medical Informatics Association", - "id": "odRiFxnB" - }, - "citation_id": "odRiFxnB" - }, - "doi:10.1136/amiajnl-2013-001935": { - "source": "doi", - "identifer": "10.1136/amiajnl-2013-001935", - "standard_citation": "doi:10.1136/amiajnl-2013-001935", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T04:02:42Z", - "timestamp": 1507003362662 - }, - "reference-count": 102, - "publisher": "Oxford University Press (OUP)", - "issue": "2", - "content-domain": { - "domain": [ - "bmj.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2014, - 3 - ] - ] - }, - "DOI": "10.1136/amiajnl-2013-001935", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2013, - 11, - 8 - ] - ], - "date-time": "2013-11-08T02:56:57Z", - "timestamp": 1383879417000 - }, - "page": "221-230", - "update-policy": "http://dx.doi.org/10.1136/crossmarkpolicy", - "source": "Crossref", - "is-referenced-by-count": 61, - "title": "A review of approaches to identifying patient phenotype cohorts using electronic health records", - "prefix": "10.1093", - "volume": "21", - "author": [ - { - "given": "Chaitanya", - "family": "Shivade", - "affiliation": [] - }, - { - "given": "Preethi", - "family": "Raghavan", - "affiliation": [] - }, - { - "given": "Eric", - "family": "Fosler-Lussier", - "affiliation": [] - }, - { - "given": "Peter J", - "family": "Embi", - "affiliation": [] - }, - { - "given": "Noemie", - "family": "Elhadad", - "affiliation": [] - }, - { - "given": "Stephen B", - "family": "Johnson", - "affiliation": [] - }, - { - "given": "Albert M", - "family": "Lai", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2014, - 3, - 1 - ] - ] - }, - "container-title": "Journal of the American Medical Informatics Association", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 1, - 23 - ] - ], - "date-time": "2017-01-23T22:56:35Z", - "timestamp": 1485212195000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 3 - ] - ] - }, - "references-count": 102, - "alternative-id": [ - "10.1136/amiajnl-2013-001935" - ], - "URL": "https://doi.org/10.1136/amiajnl-2013-001935", - "relation": {}, - "subject": [ - "Health Informatics" - ], - "container-title-short": "J Am Med Inform Assoc", - "id": "11OyzMl87" - }, - "citation_id": "11OyzMl87" - }, - "doi:10.1142/9789813207813_0050": { - "source": "doi", - "identifer": "10.1142/9789813207813_0050", - "standard_citation": "doi:10.1142/9789813207813_0050", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T03:55:38Z", - "timestamp": 1502337338203 - }, - "reference-count": 0, - "publisher": "WORLD SCIENTIFIC", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2017, - 1 - ] - ] - }, - "DOI": "10.1142/9789813207813_0050", - "type": "paper-conference", - "created": { - "date-parts": [ - [ - 2016, - 11, - 23 - ] - ], - "date-time": "2016-11-23T06:53:50Z", - "timestamp": 1479884030000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "STRATEGIES FOR EQUITABLE PHARMACOGENOMIC-GUIDED WARFARIN DOSING AMONG EUROPEAN AND AFRICAN AMERICAN INDIVIDUALS IN A CLINICAL POPULATION", - "prefix": "10.1142", - "author": [ - { - "given": "LAURA K.", - "family": "WILEY", - "affiliation": [ - { - "name": "Div. of Biomedical Informatics and Personalized Med., University of Colorado, 13001 E. 17th Pl. MS F-563 Aurora, CO 80045, USA" - } - ] - }, - { - "given": "JACOB P.", - "family": "VANHOUTEN", - "affiliation": [ - { - "name": "Dept. of Biomedical Informatics, Vanderbilt University, 2525 West End Ave Ste. 1475 Nashville, TN 37203, USA" - } - ] - }, - { - "given": "DAVID C.", - "family": "SAMUELS", - "affiliation": [ - { - "name": "Dept. of Mol. Physiology & Biophysics, Vanderbilt Genetics Inst., Vanderbilt University, 2215 Garland Ave. Nashville, TN 37232, USA" - } - ] - }, - { - "given": "MELINDA C.", - "family": "ALDRICH", - "affiliation": [ - { - "name": "Dept. of Thoracic Surgery, Div. of Epidemiology, Vanderbilt University Medical Center, 609 Oxford House Nashville, TN 37232, USA" - } - ] - }, - { - "given": "DAN M.", - "family": "RODEN", - "affiliation": [ - { - "name": "Dept. of Medicine, Vanderbilt University, 2215B Garland Ave Nashville, TN 37203, USA" - } - ] - }, - { - "given": "JOSH F.", - "family": "PETERSON", - "affiliation": [ - { - "name": "Dept. of Biomedical Informatics, Dept. of Medicine, Vanderbilt University, 2525 West End Ave Ste.1050 Nashville, TN 37203, USA" - } - ] - }, - { - "given": "JOSHUA C.", - "family": "DENNY", - "affiliation": [ - { - "name": "Dept. of Biomedical Informatics, Dept. of Medicine, Vanderbilt University, 2525 West End Ave Ste.1475 Nashville, TN 37203, USA" - } - ] - } - ], - "member": "219", - "published-online": { - "date-parts": [ - [ - 2016, - 11, - 22 - ] - ] - }, - "container-title": "Biocomputing 2017", - "original-title": [], - "link": [ - { - "URL": "http://www.worldscientific.com/doi/pdf/10.1142/9789813207813_0050", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 7, - 13 - ] - ], - "date-time": "2017-07-13T00:18:16Z", - "timestamp": 1499905096000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 11, - 22 - ] - ] - }, - "references-count": 0, - "alternative-id": [ - "10.1142/9789813207813_0050", - "10.1142/10388" - ], - "URL": "https://doi.org/10.1142/9789813207813_0050", - "relation": {}, - "id": "qe90c1CL" - }, - "citation_id": "qe90c1CL" - }, - "doi:10.1142/9789814644730_0014": { - "source": "doi", - "identifer": "10.1142/9789814644730_0014", - "standard_citation": "doi:10.1142/9789814644730_0014", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T14:04:11Z", - "timestamp": 1502287451217 - }, - "reference-count": 0, - "publisher": "WORLD SCIENTIFIC", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2014, - 11 - ] - ] - }, - "DOI": "10.1142/9789814644730_0014", - "type": "paper-conference", - "created": { - "date-parts": [ - [ - 2014, - 11, - 11 - ] - ], - "date-time": "2014-11-11T07:36:26Z", - "timestamp": 1415691386000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "UNSUPERVISED FEATURE CONSTRUCTION AND KNOWLEDGE EXTRACTION FROM GENOME-WIDE ASSAYS OF BREAST CANCER WITH DENOISING AUTOENCODERS", - "prefix": "10.1142", - "author": [ - { - "given": "JIE", - "family": "TAN", - "affiliation": [ - { - "name": "Department of Genetics, Institute for Quantitative Biomedical Sciences, Norris Cotton Cancer Center, The Geisel School of Medicine at Dartmouth, Hanover, NH 03755, USA" - } - ] - }, - { - "given": "MATTHEW", - "family": "UNG", - "affiliation": [ - { - "name": "Department of Genetics, Institute for Quantitative Biomedical Sciences, Norris Cotton Cancer Center, The Geisel School of Medicine at Dartmouth, Hanover, NH 03755, USA" - } - ] - }, - { - "given": "CHAO", - "family": "CHENG", - "affiliation": [ - { - "name": "Department of Genetics, Institute for Quantitative Biomedical Sciences, Norris Cotton Cancer Center, The Geisel School of Medicine at Dartmouth, Hanover, NH 03755, USA" - } - ] - }, - { - "given": "CASEY S", - "family": "GREENE", - "affiliation": [ - { - "name": "Department of Genetics, Institute for Quantitative Biomedical Sciences, Norris Cotton Cancer Center, The Geisel School of Medicine at Dartmouth, Hanover, NH 03755, USA" - } - ] - } - ], - "member": "219", - "published-online": { - "date-parts": [ - [ - 2014, - 11, - 10 - ] - ] - }, - "container-title": "Biocomputing 2015", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 7, - 12 - ] - ], - "date-time": "2017-07-12T18:06:23Z", - "timestamp": 1499882783000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 11 - ] - ] - }, - "references-count": 0, - "alternative-id": [ - "10.1142/9789814644730_0014", - "10.1142/9455" - ], - "URL": "https://doi.org/10.1142/9789814644730_0014", - "relation": {}, - "id": "PBiRSdXv" - }, - "citation_id": "PBiRSdXv" - }, - "doi:10.1145/1143844.1143874": { - "source": "doi", - "identifer": "10.1145/1143844.1143874", - "standard_citation": "doi:10.1145/1143844.1143874", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T08:42:17Z", - "timestamp": 1507106537605 - }, - "publisher-location": "New York, New York, USA", - "reference-count": 0, - "publisher": "ACM Press", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2006 - ] - ] - }, - "DOI": "10.1145/1143844.1143874", - "type": "paper-conference", - "created": { - "date-parts": [ - [ - 2006, - 7, - 24 - ] - ], - "date-time": "2006-07-24T16:53:01Z", - "timestamp": 1153759981000 - }, - "source": "Crossref", - "is-referenced-by-count": 516, - "title": "The relationship between Precision-Recall and ROC curves", - "prefix": "10.1145", - "author": [ - { - "given": "Jesse", - "family": "Davis", - "affiliation": [] - }, - { - "given": "Mark", - "family": "Goadrich", - "affiliation": [] - } - ], - "member": "320", - "container-title": "Proceedings of the 23rd international conference on Machine learning - ICML '06", - "original-title": [], - "link": [ - { - "URL": "http://dl.acm.org/ft_gateway.cfm?id=1143874&ftid=364254&dwn=1", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 12, - 16 - ] - ], - "date-time": "2016-12-16T17:56:58Z", - "timestamp": 1481911018000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2006 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1145/1143844.1143874", - "relation": {}, - "id": "JNnkm5Zt" - }, - "citation_id": "JNnkm5Zt" - }, - "doi:10.1145/2382577.2382579": { - "source": "doi", - "identifer": "10.1145/2382577.2382579", - "standard_citation": "doi:10.1145/2382577.2382579", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T20:40:15Z", - "timestamp": 1502397615011 - }, - "reference-count": 0, - "publisher": "Association for Computing Machinery (ACM)", - "issue": "4", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2012, - 12, - 1 - ] - ] - }, - "DOI": "10.1145/2382577.2382579", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2013, - 1, - 2 - ] - ], - "date-time": "2013-01-02T13:23:15Z", - "timestamp": 1357132995000 - }, - "page": "1-21", - "source": "Crossref", - "is-referenced-by-count": 5, - "title": "Leakage in data mining", - "prefix": "10.1145", - "volume": "6", - "author": [ - { - "given": "Shachar", - "family": "Kaufman", - "affiliation": [] - }, - { - "given": "Saharon", - "family": "Rosset", - "affiliation": [] - }, - { - "given": "Claudia", - "family": "Perlich", - "affiliation": [] - }, - { - "given": "Ori", - "family": "Stitelman", - "affiliation": [] - } - ], - "member": "320", - "container-title": "ACM Transactions on Knowledge Discovery from Data", - "original-title": [], - "link": [ - { - "URL": "http://dl.acm.org/ft_gateway.cfm?id=2382579&ftid=1328043&dwn=1", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 12, - 6 - ] - ], - "date-time": "2016-12-06T23:59:09Z", - "timestamp": 1481068749000 - }, - "score": 1.0, - "subtitle": [ - "Formulation, detection, and avoidance" - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2012, - 12, - 1 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1145/2382577.2382579", - "relation": {}, - "container-title-short": "ACM Trans. Knowl. Discov. Data", - "id": "889LsjDi" - }, - "citation_id": "889LsjDi" - }, - "doi:10.1145/2661829.2661974": { - "source": "doi", - "identifer": "10.1145/2661829.2661974", - "standard_citation": "doi:10.1145/2661829.2661974", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T13:57:45Z", - "timestamp": 1502287065125 - }, - "publisher-location": "New York, New York, USA", - "reference-count": 17, - "publisher": "ACM Press", - "license": [ - { - "URL": "http://www.acm.org/publications/policies/copyright_policy#Background", - "start": { - "date-parts": [ - [ - 2014, - 11, - 3 - ] - ], - "date-time": "2014-11-03T00:00:00Z", - "timestamp": 1414972800000 - }, - "delay-in-days": 306, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2014 - ] - ] - }, - "DOI": "10.1145/2661829.2661974", - "type": "paper-conference", - "created": { - "date-parts": [ - [ - 2014, - 11, - 7 - ] - ], - "date-time": "2014-11-07T17:10:54Z", - "timestamp": 1415380254000 - }, - "source": "Crossref", - "is-referenced-by-count": 6, - "title": "Medical Semantic Similarity with a Neural Language Model", - "prefix": "10.1145", - "author": [ - { - "given": "Lance", - "family": "De Vine", - "affiliation": [] - }, - { - "given": "Guido", - "family": "Zuccon", - "affiliation": [] - }, - { - "given": "Bevan", - "family": "Koopman", - "affiliation": [] - }, - { - "given": "Laurianne", - "family": "Sitbon", - "affiliation": [] - }, - { - "given": "Peter", - "family": "Bruza", - "affiliation": [] - } - ], - "member": "320", - "container-title": "Proceedings of the 23rd ACM International Conference on Conference on Information and Knowledge Management - CIKM '14", - "original-title": [], - "link": [ - { - "URL": "http://dl.acm.org/ft_gateway.cfm?id=2661974&ftid=1511521&dwn=1", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 22 - ] - ], - "date-time": "2017-06-22T23:48:07Z", - "timestamp": 1498175287000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014 - ] - ] - }, - "references-count": 17, - "URL": "https://doi.org/10.1145/2661829.2661974", - "relation": { - "cites": [] - }, - "id": "XQtuRkTU" - }, - "citation_id": "XQtuRkTU" - }, - "doi:10.1145/2939672.2939823": { - "source": "doi", - "identifer": "10.1145/2939672.2939823", - "standard_citation": "doi:10.1145/2939672.2939823", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T04:02:37Z", - "timestamp": 1507003357715 - }, - "publisher-location": "New York, New York, USA", - "reference-count": 39, - "publisher": "ACM Press", - "license": [ - { - "URL": "http://www.acm.org/publications/policies/copyright_policy#Background", - "start": { - "date-parts": [ - [ - 2016, - 8, - 13 - ] - ], - "date-time": "2016-08-13T00:00:00Z", - "timestamp": 1471046400000 - }, - "delay-in-days": 225, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/100006785", - "name": "Google", - "doi-asserted-by": "publisher", - "award": [] - }, - { - "DOI": "10.13039/100000001", - "name": "National Science Foundation", - "doi-asserted-by": "publisher", - "award": [ - "1418511" - ] - }, - { - "name": "Samsung Scholarship", - "award": [] - }, - { - "name": "Centers for Disease Control and Prevention", - "award": [] - }, - { - "name": "Children's Healthcare of Atlanta", - "award": [] - }, - { - "name": "UCB", - "award": [] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016 - ] - ] - }, - "DOI": "10.1145/2939672.2939823", - "type": "paper-conference", - "created": { - "date-parts": [ - [ - 2016, - 8, - 8 - ] - ], - "date-time": "2016-08-08T18:33:46Z", - "timestamp": 1470681226000 - }, - "source": "Crossref", - "is-referenced-by-count": 5, - "title": "Multi-layer Representation Learning for Medical Concepts", - "prefix": "10.1145", - "author": [ - { - "given": "Edward", - "family": "Choi", - "affiliation": [ - { - "name": "Georgia Institute of Technology, Atlanta, GA, USA" - } - ] - }, - { - "given": "Mohammad Taha", - "family": "Bahadori", - "affiliation": [ - { - "name": "Georgia Institute of Technology, Atlanta, GA, USA" - } - ] - }, - { - "given": "Elizabeth", - "family": "Searles", - "affiliation": [ - { - "name": "Children's Healthcare of Atlanta, Atlanta, GA, USA" - } - ] - }, - { - "given": "Catherine", - "family": "Coffey", - "affiliation": [ - { - "name": "Children's Healthcare of Atlanta, Atlanta, GA, USA" - } - ] - }, - { - "given": "Michael", - "family": "Thompson", - "affiliation": [ - { - "name": "Children's Healthcare of Atlanta, Atlanta, GA, USA" - } - ] - }, - { - "given": "James", - "family": "Bost", - "affiliation": [ - { - "name": "Children's Healthcare of Atlanta, Atlanta, GA, USA" - } - ] - }, - { - "given": "Javier", - "family": "Tejedor-Sojo", - "affiliation": [ - { - "name": "Children's Healthcare of Atlanta, Atlanta, GA, USA" - } - ] - }, - { - "given": "Jimeng", - "family": "Sun", - "affiliation": [ - { - "name": "Georgia Institute of Technology, Atlanta, GA, USA" - } - ] - } - ], - "member": "320", - "container-title": "Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining - KDD '16", - "original-title": [], - "link": [ - { - "URL": "http://dl.acm.org/ft_gateway.cfm?id=2939823&ftid=1776021&dwn=1", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T20:13:44Z", - "timestamp": 1498335224000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016 - ] - ] - }, - "references-count": 39, - "URL": "https://doi.org/10.1145/2939672.2939823", - "relation": { - "cites": [] - }, - "id": "TwvauiTv" - }, - "citation_id": "TwvauiTv" - }, - "doi:10.1148/radiol.2017162326": { - "source": "doi", - "identifer": "10.1148/radiol.2017162326", - "standard_citation": "doi:10.1148/radiol.2017162326", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 1 - ] - ], - "date-time": "2017-09-01T13:02:12Z", - "timestamp": 1504270932974 - }, - "reference-count": 32, - "publisher": "Radiological Society of North America (RSNA)", - "issue": "2", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2017, - 8 - ] - ] - }, - "DOI": "10.1148/radiol.2017162326", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2017, - 4, - 24 - ] - ], - "date-time": "2017-04-24T13:26:04Z", - "timestamp": 1493040364000 - }, - "page": "574-582", - "source": "Crossref", - "is-referenced-by-count": 2, - "title": "Deep Learning at Chest Radiography: Automated Classification of Pulmonary Tuberculosis by Using Convolutional Neural Networks", - "prefix": "10.1148", - "volume": "284", - "author": [ - { - "given": "Paras", - "family": "Lakhani", - "affiliation": [ - { - "name": "From the Department of Radiology, Thomas Jefferson University Hospital, Sidney Kimmel Jefferson Medical College, 132 S 10th St, Room 1080A, Main Building, Philadelphia, PA 19107-5244." - } - ] - }, - { - "given": "Baskaran", - "family": "Sundaram", - "affiliation": [ - { - "name": "From the Department of Radiology, Thomas Jefferson University Hospital, Sidney Kimmel Jefferson Medical College, 132 S 10th St, Room 1080A, Main Building, Philadelphia, PA 19107-5244." - } - ] - } - ], - "member": "1577", - "container-title": "Radiology", - "original-title": [], - "link": [ - { - "URL": "http://pubs.rsna.org/doi/pdf/10.1148/radiol.2017162326", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 7, - 19 - ] - ], - "date-time": "2017-07-19T16:55:19Z", - "timestamp": 1500483319000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 8 - ] - ] - }, - "references-count": 32, - "alternative-id": [ - "10.1148/radiol.2017162326" - ], - "URL": "https://doi.org/10.1148/radiol.2017162326", - "relation": {}, - "subject": [ - "Radiology Nuclear Medicine and imaging" - ], - "container-title-short": "Radiology", - "id": "Qve94Jra" - }, - "citation_id": "Qve94Jra" - }, - "doi:10.1155/2013/769639": { - "source": "doi", - "identifer": "10.1155/2013/769639", - "standard_citation": "doi:10.1155/2013/769639", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T07:20:08Z", - "timestamp": 1502263208771 - }, - "reference-count": 41, - "publisher": "Hindawi Limited", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/3.0/", - "start": { - "date-parts": [ - [ - 2013, - 1, - 1 - ] - ], - "date-time": "2013-01-01T00:00:00Z", - "timestamp": 1356998400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2013 - ] - ] - }, - "abstract": "The volumes of current patient data as well as their complexity make clinical decision making more challenging than ever for physicians and other care givers. This situation calls for the use of biomedical informatics methods to process data and form recommendations and/or predictions to assist such decision makers. The design, implementation, and use of biomedical informatics systems in the form of computer-aided decision support have become essential and widely used over the last two decades. This paper provides a brief review of such systems, their application protocols and methodologies, and the future challenges and directions they suggest.", - "DOI": "10.1155/2013/769639", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2013, - 2, - 4 - ] - ], - "date-time": "2013-02-04T21:05:29Z", - "timestamp": 1360011929000 - }, - "page": "1-8", - "source": "Crossref", - "is-referenced-by-count": 13, - "title": "Biomedical Informatics for Computer-Aided Decision Support Systems: A Survey", - "prefix": "10.1155", - "volume": "2013", - "author": [ - { - "ORCID": "http://orcid.org/0000-0002-8084-3563", - "authenticated-orcid": true, - "given": "Ashwin", - "family": "Belle", - "affiliation": [ - { - "name": "Department of Computer Science, Virginia Commonwealth University, Richmond, VA 23284, USA" - } - ] - }, - { - "ORCID": "http://orcid.org/0000-0001-5902-9412", - "authenticated-orcid": true, - "given": "Mark A.", - "family": "Kon", - "affiliation": [ - { - "name": "Department of Mathematics and Statistics, Boston University, Boston, MA 02215, USA" - } - ] - }, - { - "ORCID": "http://orcid.org/0000-0003-4485-6612", - "authenticated-orcid": true, - "given": "Kayvan", - "family": "Najarian", - "affiliation": [ - { - "name": "Department of Computer Science, Virginia Commonwealth University, Richmond, VA 23284, USA" - } - ] - } - ], - "member": "98", - "container-title": "The Scientific World Journal", - "original-title": [], - "link": [ - { - "URL": "http://downloads.hindawi.com/journals/tswj/2013/769639.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://downloads.hindawi.com/journals/tswj/2013/769639.xml", - "content-type": "application/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://downloads.hindawi.com/journals/tswj/2013/769639.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 21 - ] - ], - "date-time": "2017-06-21T06:55:55Z", - "timestamp": 1498028155000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2013 - ] - ] - }, - "references-count": 41, - "alternative-id": [ - "769639", - "769639" - ], - "URL": "https://doi.org/10.1155/2013/769639", - "relation": { - "cites": [] - }, - "subject": [ - "General Biochemistry, Genetics and Molecular Biology", - "General Environmental Science", - "General Medicine" - ], - "container-title-short": "The Scientific World Journal", - "id": "3JyJ3DTh" - }, - "citation_id": "3JyJ3DTh" - }, - "doi:10.1158/1078-0432.ccr-13-0583": { - "source": "doi", - "identifer": "10.1158/1078-0432.ccr-13-0583", - "standard_citation": "doi:10.1158/1078-0432.ccr-13-0583", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 20 - ] - ], - "date-time": "2017-09-20T17:42:08Z", - "timestamp": 1505929328748 - }, - "reference-count": 0, - "publisher": "American Association for Cancer Research (AACR)", - "issue": "4", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2014, - 2, - 15 - ] - ] - }, - "DOI": "10.1158/1078-0432.ccr-13-0583", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 2, - 17 - ] - ], - "date-time": "2014-02-17T05:33:27Z", - "timestamp": 1392615207000 - }, - "page": "782-790", - "source": "Crossref", - "is-referenced-by-count": 66, - "title": "New Strategies for Triple-Negative Breast Cancer--Deciphering the Heterogeneity", - "prefix": "10.1158", - "volume": "20", - "author": [ - { - "given": "I. A.", - "family": "Mayer", - "affiliation": [] - }, - { - "given": "V. G.", - "family": "Abramson", - "affiliation": [] - }, - { - "given": "B. D.", - "family": "Lehmann", - "affiliation": [] - }, - { - "given": "J. A.", - "family": "Pietenpol", - "affiliation": [] - } - ], - "member": "1086", - "published-online": { - "date-parts": [ - [ - 2014, - 2, - 16 - ] - ] - }, - "container-title": "Clinical Cancer Research", - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1158/1078-0432.CCR-13-0583", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 12, - 23 - ] - ], - "date-time": "2016-12-23T12:06:01Z", - "timestamp": 1482494761000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 2, - 15 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1158/1078-0432.ccr-13-0583", - "relation": {}, - "subject": [ - "Cancer Research", - "Oncology" - ], - "container-title-short": "Clinical Cancer Research", - "id": "pEIw87Mp" - }, - "citation_id": "pEIw87Mp" - }, - "doi:10.1186/1758-2946-5-30": { - "source": "doi", - "identifer": "10.1186/1758-2946-5-30", - "standard_citation": "doi:10.1186/1758-2946-5-30", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 16 - ] - ], - "date-time": "2017-08-16T16:23:10Z", - "timestamp": 1502900590848 - }, - "reference-count": 0, - "publisher": "Springer Nature", - "issue": "1", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2013 - ] - ] - }, - "DOI": "10.1186/1758-2946-5-30", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2013, - 6, - 22 - ] - ], - "date-time": "2013-06-22T10:14:24Z", - "timestamp": 1371896064000 - }, - "page": "30", - "source": "Crossref", - "is-referenced-by-count": 40, - "title": "Drug Repositioning: A Machine-Learning Approach through Data Integration", - "prefix": "10.1186", - "volume": "5", - "author": [ - { - "given": "Francesco", - "family": "Napolitano", - "affiliation": [] - }, - { - "given": "Yan", - "family": "Zhao", - "affiliation": [] - }, - { - "given": "Vania M", - "family": "Moreira", - "affiliation": [] - }, - { - "given": "Roberto", - "family": "Tagliaferri", - "affiliation": [] - }, - { - "given": "Juha", - "family": "Kere", - "affiliation": [] - }, - { - "given": "Mauro", - "family": "D'Amato", - "affiliation": [] - }, - { - "given": "Dario", - "family": "Greco", - "affiliation": [] - } - ], - "member": "297", - "container-title": "Journal of Cheminformatics", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2016, - 5, - 16 - ] - ], - "date-time": "2016-05-16T16:51:17Z", - "timestamp": 1463417477000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2013 - ] - ] - }, - "references-count": 0, - "alternative-id": [ - "1758-2946-5-30" - ], - "URL": "https://doi.org/10.1186/1758-2946-5-30", - "relation": {}, - "subject": [ - "Physical and Theoretical Chemistry", - "Library and Information Sciences", - "Computer Graphics and Computer-Aided Design", - "Computer Science Applications" - ], - "container-title-short": "Journal of Cheminformatics", - "id": "M1EW8Rfl" - }, - "citation_id": "M1EW8Rfl" - }, - "doi:10.1186/s12859-015-0845-0": { - "source": "doi", - "identifer": "10.1186/s12859-015-0845-0", - "standard_citation": "doi:10.1186/s12859-015-0845-0", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 11 - ] - ], - "date-time": "2017-08-11T07:57:40Z", - "timestamp": 1502438260936 - }, - "reference-count": 60, - "publisher": "Springer Nature", - "issue": "S1", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2016, - 1, - 11 - ] - ], - "date-time": "2016-01-11T00:00:00Z", - "timestamp": 1452470400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "link.springer.com" - ], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 12 - ] - ] - }, - "DOI": "10.1186/s12859-015-0845-0", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 1, - 11 - ] - ], - "date-time": "2016-01-11T10:00:45Z", - "timestamp": 1452506445000 - }, - "update-policy": "http://dx.doi.org/10.1007/springer_crossmark_policy", - "source": "Crossref", - "is-referenced-by-count": 2, - "title": "Drug repositioning for non-small cell lung cancer by using machine learning algorithms and topological graph theory", - "prefix": "10.1186", - "volume": "17", - "author": [ - { - "given": "Chien-Hung", - "family": "Huang", - "affiliation": [] - }, - { - "given": "Peter Mu-Hsin", - "family": "Chang", - "affiliation": [] - }, - { - "given": "Chia-Wei", - "family": "Hsu", - "affiliation": [] - }, - { - "given": "Chi-Ying F.", - "family": "Huang", - "affiliation": [] - }, - { - "given": "Ka-Lok", - "family": "Ng", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2016, - 1, - 11 - ] - ] - }, - "container-title": "BMC Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1186/s12859-015-0845-0.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://link.springer.com/article/10.1186/s12859-015-0845-0/fulltext.html", - "content-type": "text/html", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://link.springer.com/content/pdf/10.1186/s12859-015-0845-0", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T03:25:57Z", - "timestamp": 1498274757000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 1, - 11 - ] - ] - }, - "references-count": 60, - "alternative-id": [ - "845" - ], - "URL": "https://doi.org/10.1186/s12859-015-0845-0", - "relation": { - "cites": [] - }, - "subject": [ - "Biochemistry", - "Applied Mathematics", - "Molecular Biology", - "Structural Biology", - "Computer Science Applications" - ], - "container-title-short": "BMC Bioinformatics", - "article-number": "S2", - "id": "18lqFDKRR" - }, - "citation_id": "18lqFDKRR" - }, - "doi:10.1186/s12859-017-1561-8": { - "source": "doi", - "identifer": "10.1186/s12859-017-1561-8", - "standard_citation": "doi:10.1186/s12859-017-1561-8", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 11 - ] - ], - "date-time": "2017-09-11T10:02:12Z", - "timestamp": 1505124132078 - }, - "reference-count": 51, - "publisher": "Springer Nature", - "issue": "1", - "funder": [ - { - "DOI": "10.13039/501100003399", - "name": "Science and Technology Commission of Shanghai Municipality", - "doi-asserted-by": "publisher", - "award": [ - "16JC1404300" - ] - }, - { - "DOI": "10.13039/501100001809", - "name": "National Natural Science Foundation of China", - "doi-asserted-by": "publisher", - "award": [ - "31628003", - "61671288" - ] - } - ], - "content-domain": { - "domain": [ - "link.springer.com" - ], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2017, - 12 - ] - ] - }, - "DOI": "10.1186/s12859-017-1561-8", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2017, - 2, - 28 - ] - ], - "date-time": "2017-02-28T14:18:56Z", - "timestamp": 1488291536000 - }, - "update-policy": "http://dx.doi.org/10.1007/springer_crossmark_policy", - "source": "Crossref", - "is-referenced-by-count": 1, - "title": "RNA-protein binding motifs mining with a new hybrid deep learning based cross-domain knowledge integration approach", - "prefix": "10.1186", - "volume": "18", - "author": [ - { - "given": "Xiaoyong", - "family": "Pan", - "affiliation": [] - }, - { - "given": "Hong-Bin", - "family": "Shen", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2017, - 2, - 28 - ] - ] - }, - "container-title": "BMC Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1186/s12859-017-1561-8.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T11:33:45Z", - "timestamp": 1498390425000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 2, - 28 - ] - ] - }, - "references-count": 51, - "alternative-id": [ - "1561" - ], - "URL": "https://doi.org/10.1186/s12859-017-1561-8", - "relation": { - "has-preprint": [ - { - "id-type": "doi", - "id": "10.1101/085191", - "asserted-by": "object" - } - ], - "cites": [] - }, - "subject": [ - "Biochemistry", - "Applied Mathematics", - "Molecular Biology", - "Structural Biology", - "Computer Science Applications" - ], - "container-title-short": "BMC Bioinformatics", - "article-number": "136", - "id": "qnKdqG0P" - }, - "citation_id": "qnKdqG0P" - }, - "doi:10.1200/jco.2008.18.1370": { - "source": "doi", - "identifer": "10.1200/jco.2008.18.1370", - "standard_citation": "doi:10.1200/jco.2008.18.1370", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T07:22:12Z", - "timestamp": 1507015332880 - }, - "reference-count": 29, - "publisher": "American Society of Clinical Oncology (ASCO)", - "issue": "8", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2009, - 3, - 10 - ] - ] - }, - "DOI": "10.1200/jco.2008.18.1370", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2009, - 2, - 10 - ] - ], - "date-time": "2009-02-10T02:09:42Z", - "timestamp": 1234231782000 - }, - "page": "1160-1167", - "source": "Crossref", - "is-referenced-by-count": 1184, - "title": "Supervised Risk Predictor of Breast Cancer Based on Intrinsic Subtypes", - "prefix": "10.1200", - "volume": "27", - "author": [ - { - "given": "Joel S.", - "family": "Parker", - "affiliation": [ - { - "name": "From the Lineberger Comprehensive Cancer Center and Departments of Genetics, Pathology and Laboratory Medicine, and Department of Statistics and Operations Research, Carolina Center for Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC; Department of Pathology, University of Utah Health Sciences Center; ARUP Institute for Clinical and Experimental Pathology, Salt Lake City, UT; Genetic Pathology Evaluation Centre, Department of Pathology, Vancouver Coastal Health Research..." - } - ] - }, - { - "given": "Michael", - "family": "Mullins", - "affiliation": [ - { - "name": "From the Lineberger Comprehensive Cancer Center and Departments of Genetics, Pathology and Laboratory Medicine, and Department of Statistics and Operations Research, Carolina Center for Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC; Department of Pathology, University of Utah Health Sciences Center; ARUP Institute for Clinical and Experimental Pathology, Salt Lake City, UT; Genetic Pathology Evaluation Centre, Department of Pathology, Vancouver Coastal Health Research..." - } - ] - }, - { - "given": "Maggie C.U.", - "family": "Cheang", - "affiliation": [ - { - "name": "From the Lineberger Comprehensive Cancer Center and Departments of Genetics, Pathology and Laboratory Medicine, and Department of Statistics and Operations Research, Carolina Center for Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC; Department of Pathology, University of Utah Health Sciences Center; ARUP Institute for Clinical and Experimental Pathology, Salt Lake City, UT; Genetic Pathology Evaluation Centre, Department of Pathology, Vancouver Coastal Health Research..." - } - ] - }, - { - "given": "Samuel", - "family": "Leung", - "affiliation": [ - { - "name": "From the Lineberger Comprehensive Cancer Center and Departments of Genetics, Pathology and Laboratory Medicine, and Department of Statistics and Operations Research, Carolina Center for Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC; Department of Pathology, University of Utah Health Sciences Center; ARUP Institute for Clinical and Experimental Pathology, Salt Lake City, UT; Genetic Pathology Evaluation Centre, Department of Pathology, Vancouver Coastal Health Research..." - } - ] - }, - { - "given": "David", - "family": "Voduc", - "affiliation": [ - { - "name": "From the Lineberger Comprehensive Cancer Center and Departments of Genetics, Pathology and Laboratory Medicine, and Department of Statistics and Operations Research, Carolina Center for Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC; Department of Pathology, University of Utah Health Sciences Center; ARUP Institute for Clinical and Experimental Pathology, Salt Lake City, UT; Genetic Pathology Evaluation Centre, Department of Pathology, Vancouver Coastal Health Research..." - } - ] - }, - { - "given": "Tammi", - "family": "Vickery", - "affiliation": [ - { - "name": "From the Lineberger Comprehensive Cancer Center and Departments of Genetics, Pathology and Laboratory Medicine, and Department of Statistics and Operations Research, Carolina Center for Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC; Department of Pathology, University of Utah Health Sciences Center; ARUP Institute for Clinical and Experimental Pathology, Salt Lake City, UT; Genetic Pathology Evaluation Centre, Department of Pathology, Vancouver Coastal Health Research..." - } - ] - }, - { - "given": "Sherri", - "family": "Davies", - "affiliation": [ - { - "name": "From the Lineberger Comprehensive Cancer Center and Departments of Genetics, Pathology and Laboratory Medicine, and Department of Statistics and Operations Research, Carolina Center for Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC; Department of Pathology, University of Utah Health Sciences Center; ARUP Institute for Clinical and Experimental Pathology, Salt Lake City, UT; Genetic Pathology Evaluation Centre, Department of Pathology, Vancouver Coastal Health Research..." - } - ] - }, - { - "given": "Christiane", - "family": "Fauron", - "affiliation": [ - { - "name": "From the Lineberger Comprehensive Cancer Center and Departments of Genetics, Pathology and Laboratory Medicine, and Department of Statistics and Operations Research, Carolina Center for Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC; Department of Pathology, University of Utah Health Sciences Center; ARUP Institute for Clinical and Experimental Pathology, Salt Lake City, UT; Genetic Pathology Evaluation Centre, Department of Pathology, Vancouver Coastal Health Research..." - } - ] - }, - { - "given": "Xiaping", - "family": "He", - "affiliation": [ - { - "name": "From the Lineberger Comprehensive Cancer Center and Departments of Genetics, Pathology and Laboratory Medicine, and Department of Statistics and Operations Research, Carolina Center for Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC; Department of Pathology, University of Utah Health Sciences Center; ARUP Institute for Clinical and Experimental Pathology, Salt Lake City, UT; Genetic Pathology Evaluation Centre, Department of Pathology, Vancouver Coastal Health Research..." - } - ] - }, - { - "given": "Zhiyuan", - "family": "Hu", - "affiliation": [ - { - "name": "From the Lineberger Comprehensive Cancer Center and Departments of Genetics, Pathology and Laboratory Medicine, and Department of Statistics and Operations Research, Carolina Center for Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC; Department of Pathology, University of Utah Health Sciences Center; ARUP Institute for Clinical and Experimental Pathology, Salt Lake City, UT; Genetic Pathology Evaluation Centre, Department of Pathology, Vancouver Coastal Health Research..." - } - ] - }, - { - "given": "John F.", - "family": "Quackenbush", - "affiliation": [ - { - "name": "From the Lineberger Comprehensive Cancer Center and Departments of Genetics, Pathology and Laboratory Medicine, and Department of Statistics and Operations Research, Carolina Center for Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC; Department of Pathology, University of Utah Health Sciences Center; ARUP Institute for Clinical and Experimental Pathology, Salt Lake City, UT; Genetic Pathology Evaluation Centre, Department of Pathology, Vancouver Coastal Health Research..." - } - ] - }, - { - "given": "Inge J.", - "family": "Stijleman", - "affiliation": [ - { - "name": "From the Lineberger Comprehensive Cancer Center and Departments of Genetics, Pathology and Laboratory Medicine, and Department of Statistics and Operations Research, Carolina Center for Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC; Department of Pathology, University of Utah Health Sciences Center; ARUP Institute for Clinical and Experimental Pathology, Salt Lake City, UT; Genetic Pathology Evaluation Centre, Department of Pathology, Vancouver Coastal Health Research..." - } - ] - }, - { - "given": "Juan", - "family": "Palazzo", - "affiliation": [ - { - "name": "From the Lineberger Comprehensive Cancer Center and Departments of Genetics, Pathology and Laboratory Medicine, and Department of Statistics and Operations Research, Carolina Center for Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC; Department of Pathology, University of Utah Health Sciences Center; ARUP Institute for Clinical and Experimental Pathology, Salt Lake City, UT; Genetic Pathology Evaluation Centre, Department of Pathology, Vancouver Coastal Health Research..." - } - ] - }, - { - "given": "J.S.", - "family": "Marron", - "affiliation": [ - { - "name": "From the Lineberger Comprehensive Cancer Center and Departments of Genetics, Pathology and Laboratory Medicine, and Department of Statistics and Operations Research, Carolina Center for Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC; Department of Pathology, University of Utah Health Sciences Center; ARUP Institute for Clinical and Experimental Pathology, Salt Lake City, UT; Genetic Pathology Evaluation Centre, Department of Pathology, Vancouver Coastal Health Research..." - } - ] - }, - { - "given": "Andrew B.", - "family": "Nobel", - "affiliation": [ - { - "name": "From the Lineberger Comprehensive Cancer Center and Departments of Genetics, Pathology and Laboratory Medicine, and Department of Statistics and Operations Research, Carolina Center for Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC; Department of Pathology, University of Utah Health Sciences Center; ARUP Institute for Clinical and Experimental Pathology, Salt Lake City, UT; Genetic Pathology Evaluation Centre, Department of Pathology, Vancouver Coastal Health Research..." - } - ] - }, - { - "given": "Elaine", - "family": "Mardis", - "affiliation": [ - { - "name": "From the Lineberger Comprehensive Cancer Center and Departments of Genetics, Pathology and Laboratory Medicine, and Department of Statistics and Operations Research, Carolina Center for Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC; Department of Pathology, University of Utah Health Sciences Center; ARUP Institute for Clinical and Experimental Pathology, Salt Lake City, UT; Genetic Pathology Evaluation Centre, Department of Pathology, Vancouver Coastal Health Research..." - } - ] - }, - { - "given": "Torsten O.", - "family": "Nielsen", - "affiliation": [ - { - "name": "From the Lineberger Comprehensive Cancer Center and Departments of Genetics, Pathology and Laboratory Medicine, and Department of Statistics and Operations Research, Carolina Center for Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC; Department of Pathology, University of Utah Health Sciences Center; ARUP Institute for Clinical and Experimental Pathology, Salt Lake City, UT; Genetic Pathology Evaluation Centre, Department of Pathology, Vancouver Coastal Health Research..." - } - ] - }, - { - "given": "Matthew J.", - "family": "Ellis", - "affiliation": [ - { - "name": "From the Lineberger Comprehensive Cancer Center and Departments of Genetics, Pathology and Laboratory Medicine, and Department of Statistics and Operations Research, Carolina Center for Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC; Department of Pathology, University of Utah Health Sciences Center; ARUP Institute for Clinical and Experimental Pathology, Salt Lake City, UT; Genetic Pathology Evaluation Centre, Department of Pathology, Vancouver Coastal Health Research..." - } - ] - }, - { - "given": "Charles M.", - "family": "Perou", - "affiliation": [ - { - "name": "From the Lineberger Comprehensive Cancer Center and Departments of Genetics, Pathology and Laboratory Medicine, and Department of Statistics and Operations Research, Carolina Center for Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC; Department of Pathology, University of Utah Health Sciences Center; ARUP Institute for Clinical and Experimental Pathology, Salt Lake City, UT; Genetic Pathology Evaluation Centre, Department of Pathology, Vancouver Coastal Health Research..." - } - ] - }, - { - "given": "Philip S.", - "family": "Bernard", - "affiliation": [ - { - "name": "From the Lineberger Comprehensive Cancer Center and Departments of Genetics, Pathology and Laboratory Medicine, and Department of Statistics and Operations Research, Carolina Center for Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC; Department of Pathology, University of Utah Health Sciences Center; ARUP Institute for Clinical and Experimental Pathology, Salt Lake City, UT; Genetic Pathology Evaluation Centre, Department of Pathology, Vancouver Coastal Health Research..." - } - ] - } - ], - "member": "233", - "container-title": "Journal of Clinical Oncology", - "original-title": [], - "link": [ - { - "URL": "http://ascopubs.org/doi/pdf/10.1200/JCO.2008.18.1370", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 4, - 3 - ] - ], - "date-time": "2017-04-03T13:31:52Z", - "timestamp": 1491226312000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2009, - 3, - 10 - ] - ] - }, - "references-count": 29, - "alternative-id": [ - "10.1200/JCO.2008.18.1370" - ], - "URL": "https://doi.org/10.1200/jco.2008.18.1370", - "relation": {}, - "subject": [ - "Cancer Research", - "Oncology" - ], - "container-title-short": "JCO", - "id": "lnK82Ey6" - }, - "citation_id": "lnK82Ey6" - }, - "doi:10.1208/s12248-012-9322-0": { - "source": "doi", - "identifer": "10.1208/s12248-012-9322-0", - "standard_citation": "doi:10.1208/s12248-012-9322-0", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 1 - ] - ], - "date-time": "2017-09-01T10:22:12Z", - "timestamp": 1504261332733 - }, - "reference-count": 97, - "publisher": "American Association of Pharmaceutical Scientists (AAPS)", - "issue": "1", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2012, - 3 - ] - ] - }, - "DOI": "10.1208/s12248-012-9322-0", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2012, - 1, - 26 - ] - ], - "date-time": "2012-01-26T16:24:15Z", - "timestamp": 1327595055000 - }, - "page": "133-141", - "source": "Crossref", - "is-referenced-by-count": 145, - "title": "Structure-Based Virtual Screening for Drug Discovery: a Problem-Centric Review", - "prefix": "10.1208", - "volume": "14", - "author": [ - { - "given": "Tiejun", - "family": "Cheng", - "affiliation": [] - }, - { - "given": "Qingliang", - "family": "Li", - "affiliation": [] - }, - { - "given": "Zhigang", - "family": "Zhou", - "affiliation": [] - }, - { - "given": "Yanli", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Stephen H.", - "family": "Bryant", - "affiliation": [] - } - ], - "member": "225", - "published-online": { - "date-parts": [ - [ - 2012, - 1, - 27 - ] - ] - }, - "container-title": "The AAPS Journal", - "original-title": [], - "link": [ - { - "URL": "http://www.springerlink.com/index/pdf/10.1208/s12248-012-9322-0", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 20 - ] - ], - "date-time": "2017-06-20T13:22:17Z", - "timestamp": 1497964937000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2012, - 1, - 27 - ] - ] - }, - "references-count": 97, - "alternative-id": [ - "9322" - ], - "URL": "https://doi.org/10.1208/s12248-012-9322-0", - "relation": {}, - "container-title-short": "AAPS J", - "id": "13iyYvEcB" - }, - "citation_id": "13iyYvEcB" - }, - "doi:10.1371/journal.pbio.1001661": { - "source": "doi", - "identifer": "10.1371/journal.pbio.1001661", - "standard_citation": "doi:10.1371/journal.pbio.1001661", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 18 - ] - ], - "date-time": "2017-09-18T14:42:11Z", - "timestamp": 1505745731387 - }, - "reference-count": 31, - "publisher": "Public Library of Science (PLoS)", - "issue": "9", - "license": [ - { - "URL": "https://creativecommons.org/publicdomain/zero/1.0/", - "start": { - "date-parts": [ - [ - 2013, - 9, - 17 - ] - ], - "date-time": "2013-09-17T00:00:00Z", - "timestamp": 1379376000000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "www.plosbiology.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pbio.1001661", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2013, - 9, - 17 - ] - ], - "date-time": "2013-09-17T20:51:16Z", - "timestamp": 1379451076000 - }, - "page": "e1001661", - "update-policy": "http://dx.doi.org/10.1371/journal.pbio.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 48, - "title": "Generalization and Dilution of Association Results from European GWAS in Populations of Non-European Ancestry: The PAGE Study", - "prefix": "10.1371", - "volume": "11", - "author": [ - { - "given": "Christopher S.", - "family": "Carlson", - "affiliation": [] - }, - { - "given": "Tara C.", - "family": "Matise", - "affiliation": [] - }, - { - "given": "Kari E.", - "family": "North", - "affiliation": [] - }, - { - "given": "Christopher A.", - "family": "Haiman", - "affiliation": [] - }, - { - "given": "Megan D.", - "family": "Fesinmeyer", - "affiliation": [] - }, - { - "given": "Steven", - "family": "Buyske", - "affiliation": [] - }, - { - "given": "Fredrick R.", - "family": "Schumacher", - "affiliation": [] - }, - { - "given": "Ulrike", - "family": "Peters", - "affiliation": [] - }, - { - "given": "Nora", - "family": "Franceschini", - "affiliation": [] - }, - { - "given": "Marylyn D.", - "family": "Ritchie", - "affiliation": [] - }, - { - "given": "David J.", - "family": "Duggan", - "affiliation": [] - }, - { - "given": "Kylee L.", - "family": "Spencer", - "affiliation": [] - }, - { - "given": "Logan", - "family": "Dumitrescu", - "affiliation": [] - }, - { - "given": "Charles B.", - "family": "Eaton", - "affiliation": [] - }, - { - "given": "Fridtjof", - "family": "Thomas", - "affiliation": [] - }, - { - "given": "Alicia", - "family": "Young", - "affiliation": [] - }, - { - "given": "Cara", - "family": "Carty", - "affiliation": [] - }, - { - "given": "Gerardo", - "family": "Heiss", - "affiliation": [] - }, - { - "given": "Loic", - "family": "Le Marchand", - "affiliation": [] - }, - { - "given": "Dana C.", - "family": "Crawford", - "affiliation": [] - }, - { - "given": "Lucia A.", - "family": "Hindorff", - "affiliation": [] - }, - { - "given": "Charles L.", - "family": "Kooperberg", - "affiliation": [] - }, - { - "name": "for the PAGE Consortium", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2013, - 9, - 17 - ] - ] - }, - "container-title": "PLoS Biology", - "original-title": [], - "link": [ - { - "URL": "http://dx.plos.org/10.1371/journal.pbio.1001661", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 21 - ] - ], - "date-time": "2017-06-21T22:04:33Z", - "timestamp": 1498082673000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Greg", - "family": "Gibson", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2013, - 9, - 17 - ] - ] - }, - "references-count": 31, - "URL": "https://doi.org/10.1371/journal.pbio.1001661", - "relation": { - "cites": [] - }, - "subject": [ - "General Biochemistry, Genetics and Molecular Biology", - "General Immunology and Microbiology", - "General Neuroscience", - "General Agricultural and Biological Sciences" - ], - "container-title-short": "PLoS Biol", - "id": "dKwyEWWF" - }, - "citation_id": "dKwyEWWF" - }, - "doi:10.1371/journal.pbio.1002195": { - "source": "doi", - "identifer": "10.1371/journal.pbio.1002195", - "standard_citation": "doi:10.1371/journal.pbio.1002195", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 2 - ] - ], - "date-time": "2017-10-02T15:42:19Z", - "timestamp": 1506958939783 - }, - "reference-count": 52, - "publisher": "Public Library of Science (PLoS)", - "issue": "7", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2015, - 7, - 7 - ] - ], - "date-time": "2015-07-07T00:00:00Z", - "timestamp": 1436227200000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "www.plosbiology.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pbio.1002195", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 7, - 7 - ] - ], - "date-time": "2015-07-07T17:42:57Z", - "timestamp": 1436290977000 - }, - "page": "e1002195", - "update-policy": "http://dx.doi.org/10.1371/journal.pbio.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 135, - "title": "Big Data: Astronomical or Genomical?", - "prefix": "10.1371", - "volume": "13", - "author": [ - { - "given": "Zachary D.", - "family": "Stephens", - "affiliation": [] - }, - { - "given": "Skylar Y.", - "family": "Lee", - "affiliation": [] - }, - { - "given": "Faraz", - "family": "Faghri", - "affiliation": [] - }, - { - "given": "Roy H.", - "family": "Campbell", - "affiliation": [] - }, - { - "given": "Chengxiang", - "family": "Zhai", - "affiliation": [] - }, - { - "given": "Miles J.", - "family": "Efron", - "affiliation": [] - }, - { - "given": "Ravishankar", - "family": "Iyer", - "affiliation": [] - }, - { - "given": "Michael C.", - "family": "Schatz", - "affiliation": [] - }, - { - "given": "Saurabh", - "family": "Sinha", - "affiliation": [] - }, - { - "given": "Gene E.", - "family": "Robinson", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2015, - 7, - 7 - ] - ] - }, - "container-title": "PLOS Biology", - "original-title": [], - "link": [ - { - "URL": "http://dx.plos.org/10.1371/journal.pbio.1002195", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 23 - ] - ], - "date-time": "2017-06-23T15:16:20Z", - "timestamp": 1498230980000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 7, - 7 - ] - ] - }, - "references-count": 52, - "URL": "https://doi.org/10.1371/journal.pbio.1002195", - "relation": { - "cites": [] - }, - "subject": [ - "General Biochemistry, Genetics and Molecular Biology", - "General Immunology and Microbiology", - "General Neuroscience", - "General Agricultural and Biological Sciences" - ], - "container-title-short": "PLoS Biol", - "id": "13bxiY1vo" - }, - "citation_id": "13bxiY1vo" - }, - "doi:10.1371/journal.pcbi.1002957": { - "source": "doi", - "identifer": "10.1371/journal.pcbi.1002957", - "standard_citation": "doi:10.1371/journal.pcbi.1002957", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T10:23:39Z", - "timestamp": 1502274219005 - }, - "reference-count": 96, - "publisher": "Public Library of Science (PLoS)", - "issue": "3", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2013, - 3, - 14 - ] - ], - "date-time": "2013-03-14T00:00:00Z", - "timestamp": 1363219200000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "www.ploscompbiol.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pcbi.1002957", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2013, - 3, - 14 - ] - ], - "date-time": "2013-03-14T23:39:22Z", - "timestamp": 1363304362000 - }, - "page": "e1002957", - "update-policy": "http://dx.doi.org/10.1371/journal.pcbi.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 22, - "title": "Functional Knowledge Transfer for High-accuracy Prediction of Under-studied Biological Processes", - "prefix": "10.1371", - "volume": "9", - "author": [ - { - "given": "Christopher Y.", - "family": "Park", - "affiliation": [] - }, - { - "given": "Aaron K.", - "family": "Wong", - "affiliation": [] - }, - { - "given": "Casey S.", - "family": "Greene", - "affiliation": [] - }, - { - "given": "Jessica", - "family": "Rowland", - "affiliation": [] - }, - { - "given": "Yuanfang", - "family": "Guan", - "affiliation": [] - }, - { - "given": "Lars A.", - "family": "Bongo", - "affiliation": [] - }, - { - "given": "Rebecca D.", - "family": "Burdine", - "affiliation": [] - }, - { - "given": "Olga G.", - "family": "Troyanskaya", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2013, - 3, - 14 - ] - ] - }, - "container-title": "PLoS Computational Biology", - "original-title": [], - "link": [ - { - "URL": "http://dx.plos.org/10.1371/journal.pcbi.1002957", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 21 - ] - ], - "date-time": "2017-06-21T09:24:10Z", - "timestamp": 1498037050000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Greg", - "family": "Tucker-Kellogg", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2013, - 3, - 14 - ] - ] - }, - "references-count": 96, - "URL": "https://doi.org/10.1371/journal.pcbi.1002957", - "relation": { - "cites": [] - }, - "subject": [ - "Ecology", - "Modelling and Simulation", - "Computational Theory and Mathematics", - "Genetics", - "Ecology, Evolution, Behavior and Systematics", - "Molecular Biology", - "Cellular and Molecular Neuroscience" - ], - "container-title-short": "PLoS Comput Biol", - "id": "5tvnB4uW" - }, - "citation_id": "5tvnB4uW" - }, - "doi:10.1371/journal.pcbi.1003711": { - "source": "doi", - "identifer": "10.1371/journal.pcbi.1003711", - "standard_citation": "doi:10.1371/journal.pcbi.1003711", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T14:02:16Z", - "timestamp": 1507039336014 - }, - "reference-count": 38, - "publisher": "Public Library of Science (PLoS)", - "issue": "7", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2014, - 7, - 17 - ] - ], - "date-time": "2014-07-17T00:00:00Z", - "timestamp": 1405555200000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "www.ploscompbiol.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pcbi.1003711", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 7, - 17 - ] - ], - "date-time": "2014-07-17T18:43:05Z", - "timestamp": 1405622585000 - }, - "page": "e1003711", - "update-policy": "http://dx.doi.org/10.1371/journal.pcbi.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 51, - "title": "Enhanced Regulatory Sequence Prediction Using Gapped k-mer Features", - "prefix": "10.1371", - "volume": "10", - "author": [ - { - "given": "Mahmoud", - "family": "Ghandi", - "affiliation": [] - }, - { - "given": "Dongwon", - "family": "Lee", - "affiliation": [] - }, - { - "given": "Morteza", - "family": "Mohammad-Noori", - "affiliation": [] - }, - { - "given": "Michael A.", - "family": "Beer", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2014, - 7, - 17 - ] - ] - }, - "container-title": "PLoS Computational Biology", - "original-title": [], - "link": [ - { - "URL": "http://dx.plos.org/10.1371/journal.pcbi.1003711", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 22 - ] - ], - "date-time": "2017-06-22T16:55:15Z", - "timestamp": 1498150515000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Quaid", - "family": "Morris", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 7, - 17 - ] - ] - }, - "references-count": 38, - "URL": "https://doi.org/10.1371/journal.pcbi.1003711", - "relation": { - "cites": [] - }, - "subject": [ - "Ecology", - "Modelling and Simulation", - "Computational Theory and Mathematics", - "Genetics", - "Ecology, Evolution, Behavior and Systematics", - "Molecular Biology", - "Cellular and Molecular Neuroscience" - ], - "container-title-short": "PLoS Comput Biol", - "id": "JxuQvvyk" - }, - "citation_id": "JxuQvvyk" - }, - "doi:10.1371/journal.pcbi.1003889": { - "source": "doi", - "identifer": "10.1371/journal.pcbi.1003889", - "standard_citation": "doi:10.1371/journal.pcbi.1003889", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 29 - ] - ], - "date-time": "2017-09-29T17:02:14Z", - "timestamp": 1506704534033 - }, - "reference-count": 50, - "publisher": "Public Library of Science (PLoS)", - "issue": "11", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2014, - 11, - 6 - ] - ], - "date-time": "2014-11-06T00:00:00Z", - "timestamp": 1415232000000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "www.ploscompbiol.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pcbi.1003889", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 11, - 6 - ] - ], - "date-time": "2014-11-06T19:07:02Z", - "timestamp": 1415300822000 - }, - "page": "e1003889", - "update-policy": "http://dx.doi.org/10.1371/journal.pcbi.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 45, - "title": "Improved Contact Predictions Using the Recognition of Protein Like Contact Patterns", - "prefix": "10.1371", - "volume": "10", - "author": [ - { - "given": "Marcin J.", - "family": "Skwark", - "affiliation": [] - }, - { - "given": "Daniele", - "family": "Raimondi", - "affiliation": [] - }, - { - "given": "Mirco", - "family": "Michel", - "affiliation": [] - }, - { - "given": "Arne", - "family": "Elofsson", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2014, - 11, - 6 - ] - ] - }, - "container-title": "PLoS Computational Biology", - "original-title": [], - "link": [ - { - "URL": "http://dx.plos.org/10.1371/journal.pcbi.1003889", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 22 - ] - ], - "date-time": "2017-06-22T23:44:00Z", - "timestamp": 1498175040000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Guanghong", - "family": "Wei", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 11, - 6 - ] - ] - }, - "references-count": 50, - "URL": "https://doi.org/10.1371/journal.pcbi.1003889", - "relation": { - "cites": [] - }, - "subject": [ - "Ecology", - "Modelling and Simulation", - "Computational Theory and Mathematics", - "Genetics", - "Ecology, Evolution, Behavior and Systematics", - "Molecular Biology", - "Cellular and Molecular Neuroscience" - ], - "container-title-short": "PLoS Comput Biol", - "id": "F13xtRbV" - }, - "citation_id": "F13xtRbV" - }, - "doi:10.1371/journal.pcbi.1005135": { - "source": "doi", - "identifer": "10.1371/journal.pcbi.1005135", - "standard_citation": "doi:10.1371/journal.pcbi.1005135", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 27 - ] - ], - "date-time": "2017-09-27T23:22:14Z", - "timestamp": 1506554534835 - }, - "reference-count": 55, - "publisher": "Public Library of Science (PLoS)", - "issue": "10", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2016, - 10, - 7 - ] - ], - "date-time": "2016-10-07T00:00:00Z", - "timestamp": 1475798400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/100000092", - "name": "U.S. National Library of Medicine", - "doi-asserted-by": "publisher", - "award": [ - "R01LM011986" - ] - }, - { - "DOI": "10.13039/100000001", - "name": "National Science Foundation", - "doi-asserted-by": "publisher", - "award": [ - "CNS-0855217", - "ACI-1126113", - "CNS-0958379" - ] - }, - { - "name": "UNI's Summer Fellowship", - "award": [] - } - ], - "content-domain": { - "domain": [ - "www.ploscompbiol.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pcbi.1005135", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 10, - 7 - ] - ], - "date-time": "2016-10-07T17:47:03Z", - "timestamp": 1475862423000 - }, - "page": "e1005135", - "update-policy": "http://dx.doi.org/10.1371/journal.pcbi.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 6, - "title": "Large-Scale Off-Target Identification Using Fast and Accurate Dual Regularized One-Class Collaborative Filtering and Its Application to Drug Repurposing", - "prefix": "10.1371", - "volume": "12", - "author": [ - { - "ORCID": "http://orcid.org/0000-0002-8420-4750", - "authenticated-orcid": true, - "given": "Hansaim", - "family": "Lim", - "affiliation": [] - }, - { - "given": "Aleksandar", - "family": "Poleksic", - "affiliation": [] - }, - { - "given": "Yuan", - "family": "Yao", - "affiliation": [] - }, - { - "given": "Hanghang", - "family": "Tong", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-0719-8289", - "authenticated-orcid": false, - "given": "Di", - "family": "He", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0003-3892-384X", - "authenticated-orcid": false, - "given": "Luke", - "family": "Zhuang", - "affiliation": [] - }, - { - "given": "Patrick", - "family": "Meng", - "affiliation": [] - }, - { - "given": "Lei", - "family": "Xie", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2016, - 10, - 7 - ] - ] - }, - "container-title": "PLOS Computational Biology", - "original-title": [], - "link": [ - { - "URL": "http://dx.plos.org/10.1371/journal.pcbi.1005135", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T00:37:26Z", - "timestamp": 1498351046000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Avner", - "family": "Schlessinger", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 10, - 7 - ] - ] - }, - "references-count": 55, - "URL": "https://doi.org/10.1371/journal.pcbi.1005135", - "relation": { - "cites": [] - }, - "subject": [ - "Ecology", - "Modelling and Simulation", - "Computational Theory and Mathematics", - "Genetics", - "Ecology, Evolution, Behavior and Systematics", - "Molecular Biology", - "Cellular and Molecular Neuroscience" - ], - "container-title-short": "PLoS Comput Biol", - "id": "1SIuofeg" - }, - "citation_id": "1SIuofeg" - }, - "doi:10.1371/journal.pcbi.1005177": { - "source": "doi", - "identifer": "10.1371/journal.pcbi.1005177", - "standard_citation": "doi:10.1371/journal.pcbi.1005177", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 11 - ] - ], - "date-time": "2017-09-11T23:22:24Z", - "timestamp": 1505172144245 - }, - "reference-count": 66, - "publisher": "Public Library of Science (PLoS)", - "issue": "11", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2016, - 11, - 4 - ] - ], - "date-time": "2016-11-04T00:00:00Z", - "timestamp": 1478217600000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/100000057", - "name": "National Institute of General Medical Sciences", - "doi-asserted-by": "publisher", - "award": [ - "1F32GM119319-01", - "P50 GM107615" - ] - }, - { - "DOI": "10.13039/100000861", - "name": "Burroughs Wellcome Fund", - "doi-asserted-by": "publisher", - "award": [ - "Postdoctoral Enrichment Program" - ] - }, - { - "name": "Paul G. Allen Family Foundation (US)", - "award": [ - "Allen Distinguished Investigator", - "Allen Discovery Center" - ] - }, - { - "name": "Paul G. Allen Family Foundation (US)", - "award": [ - "Allen Distinguished Investigator", - "Allen Discovery Center" - ] - }, - { - "name": "NIH Office of the Director (US)", - "award": [ - "5DP1LM01150" - ] - }, - { - "name": "U.S. Department of Energy (US)", - "award": [ - "DE-FG02-97ER25308" - ] - }, - { - "name": "Siebel Foundation", - "award": [ - "Graduate Scholarship" - ] - }, - { - "name": "Stanford University School of Medicine (US)", - "award": [ - "Stanford Biomedical Big Data Science Postdoctoral Fellowship" - ] - } - ], - "content-domain": { - "domain": [ - "www.ploscompbiol.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pcbi.1005177", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 11, - 4 - ] - ], - "date-time": "2016-11-04T17:27:54Z", - "timestamp": 1478280474000 - }, - "page": "e1005177", - "update-policy": "http://dx.doi.org/10.1371/journal.pcbi.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 6, - "title": "Deep Learning Automates the Quantitative Analysis of Individual Cells in Live-Cell Imaging Experiments", - "prefix": "10.1371", - "volume": "12", - "author": [ - { - "given": "David A.", - "family": "Van Valen", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-9709-5549", - "authenticated-orcid": true, - "given": "Takamasa", - "family": "Kudo", - "affiliation": [] - }, - { - "given": "Keara M.", - "family": "Lane", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-7257-0315", - "authenticated-orcid": true, - "given": "Derek N.", - "family": "Macklin", - "affiliation": [] - }, - { - "given": "Nicolas T.", - "family": "Quach", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-7197-6292", - "authenticated-orcid": true, - "given": "Mialy M.", - "family": "DeFelice", - "affiliation": [] - }, - { - "given": "Inbal", - "family": "Maayan", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-6039-748X", - "authenticated-orcid": true, - "given": "Yu", - "family": "Tanouchi", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0001-9418-9577", - "authenticated-orcid": true, - "given": "Euan A.", - "family": "Ashley", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-5993-8912", - "authenticated-orcid": true, - "given": "Markus W.", - "family": "Covert", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2016, - 11, - 4 - ] - ] - }, - "container-title": "PLOS Computational Biology", - "original-title": [], - "link": [ - { - "URL": "http://dx.plos.org/10.1371/journal.pcbi.1005177", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T02:52:27Z", - "timestamp": 1498359147000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Martin", - "family": "Meier-Schellersheim", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 11, - 4 - ] - ] - }, - "references-count": 66, - "URL": "https://doi.org/10.1371/journal.pcbi.1005177", - "relation": { - "cites": [] - }, - "subject": [ - "Ecology", - "Modelling and Simulation", - "Computational Theory and Mathematics", - "Genetics", - "Ecology, Evolution, Behavior and Systematics", - "Molecular Biology", - "Cellular and Molecular Neuroscience" - ], - "container-title-short": "PLoS Comput Biol", - "id": "40EG4ZEU" - }, - "citation_id": "40EG4ZEU" - }, - "doi:10.1371/journal.pcbi.1005219": { - "source": "doi", - "identifer": "10.1371/journal.pcbi.1005219", - "standard_citation": "doi:10.1371/journal.pcbi.1005219", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T04:14:12Z", - "timestamp": 1502338452756 - }, - "reference-count": 76, - "publisher": "Public Library of Science (PLoS)", - "issue": "11", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2016, - 11, - 28 - ] - ], - "date-time": "2016-11-28T00:00:00Z", - "timestamp": 1480291200000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/501100001871", - "name": "Fundação para a Ciência e a Tecnologia", - "doi-asserted-by": "publisher", - "award": [ - "SFRH/BD/86343/2012" - ] - } - ], - "content-domain": { - "domain": [ - "www.ploscompbiol.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pcbi.1005219", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 11, - 28 - ] - ], - "date-time": "2016-11-28T18:31:17Z", - "timestamp": 1480357877000 - }, - "page": "e1005219", - "update-policy": "http://dx.doi.org/10.1371/journal.pcbi.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 2, - "title": "Computational Discovery of Putative Leads for Drug Repositioning through Drug-Target Interaction Prediction", - "prefix": "10.1371", - "volume": "12", - "author": [ - { - "given": "Edgar D.", - "family": "Coelho", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0003-4937-2334", - "authenticated-orcid": true, - "given": "Joel P.", - "family": "Arrais", - "affiliation": [] - }, - { - "given": "José Luís", - "family": "Oliveira", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2016, - 11, - 28 - ] - ] - }, - "container-title": "PLOS Computational Biology", - "original-title": [], - "link": [ - { - "URL": "http://dx.plos.org/10.1371/journal.pcbi.1005219", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T04:39:19Z", - "timestamp": 1498365559000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "James M.", - "family": "Briggs", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 11, - 28 - ] - ] - }, - "references-count": 76, - "URL": "https://doi.org/10.1371/journal.pcbi.1005219", - "relation": { - "cites": [] - }, - "subject": [ - "Ecology", - "Modelling and Simulation", - "Computational Theory and Mathematics", - "Genetics", - "Ecology, Evolution, Behavior and Systematics", - "Molecular Biology", - "Cellular and Molecular Neuroscience" - ], - "container-title-short": "PLoS Comput Biol", - "id": "tOpadZQw" - }, - "citation_id": "tOpadZQw" - }, - "doi:10.1371/journal.pcbi.1005324": { - "source": "doi", - "identifer": "10.1371/journal.pcbi.1005324", - "standard_citation": "doi:10.1371/journal.pcbi.1005324", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 27 - ] - ], - "date-time": "2017-09-27T02:42:16Z", - "timestamp": 1506480136884 - }, - "update-to": [ - { - "updated": { - "date-parts": [ - [ - 2017, - 1, - 20 - ] - ], - "date-time": "2017-01-20T00:00:00Z", - "timestamp": 1484870400000 - }, - "DOI": "10.1371/journal.pcbi.1005324", - "type": "new_version", - "label": "New version" - } - ], - "reference-count": 49, - "publisher": "Public Library of Science (PLoS)", - "issue": "1", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2017, - 1, - 5 - ] - ], - "date-time": "2017-01-05T00:00:00Z", - "timestamp": 1483574400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/100000057", - "name": "National Institute of General Medical Sciences", - "doi-asserted-by": "publisher", - "award": [ - "R01GM089753" - ] - }, - { - "DOI": "10.13039/100000076", - "name": "Directorate for Biological Sciences", - "doi-asserted-by": "publisher", - "award": [ - "DBI-1564955" - ] - } - ], - "content-domain": { - "domain": [ - "www.ploscompbiol.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pcbi.1005324", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2017, - 1, - 9 - ] - ], - "date-time": "2017-01-09T19:57:34Z", - "timestamp": 1483991854000 - }, - "page": "e1005324", - "update-policy": "http://dx.doi.org/10.1371/journal.pcbi.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 20, - "title": "Accurate De Novo Prediction of Protein Contact Map by Ultra-Deep Learning Model", - "prefix": "10.1371", - "volume": "13", - "author": [ - { - "given": "Sheng", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Siqi", - "family": "Sun", - "affiliation": [] - }, - { - "given": "Zhen", - "family": "Li", - "affiliation": [] - }, - { - "given": "Renyu", - "family": "Zhang", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0001-7111-4839", - "authenticated-orcid": true, - "given": "Jinbo", - "family": "Xu", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2017, - 1, - 5 - ] - ] - }, - "container-title": "PLOS Computational Biology", - "original-title": [], - "link": [ - { - "URL": "http://dx.plos.org/10.1371/journal.pcbi.1005324", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T07:42:09Z", - "timestamp": 1498376529000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Avner", - "family": "Schlessinger", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 1, - 5 - ] - ] - }, - "references-count": 49, - "URL": "https://doi.org/10.1371/journal.pcbi.1005324", - "relation": { - "has-preprint": [ - { - "id-type": "doi", - "id": "10.1101/073239", - "asserted-by": "object" - } - ], - "cites": [] - }, - "subject": [ - "Ecology", - "Modelling and Simulation", - "Computational Theory and Mathematics", - "Genetics", - "Ecology, Evolution, Behavior and Systematics", - "Molecular Biology", - "Cellular and Molecular Neuroscience" - ], - "container-title-short": "PLoS Comput Biol", - "id": "BhfjKSY3" - }, - "citation_id": "BhfjKSY3" - }, - "doi:10.1371/journal.pone.0028766": { - "source": "doi", - "identifer": "10.1371/journal.pone.0028766", - "standard_citation": "doi:10.1371/journal.pone.0028766", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T11:02:15Z", - "timestamp": 1507114935021 - }, - "reference-count": 82, - "publisher": "Public Library of Science (PLoS)", - "issue": "12", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2011, - 12, - 7 - ] - ], - "date-time": "2011-12-07T00:00:00Z", - "timestamp": 1323216000000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "www.plosone.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pone.0028766", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2011, - 12, - 7 - ] - ], - "date-time": "2011-12-07T22:09:50Z", - "timestamp": 1323295790000 - }, - "page": "e28766", - "update-policy": "http://dx.doi.org/10.1371/journal.pone.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 292, - "title": "Protein 3D Structure Computed from Evolutionary Sequence Variation", - "prefix": "10.1371", - "volume": "6", - "author": [ - { - "given": "Debora S.", - "family": "Marks", - "affiliation": [] - }, - { - "given": "Lucy J.", - "family": "Colwell", - "affiliation": [] - }, - { - "given": "Robert", - "family": "Sheridan", - "affiliation": [] - }, - { - "given": "Thomas A.", - "family": "Hopf", - "affiliation": [] - }, - { - "given": "Andrea", - "family": "Pagnani", - "affiliation": [] - }, - { - "given": "Riccardo", - "family": "Zecchina", - "affiliation": [] - }, - { - "given": "Chris", - "family": "Sander", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2011, - 12, - 7 - ] - ] - }, - "container-title": "PLoS ONE", - "original-title": [], - "link": [ - { - "URL": "http://dx.plos.org/10.1371/journal.pone.0028766", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 20 - ] - ], - "date-time": "2017-06-20T10:58:21Z", - "timestamp": 1497956301000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Andrej", - "family": "Sali", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2011, - 12, - 7 - ] - ] - }, - "references-count": 82, - "URL": "https://doi.org/10.1371/journal.pone.0028766", - "relation": { - "cites": [] - }, - "subject": [ - "General Biochemistry, Genetics and Molecular Biology", - "General Agricultural and Biological Sciences", - "General Medicine" - ], - "container-title-short": "PLoS ONE", - "id": "10dNuD89l" - }, - "citation_id": "10dNuD89l" - }, - "doi:10.1371/journal.pone.0032235": { - "source": "doi", - "identifer": "10.1371/journal.pone.0032235", - "standard_citation": "doi:10.1371/journal.pone.0032235", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 15 - ] - ], - "date-time": "2017-08-15T11:02:13Z", - "timestamp": 1502794933439 - }, - "reference-count": 58, - "publisher": "Public Library of Science (PLoS)", - "issue": "3", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2012, - 3, - 26 - ] - ], - "date-time": "2012-03-26T00:00:00Z", - "timestamp": 1332720000000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "www.plosone.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pone.0032235", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2012, - 3, - 26 - ] - ], - "date-time": "2012-03-26T21:20:23Z", - "timestamp": 1332796823000 - }, - "page": "e32235", - "update-policy": "http://dx.doi.org/10.1371/journal.pone.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 12, - "title": "A Unified Multitask Architecture for Predicting Local Protein Properties", - "prefix": "10.1371", - "volume": "7", - "author": [ - { - "given": "Yanjun", - "family": "Qi", - "affiliation": [] - }, - { - "given": "Merja", - "family": "Oja", - "affiliation": [] - }, - { - "given": "Jason", - "family": "Weston", - "affiliation": [] - }, - { - "given": "William Stafford", - "family": "Noble", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2012, - 3, - 26 - ] - ] - }, - "container-title": "PLoS ONE", - "original-title": [], - "link": [ - { - "URL": "http://dx.plos.org/10.1371/journal.pone.0032235", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 20 - ] - ], - "date-time": "2017-06-20T16:08:12Z", - "timestamp": 1497974892000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Christos A.", - "family": "Ouzounis", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2012, - 3, - 26 - ] - ] - }, - "references-count": 58, - "URL": "https://doi.org/10.1371/journal.pone.0032235", - "relation": { - "cites": [] - }, - "subject": [ - "General Biochemistry, Genetics and Molecular Biology", - "General Agricultural and Biological Sciences", - "General Medicine" - ], - "container-title-short": "PLoS ONE", - "id": "1AlhRKQbe" - }, - "citation_id": "1AlhRKQbe" - }, - "doi:10.1371/journal.pone.0061318": { - "source": "doi", - "identifer": "10.1371/journal.pone.0061318", - "standard_citation": "doi:10.1371/journal.pone.0061318", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T07:22:12Z", - "timestamp": 1507015332882 - }, - "reference-count": 41, - "publisher": "Public Library of Science (PLoS)", - "issue": "4", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2013, - 4, - 30 - ] - ], - "date-time": "2013-04-30T00:00:00Z", - "timestamp": 1367280000000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "www.plosone.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pone.0061318", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2013, - 5, - 2 - ] - ], - "date-time": "2013-05-02T00:48:05Z", - "timestamp": 1367455685000 - }, - "page": "e61318", - "update-policy": "http://dx.doi.org/10.1371/journal.pone.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 54, - "title": "Machine Learning Prediction of Cancer Cell Sensitivity to Drugs Based on Genomic and Chemical Properties", - "prefix": "10.1371", - "volume": "8", - "author": [ - { - "given": "Michael P.", - "family": "Menden", - "affiliation": [] - }, - { - "given": "Francesco", - "family": "Iorio", - "affiliation": [] - }, - { - "given": "Mathew", - "family": "Garnett", - "affiliation": [] - }, - { - "given": "Ultan", - "family": "McDermott", - "affiliation": [] - }, - { - "given": "Cyril H.", - "family": "Benes", - "affiliation": [] - }, - { - "given": "Pedro J.", - "family": "Ballester", - "affiliation": [] - }, - { - "given": "Julio", - "family": "Saez-Rodriguez", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2013, - 4, - 30 - ] - ] - }, - "container-title": "PLoS ONE", - "original-title": [], - "link": [ - { - "URL": "http://dx.plos.org/10.1371/journal.pone.0061318", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 21 - ] - ], - "date-time": "2017-06-21T12:21:43Z", - "timestamp": 1498047703000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Gajendra P. S.", - "family": "Raghava", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2013, - 4, - 30 - ] - ] - }, - "references-count": 41, - "URL": "https://doi.org/10.1371/journal.pone.0061318", - "relation": { - "cites": [] - }, - "subject": [ - "General Biochemistry, Genetics and Molecular Biology", - "General Agricultural and Biological Sciences", - "General Medicine" - ], - "container-title-short": "PLoS ONE", - "id": "QcwZC8wG" - }, - "citation_id": "QcwZC8wG" - }, - "doi:10.1371/journal.pone.0066341": { - "source": "doi", - "identifer": "10.1371/journal.pone.0066341", - "standard_citation": "doi:10.1371/journal.pone.0066341", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T04:02:42Z", - "timestamp": 1507003362623 - }, - "reference-count": 63, - "publisher": "Public Library of Science (PLoS)", - "issue": "6", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2013, - 6, - 24 - ] - ], - "date-time": "2013-06-24T00:00:00Z", - "timestamp": 1372032000000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "www.plosone.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pone.0066341", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2013, - 6, - 24 - ] - ], - "date-time": "2013-06-24T21:19:25Z", - "timestamp": 1372108765000 - }, - "page": "e66341", - "update-policy": "http://dx.doi.org/10.1371/journal.pone.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 39, - "title": "Computational Phenotype Discovery Using Unsupervised Feature Learning over Noisy, Sparse, and Irregular Clinical Data", - "prefix": "10.1371", - "volume": "8", - "author": [ - { - "given": "Thomas A.", - "family": "Lasko", - "affiliation": [] - }, - { - "given": "Joshua C.", - "family": "Denny", - "affiliation": [] - }, - { - "given": "Mia A.", - "family": "Levy", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2013, - 6, - 24 - ] - ] - }, - "container-title": "PLoS ONE", - "original-title": [], - "link": [ - { - "URL": "http://dx.plos.org/10.1371/journal.pone.0066341", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 21 - ] - ], - "date-time": "2017-06-21T15:09:29Z", - "timestamp": 1498057769000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Joseph", - "family": "Devaney", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2013, - 6, - 24 - ] - ] - }, - "references-count": 63, - "URL": "https://doi.org/10.1371/journal.pone.0066341", - "relation": { - "cites": [] - }, - "subject": [ - "General Biochemistry, Genetics and Molecular Biology", - "General Agricultural and Biological Sciences", - "General Medicine" - ], - "container-title-short": "PLoS ONE", - "id": "FLX0o7bL" - }, - "citation_id": "FLX0o7bL" - }, - "doi:10.1371/journal.pone.0171410": { - "source": "doi", - "identifer": "10.1371/journal.pone.0171410", - "standard_citation": "doi:10.1371/journal.pone.0171410", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T22:33:58Z", - "timestamp": 1502404438610 - }, - "reference-count": 43, - "publisher": "Public Library of Science (PLoS)", - "issue": "2", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2017, - 2, - 3 - ] - ], - "date-time": "2017-02-03T00:00:00Z", - "timestamp": 1486080000000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "www.plosone.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pone.0171410", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2017, - 2, - 3 - ] - ], - "date-time": "2017-02-03T18:45:18Z", - "timestamp": 1486147518000 - }, - "page": "e0171410", - "update-policy": "http://dx.doi.org/10.1371/journal.pone.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Recognition of prokaryotic and eukaryotic promoters using convolutional deep learning neural networks", - "prefix": "10.1371", - "volume": "12", - "author": [ - { - "given": "Ramzan Kh.", - "family": "Umarov", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0001-8885-493X", - "authenticated-orcid": true, - "given": "Victor V.", - "family": "Solovyev", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2017, - 2, - 3 - ] - ] - }, - "container-title": "PLOS ONE", - "original-title": [], - "link": [ - { - "URL": "http://dx.plos.org/10.1371/journal.pone.0171410", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T09:43:42Z", - "timestamp": 1498383822000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Igor B.", - "family": "Rogozin", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 2, - 3 - ] - ] - }, - "references-count": 43, - "URL": "https://doi.org/10.1371/journal.pone.0171410", - "relation": { - "cites": [] - }, - "subject": [ - "General Biochemistry, Genetics and Molecular Biology", - "General Agricultural and Biological Sciences", - "General Medicine" - ], - "container-title-short": "PLoS ONE", - "id": "as2HfoSh" - }, - "citation_id": "as2HfoSh" - }, - "doi:10.15252/msb.20156651": { - "source": "doi", - "identifer": "10.15252/msb.20156651", - "standard_citation": "doi:10.15252/msb.20156651", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T04:02:42Z", - "timestamp": 1507003362640 - }, - "reference-count": 0, - "publisher": "EMBO", - "issue": "7", - "funder": [ - { - "DOI": "10.13039/100004410", - "name": "European Molecular Biology Organization", - "doi-asserted-by": "publisher", - "award": [] - }, - { - "DOI": "10.13039/501100000780", - "name": "European Commission", - "doi-asserted-by": "publisher", - "award": [] - }, - { - "DOI": "10.13039/501100002301", - "name": "Eesti Teadusagentuur", - "doi-asserted-by": "publisher", - "award": [ - "IUT34‐4" - ] - }, - { - "DOI": "10.13039/100004440", - "name": "Wellcome Trust", - "doi-asserted-by": "publisher", - "award": [] - }, - { - "DOI": "10.13039/501100000781", - "name": "European Research Council", - "doi-asserted-by": "publisher", - "award": [ - "N635290" - ] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 7 - ] - ] - }, - "DOI": "10.15252/msb.20156651", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 7, - 30 - ] - ], - "date-time": "2016-07-30T00:38:24Z", - "timestamp": 1469839104000 - }, - "page": "878", - "source": "Crossref", - "is-referenced-by-count": 33, - "title": "Deep learning for computational biology", - "prefix": "10.15252", - "volume": "12", - "author": [ - { - "given": "Christof", - "family": "Angermueller", - "affiliation": [] - }, - { - "given": "Tanel", - "family": "Pärnamaa", - "affiliation": [] - }, - { - "given": "Leopold", - "family": "Parts", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-8818-7193", - "authenticated-orcid": false, - "given": "Oliver", - "family": "Stegle", - "affiliation": [] - } - ], - "member": "79", - "published-online": { - "date-parts": [ - [ - 2016, - 7, - 29 - ] - ] - }, - "container-title": "Molecular Systems Biology", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2016, - 7, - 30 - ] - ], - "date-time": "2016-07-30T00:38:29Z", - "timestamp": 1469839109000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 7 - ] - ] - }, - "references-count": 0, - "alternative-id": [ - "10.15252/msb.20156651" - ], - "URL": "https://doi.org/10.15252/msb.20156651", - "relation": {}, - "subject": [ - "General Biochemistry, Genetics and Molecular Biology", - "Computational Theory and Mathematics", - "General Immunology and Microbiology", - "Applied Mathematics", - "General Agricultural and Biological Sciences", - "Information Systems" - ], - "container-title-short": "Mol Syst Biol", - "id": "irSe12Sm" - }, - "citation_id": "irSe12Sm" - }, - "doi:10.1561/0400000042": { - "source": "doi", - "identifer": "10.1561/0400000042", - "standard_citation": "doi:10.1561/0400000042", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T04:22:16Z", - "timestamp": 1507004536534 - }, - "reference-count": 0, - "publisher": "Now Publishers", - "issue": "3-4", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2013 - ] - ] - }, - "DOI": "10.1561/0400000042", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 8, - 11 - ] - ], - "date-time": "2014-08-11T09:35:55Z", - "timestamp": 1407749755000 - }, - "page": "211-407", - "source": "Crossref", - "is-referenced-by-count": 102, - "title": "The Algorithmic Foundations of Differential Privacy", - "prefix": "10.1561", - "volume": "9", - "author": [ - { - "given": "Cynthia", - "family": "Dwork", - "affiliation": [] - }, - { - "given": "Aaron", - "family": "Roth", - "affiliation": [] - } - ], - "member": "432", - "container-title": "Foundations and Trends® in Theoretical Computer Science", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2014, - 8, - 11 - ] - ], - "date-time": "2014-08-11T09:35:55Z", - "timestamp": 1407749755000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2013 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1561/0400000042", - "relation": {}, - "subject": [ - "Theoretical Computer Science" - ], - "container-title-short": "FNT in Theoretical Computer Science", - "id": "v8Lp4ibI" - }, - "citation_id": "v8Lp4ibI" - }, - "doi:10.1590/2176-9451.19.4.027-029.ebo": { - "source": "doi", - "identifer": "10.1590/2176-9451.19.4.027-029.ebo", - "standard_citation": "doi:10.1590/2176-9451.19.4.027-029.ebo", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T11:30:18Z", - "timestamp": 1502278218386 - }, - "reference-count": 3, - "publisher": "FapUNIFESP (SciELO)", - "issue": "4", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2014, - 8 - ] - ] - }, - "abstract": "Sample size calculation is part of the early stages of conducting an epidemiological, clinical or lab study. In preparing a scientific paper, there are ethical and methodological indications for its use. Two investigations conducted with the same methodology and achieving equivalent results, but different only in terms of sample size, may point the researcher in different directions when it comes to making clinical decisions. Therefore, ideally, samples should not be small and, contrary to what one might think, should not be excessive. The aim of this paper is to discuss in clinical language the main implications of the sample size when interpreting a study.", - "DOI": "10.1590/2176-9451.19.4.027-029.ebo", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 10, - 11 - ] - ], - "date-time": "2014-10-11T18:46:34Z", - "timestamp": 1413053194000 - }, - "page": "27-29", - "source": "Crossref", - "is-referenced-by-count": 4, - "title": "How sample size influences research outcomes", - "prefix": "10.1590", - "volume": "19", - "author": [ - { - "given": "Jorge", - "family": "Faber", - "affiliation": [ - { - "name": "University of Brasília" - } - ] - }, - { - "given": "Lilian Martins", - "family": "Fonseca", - "affiliation": [ - { - "name": "University of Brasília" - } - ] - } - ], - "member": "530", - "container-title": "Dental Press Journal of Orthodontics", - "original-title": [], - "link": [ - { - "URL": "http://www.scielo.br/pdf/dpjo/v19n4/2176-9451-dpjo-19-04-00027.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 5, - 27 - ] - ], - "date-time": "2017-05-27T22:09:08Z", - "timestamp": 1495922948000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 8 - ] - ] - }, - "references-count": 3, - "alternative-id": [ - "S2176-94512014000400027" - ], - "URL": "https://doi.org/10.1590/2176-9451.19.4.027-029.ebo", - "relation": { - "cites": [] - }, - "subject": [ - "Orthodontics", - "Oral Surgery" - ], - "container-title-short": "Dental Press J. Orthod.", - "id": "Fx5qVQlk" - }, - "citation_id": "Fx5qVQlk" - }, - "doi:10.2307/2281868": { - "source": "doi", - "identifer": "10.2307/2281868", - "standard_citation": "doi:10.2307/2281868", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T10:02:18Z", - "timestamp": 1507111338766 - }, - "reference-count": 28, - "publisher": "Informa UK Limited", - "issue": "282", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 1958, - 6 - ] - ] - }, - "DOI": "10.1080/01621459.1958.10501452", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2012, - 4, - 12 - ] - ], - "date-time": "2012-04-12T17:59:40Z", - "timestamp": 1334253580000 - }, - "page": "457-481", - "source": "Crossref", - "is-referenced-by-count": 15699, - "title": "Nonparametric Estimation from Incomplete Observations", - "prefix": "10.1080", - "volume": "53", - "author": [ - { - "given": "E. L.", - "family": "Kaplan", - "affiliation": [] - }, - { - "given": "Paul", - "family": "Meier", - "affiliation": [] - } - ], - "member": "301", - "container-title": "Journal of the American Statistical Association", - "original-title": [], - "link": [ - { - "URL": "http://www.tandfonline.com/doi/pdf/10.1080/01621459.1958.10501452", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 20 - ] - ], - "date-time": "2017-06-20T17:11:02Z", - "timestamp": 1497978662000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 1958, - 6 - ] - ] - }, - "references-count": 28, - "alternative-id": [ - "10.1080/01621459.1958.10501452" - ], - "URL": "https://doi.org/10.1080/01621459.1958.10501452", - "relation": { - "cites": [] - }, - "subject": [ - "Statistics, Probability and Uncertainty", - "Statistics and Probability" - ], - "container-title-short": "Journal of the American Statistical Association", - "id": "6RHepB1T" - }, - "citation_id": "6RHepB1T" - }, - "doi:10.2337/diab.35.2.237": { - "source": "doi", - "identifer": "10.2337/diab.35.2.237", - "standard_citation": "doi:10.2337/diab.35.2.237", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 26 - ] - ], - "date-time": "2017-08-26T09:45:12Z", - "timestamp": 1503740712700 - }, - "reference-count": 0, - "publisher": "American Diabetes Association", - "issue": "2", - "content-domain": { - "domain": [ - "diabetesjournals.org" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 1986, - 2, - 1 - ] - ] - }, - "DOI": "10.2337/diab.35.2.237", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2013, - 9, - 19 - ] - ], - "date-time": "2013-09-19T17:16:02Z", - "timestamp": 1379610962000 - }, - "page": "237-241", - "update-policy": "http://dx.doi.org/10.2337/crossmarkpolicy", - "source": "Crossref", - "is-referenced-by-count": 58, - "title": "Islet Cell Antibodies Identify Latent Type I Diabetes in Patients Aged 35-75 Years at Diagnosis", - "prefix": "10.2337", - "volume": "35", - "author": [ - { - "given": "L. C.", - "family": "Groop", - "affiliation": [] - }, - { - "given": "G. F.", - "family": "Bottazzo", - "affiliation": [] - }, - { - "given": "D.", - "family": "Doniach", - "affiliation": [] - } - ], - "member": "1167", - "container-title": "Diabetes", - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.2337/diab.35.2.237", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 12, - 23 - ] - ], - "date-time": "2016-12-23T01:52:38Z", - "timestamp": 1482457958000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 1986, - 2, - 1 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.2337/diab.35.2.237", - "relation": {}, - "subject": [ - "Internal Medicine", - "Endocrinology, Diabetes and Metabolism" - ], - "container-title-short": "Diabetes", - "id": "ws1zvGoZ" - }, - "citation_id": "ws1zvGoZ" - }, - "doi:10.2337/diabetes.54.suppl_2.s68": { - "source": "doi", - "identifer": "10.2337/diabetes.54.suppl_2.s68", - "standard_citation": "doi:10.2337/diabetes.54.suppl_2.s68", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 29 - ] - ], - "date-time": "2017-09-29T07:02:18Z", - "timestamp": 1506668538366 - }, - "reference-count": 46, - "publisher": "American Diabetes Association", - "issue": "Supplement 2", - "content-domain": { - "domain": [ - "diabetesjournals.org" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2005, - 12, - 1 - ] - ] - }, - "DOI": "10.2337/diabetes.54.suppl_2.s68", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2007, - 3, - 6 - ] - ], - "date-time": "2007-03-06T19:04:22Z", - "timestamp": 1173207862000 - }, - "page": "S68-S72", - "update-policy": "http://dx.doi.org/10.2337/crossmarkpolicy", - "source": "Crossref", - "is-referenced-by-count": 71, - "title": "Latent Autoimmune Diabetes in Adults: Definition, Prevalence,  -Cell Function, and Treatment", - "prefix": "10.2337", - "volume": "54", - "author": [ - { - "given": "G.", - "family": "Stenstrom", - "affiliation": [] - }, - { - "given": "A.", - "family": "Gottsater", - "affiliation": [] - }, - { - "given": "E.", - "family": "Bakhtadze", - "affiliation": [] - }, - { - "given": "B.", - "family": "Berger", - "affiliation": [] - }, - { - "given": "G.", - "family": "Sundkvist", - "affiliation": [] - } - ], - "member": "1167", - "published-online": { - "date-parts": [ - [ - 2005, - 11, - 23 - ] - ] - }, - "container-title": "Diabetes", - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.2337/diabetes.54.suppl_2.S68", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 17 - ] - ], - "date-time": "2017-06-17T15:01:02Z", - "timestamp": 1497711662000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2005, - 11, - 23 - ] - ] - }, - "references-count": 46, - "URL": "https://doi.org/10.2337/diabetes.54.suppl_2.s68", - "relation": {}, - "subject": [ - "Internal Medicine", - "Endocrinology, Diabetes and Metabolism" - ], - "container-title-short": "Diabetes", - "id": "WuOqsORY" - }, - "citation_id": "WuOqsORY" - }, - "doi:10.3109/10409238.2015.1135868": { - "source": "doi", - "identifer": "10.3109/10409238.2015.1135868", - "standard_citation": "doi:10.3109/10409238.2015.1135868", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 31 - ] - ], - "date-time": "2017-08-31T16:02:17Z", - "timestamp": 1504195337888 - }, - "reference-count": 90, - "publisher": "Informa UK Limited", - "issue": "2", - "content-domain": { - "domain": [ - "www.tandfonline.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2016, - 3, - 3 - ] - ] - }, - "DOI": "10.3109/10409238.2015.1135868", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 1, - 25 - ] - ], - "date-time": "2016-01-25T07:02:47Z", - "timestamp": 1453705367000 - }, - "page": "102-109", - "update-policy": "http://dx.doi.org/10.1080/tandf_crossmark_01", - "source": "Crossref", - "is-referenced-by-count": 8, - "title": "Computer vision for high content screening", - "prefix": "10.3109", - "volume": "51", - "author": [ - { - "given": "Oren Z.", - "family": "Kraus", - "affiliation": [] - }, - { - "given": "Brendan J.", - "family": "Frey", - "affiliation": [] - } - ], - "member": "301", - "published-online": { - "date-parts": [ - [ - 2016, - 1, - 24 - ] - ] - }, - "container-title": "Critical Reviews in Biochemistry and Molecular Biology", - "original-title": [], - "link": [ - { - "URL": "http://www.tandfonline.com/doi/pdf/10.3109/10409238.2015.1135868", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 1, - 6 - ] - ], - "date-time": "2017-01-06T12:05:23Z", - "timestamp": 1483704323000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 1, - 24 - ] - ] - }, - "references-count": 90, - "alternative-id": [ - "10.3109/10409238.2015.1135868" - ], - "URL": "https://doi.org/10.3109/10409238.2015.1135868", - "relation": { - "cites": [] - }, - "subject": [ - "Biochemistry", - "Molecular Biology" - ], - "container-title-short": "Critical Reviews in Biochemistry and Molecular Biology", - "assertion": [ - { - "value": "The publishing and review policy for this title is described in its Aims & Scope.", - "order": 1, - "name": "peerreview_statement", - "label": "Peer Review Statement" - }, - { - "value": "http://www.tandfonline.com/action/journalInformation?show=aimsScope&journalCode=ibmg20", - "URL": "http://www.tandfonline.com/action/journalInformation?show=aimsScope&journalCode=ibmg20", - "order": 2, - "name": "aims_and_scope_url", - "label": "Aim & Scope" - } - ], - "id": "MmRGFVUu" - }, - "citation_id": "MmRGFVUu" - }, - "doi:10.3389/fgene.2014.00342": { - "source": "doi", - "identifer": "10.3389/fgene.2014.00342", - "standard_citation": "doi:10.3389/fgene.2014.00342", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T10:38:25Z", - "timestamp": 1502275105709 - }, - "reference-count": 21, - "publisher": "Frontiers Media SA", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.3389/fgene.2014.00342", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 9, - 30 - ] - ], - "date-time": "2014-09-30T15:41:13Z", - "timestamp": 1412091673000 - }, - "source": "Crossref", - "is-referenced-by-count": 2, - "title": "Large-scale integration of small molecule-induced genome-wide transcriptional responses, Kinome-wide binding affinities and cell-growth inhibition profiles reveal global trends characterizing systems-level drug action", - "prefix": "10.3389", - "volume": "5", - "author": [ - { - "given": "DuÅ¡ica", - "family": "Vidović", - "affiliation": [] - }, - { - "given": "Amar", - "family": "Koleti", - "affiliation": [] - }, - { - "given": "Stephan C.", - "family": "Schürer", - "affiliation": [] - } - ], - "member": "1965", - "published-online": { - "date-parts": [ - [ - 2014, - 9, - 30 - ] - ] - }, - "container-title": "Frontiers in Genetics", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2015, - 2, - 18 - ] - ], - "date-time": "2015-02-18T20:19:46Z", - "timestamp": 1424290786000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 9, - 30 - ] - ] - }, - "references-count": 21, - "URL": "https://doi.org/10.3389/fgene.2014.00342", - "relation": { - "cites": [] - }, - "subject": [ - "Genetics(clinical)", - "Molecular Medicine", - "Genetics" - ], - "container-title-short": "Front. Genet.", - "id": "ppGS5h4v" - }, - "citation_id": "ppGS5h4v" - }, - "pmid:21347133": { - "source": "pmid", - "identifer": "21347133", - "standard_citation": "pmid:21347133", - "citeproc": { - "source": "PMC", - "accessed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ] - }, - "id": "y9ONtSZ9", - "title": "Secondary Use of EHR: Data Quality Issues and Informatics Opportunities", - "author": [ - { - "family": "Botsis", - "given": "Taxiarchis" - }, - { - "family": "Hartvigsen", - "given": "Gunnar" - }, - { - "family": "Chen", - "given": "Fei" - }, - { - "family": "Weng", - "given": "Chunhua" - } - ], - "container-title-short": "Summit on Translat Bioinforma", - "container-title": "Summit on Translational Bioinformatics", - "publisher": "American Medical Informatics Association", - "issued": { - "date-parts": [ - [ - 2010 - ] - ] - }, - "page": "1-5", - "volume": "2010", - "PMID": "21347133", - "PMCID": "PMC3041534", - "type": "article-journal", - "URL": "https://www.ncbi.nlm.nih.gov/pubmed/21347133" - }, - "citation_id": "y9ONtSZ9" - }, - "pmid:24159271": { - "source": "pmid", - "identifer": "24159271", - "standard_citation": "pmid:24159271", - "citeproc": { - "source": "PMC", - "accessed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ] - }, - "id": "11sli93ov", - "title": "Impact of Electronic Health Record Systems on Information Integrity: Quality and Safety Implications", - "author": [ - { - "family": "Bowman", - "given": "Sue" - } - ], - "container-title-short": "Perspect Health Inf Manag", - "container-title": "Perspectives in Health Information Management", - "publisher": "American Health Information Management Association", - "issued": { - "date-parts": [ - [ - 2013 - ] - ] - }, - "page": "1c", - "volume": "10", - "issue": "Fall", - "PMID": "24159271", - "PMCID": "PMC3797550", - "type": "article-journal", - "URL": "https://www.ncbi.nlm.nih.gov/pubmed/24159271" - }, - "citation_id": "11sli93ov" - }, - "pmid:26958203": { - "source": "pmid", - "identifer": "26958203", - "standard_citation": "pmid:26958203", - "citeproc": { - "source": "PMC", - "accessed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ] - }, - "id": "FUIfIdE", - "title": "Causal Phenotype Discovery via Deep Networks", - "author": [ - { - "family": "Kale", - "given": "David C." - }, - { - "family": "Che", - "given": "Zhengping" - }, - { - "family": "Bahadori", - "given": "Mohammad Taha" - }, - { - "family": "Li", - "given": "Wenzhe" - }, - { - "family": "Liu", - "given": "Yan" - }, - { - "family": "Wetzel", - "given": "Randall" - } - ], - "container-title-short": "AMIA Annu Symp Proc", - "container-title": "AMIA Annual Symposium Proceedings", - "publisher": "American Medical Informatics Association", - "issued": { - "date-parts": [ - [ - 2015 - ] - ] - }, - "page": "677-686", - "volume": "2015", - "PMID": "26958203", - "PMCID": "PMC4765623", - "type": "article-journal", - "URL": "https://www.ncbi.nlm.nih.gov/pubmed/26958203" - }, - "citation_id": "FUIfIdE" - }, - "pmid:27134610": { - "source": "pmid", - "identifer": "27134610", - "standard_citation": "pmid:27134610", - "citeproc": { - "source": "PMC", - "accessed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ] - }, - "id": "4rTluXLs", - "title": "Why Patient Matching Is a Challenge: Research on Master Patient Index (MPI) Data Discrepancies in Key Identifying Fields", - "author": [ - { - "family": "Just", - "given": "Beth Haenke" - }, - { - "family": "Marc", - "given": "David" - }, - { - "family": "Munns", - "given": "Megan" - }, - { - "family": "Sandefer", - "given": "Ryan" - } - ], - "container-title-short": "Perspect Health Inf Manag", - "container-title": "Perspectives in Health Information Management", - "publisher": "American Health Information Management Association", - "issued": { - "date-parts": [ - [ - 2016 - ] - ] - }, - "page": "1e", - "volume": "13", - "issue": "Spring", - "PMID": "27134610", - "PMCID": "PMC4832129", - "type": "article-journal", - "URL": "https://www.ncbi.nlm.nih.gov/pubmed/27134610" - }, - "citation_id": "4rTluXLs" - }, - "doi:10.1101/gr.634603": { - "source": "doi", - "identifer": "10.1101/gr.634603", - "standard_citation": "doi:10.1101/gr.634603", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 26 - ] - ], - "date-time": "2017-09-26T20:22:12Z", - "timestamp": 1506457332912 - }, - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "issue": "4", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2003, - 4, - 1 - ] - ] - }, - "DOI": "10.1101/gr.634603", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2003, - 4, - 2 - ] - ], - "date-time": "2003-04-02T03:44:29Z", - "timestamp": 1049255069000 - }, - "page": "693-702", - "source": "Crossref", - "is-referenced-by-count": 116, - "title": "Informatics for Unveiling Hidden Genome Signatures", - "prefix": "10.1101", - "volume": "13", - "author": [ - { - "given": "T.", - "family": "Abe", - "affiliation": [] - } - ], - "member": "246", - "container-title": "Genome Research", - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/gr.634603", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 2, - 15 - ] - ], - "date-time": "2017-02-15T17:54:04Z", - "timestamp": 1487181244000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2003, - 4, - 1 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/gr.634603", - "relation": {}, - "subject": [ - "Genetics(clinical)", - "Genetics" - ], - "id": "1HhqhBwrM" - }, - "citation_id": "1HhqhBwrM" - }, - "doi:10.1167/iovs.16-19964": { - "source": "doi", - "identifer": "10.1167/iovs.16-19964", - "standard_citation": "doi:10.1167/iovs.16-19964", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 23 - ] - ], - "date-time": "2017-09-23T04:42:12Z", - "timestamp": 1506141732748 - }, - "reference-count": 0, - "publisher": "Association for Research in Vision and Ophthalmology (ARVO)", - "issue": "13", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by-nc-nd/4.0/", - "start": { - "date-parts": [ - [ - 2016, - 10, - 4 - ] - ], - "date-time": "2016-10-04T00:00:00Z", - "timestamp": 1475539200000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 10, - 4 - ] - ] - }, - "DOI": "10.1167/iovs.16-19964", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 10, - 4 - ] - ], - "date-time": "2016-10-04T19:50:15Z", - "timestamp": 1475610615000 - }, - "page": "5200", - "source": "Crossref", - "is-referenced-by-count": 13, - "title": "Improved Automated Detection of Diabetic Retinopathy on a Publicly Available Dataset Through Integration of Deep Learning", - "prefix": "10.1167", - "volume": "57", - "author": [ - { - "given": "Michael David", - "family": "Abràmoff", - "affiliation": [ - { - "name": "Department of Ophthalmology and Visual Sciences, University of Iowa Hospitals and Clinics, Iowa City, Iowa, United States 2Iowa City Veterans Affairs Medical Center, Iowa City, Iowa, United States 3IDx LLC, Iowa City, Iowa, United States" - } - ] - }, - { - "given": "Yiyue", - "family": "Lou", - "affiliation": [ - { - "name": "Department of Biostatistics, College of Public Health, University of Iowa, Iowa City, Iowa, United States" - } - ] - }, - { - "given": "Ali", - "family": "Erginay", - "affiliation": [ - { - "name": "Service d' Ophtalmologie, Hôpital Lariboisière, APHP, Paris, France" - } - ] - }, - { - "given": "Warren", - "family": "Clarida", - "affiliation": [ - { - "name": "IDx LLC, Iowa City, Iowa, United States" - } - ] - }, - { - "given": "Ryan", - "family": "Amelon", - "affiliation": [ - { - "name": "IDx LLC, Iowa City, Iowa, United States" - } - ] - }, - { - "given": "James C.", - "family": "Folk", - "affiliation": [ - { - "name": "Department of Ophthalmology and Visual Sciences, University of Iowa Hospitals and Clinics, Iowa City, Iowa, United States 3IDx LLC, Iowa City, Iowa, United States" - } - ] - }, - { - "given": "Meindert", - "family": "Niemeijer", - "affiliation": [ - { - "name": "IDx LLC, Iowa City, Iowa, United States" - } - ] - } - ], - "member": "27", - "container-title": "Investigative Opthalmology & Visual Science", - "original-title": [], - "link": [ - { - "URL": "http://iovs.arvojournals.org/article.aspx?doi=10.1167/iovs.16-19964", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 1, - 7 - ] - ], - "date-time": "2017-01-07T05:06:15Z", - "timestamp": 1483765575000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 10, - 4 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1167/iovs.16-19964", - "relation": {}, - "container-title-short": "Invest. Ophthalmol. Vis. Sci.", - "id": "e3vyHBV2" - }, - "citation_id": "e3vyHBV2" - }, - "doi:10.7554/elife.05005": { - "source": "doi", - "identifer": "10.7554/elife.05005", - "standard_citation": "doi:10.7554/elife.05005", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T14:02:16Z", - "timestamp": 1507039336007 - }, - "reference-count": 123, - "publisher": "eLife Sciences Organisation, Ltd.", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2015, - 8, - 12 - ] - ], - "date-time": "2015-08-12T00:00:00Z", - "timestamp": 1439337600000 - }, - "delay-in-days": 0, - "content-version": "vor" - }, - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2015, - 8, - 12 - ] - ], - "date-time": "2015-08-12T00:00:00Z", - "timestamp": 1439337600000 - }, - "delay-in-days": 0, - "content-version": "am" - }, - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2015, - 8, - 12 - ] - ], - "date-time": "2015-08-12T00:00:00Z", - "timestamp": 1439337600000 - }, - "delay-in-days": 0, - "content-version": "tdm" - } - ], - "funder": [ - { - "DOI": "10.13039/100000002", - "name": "National Institutes of Health", - "doi-asserted-by": "publisher", - "award": [ - "GM067031" - ] - }, - { - "DOI": "10.13039/100000011", - "name": "Howard Hughes Medical Institute", - "doi-asserted-by": "publisher", - "award": [ - "Investigator" - ] - }, - { - "DOI": "10.13039/100000001", - "name": "National Science Foundation", - "doi-asserted-by": "publisher", - "award": [ - "Graduate Research Fellowship" - ] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.7554/elife.05005", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 8, - 12 - ] - ], - "date-time": "2015-08-12T09:14:46Z", - "timestamp": 1439370886000 - }, - "source": "Crossref", - "is-referenced-by-count": 388, - "title": "Predicting effective microRNA target sites in mammalian mRNAs", - "prefix": "10.7554", - "volume": "4", - "author": [ - { - "given": "Vikram", - "family": "Agarwal", - "affiliation": [] - }, - { - "given": "George W", - "family": "Bell", - "affiliation": [] - }, - { - "given": "Jin-Wu", - "family": "Nam", - "affiliation": [] - }, - { - "given": "David P", - "family": "Bartel", - "affiliation": [] - } - ], - "member": "4374", - "published-online": { - "date-parts": [ - [ - 2015, - 8, - 12 - ] - ] - }, - "container-title": "eLife", - "original-title": [], - "link": [ - { - "URL": "http://elifesciences.org/lookup/doi/10.7554/eLife.05005", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 23 - ] - ], - "date-time": "2017-06-23T17:28:53Z", - "timestamp": 1498238933000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 8, - 12 - ] - ] - }, - "references-count": 123, - "alternative-id": [ - "10.7554/eLife.05005" - ], - "URL": "https://doi.org/10.7554/elife.05005", - "relation": { - "cites": [] - }, - "subject": [ - "General Biochemistry, Genetics and Molecular Biology", - "General Immunology and Microbiology", - "General Neuroscience", - "General Medicine" - ], - "id": "12vPQi3gp" - }, - "citation_id": "12vPQi3gp" - }, - "doi:10.1038/nbt.3300": { - "source": "doi", - "identifer": "10.1038/nbt.3300", - "standard_citation": "doi:10.1038/nbt.3300", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T14:02:16Z", - "timestamp": 1507039336004 - }, - "reference-count": 51, - "publisher": "Springer Nature", - "issue": "8", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/nbt.3300", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 7, - 27 - ] - ], - "date-time": "2015-07-27T15:41:12Z", - "timestamp": 1438011672000 - }, - "page": "831-838", - "source": "Crossref", - "is-referenced-by-count": 131, - "title": "Predicting the sequence specificities of DNA- and RNA-binding proteins by deep learning", - "prefix": "10.1038", - "volume": "33", - "author": [ - { - "given": "Babak", - "family": "Alipanahi", - "affiliation": [] - }, - { - "given": "Andrew", - "family": "Delong", - "affiliation": [] - }, - { - "given": "Matthew T", - "family": "Weirauch", - "affiliation": [] - }, - { - "given": "Brendan J", - "family": "Frey", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2015, - 7, - 27 - ] - ] - }, - "container-title": "Nature Biotechnology", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2015, - 8, - 7 - ] - ], - "date-time": "2015-08-07T16:59:41Z", - "timestamp": 1438966781000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 7, - 27 - ] - ] - }, - "references-count": 51, - "alternative-id": [ - "nbt.3300" - ], - "URL": "https://doi.org/10.1038/nbt.3300", - "relation": { - "cites": [] - }, - "subject": [ - "Biotechnology", - "Molecular Medicine", - "Applied Microbiology and Biotechnology", - "Bioengineering", - "Biomedical Engineering" - ], - "container-title-short": "Nat Biotechnol", - "id": "jJHZHWrl" - }, - "citation_id": "jJHZHWrl" - }, - "doi:10.1021/acscentsci.6b00367": { - "source": "doi", - "identifer": "10.1021/acscentsci.6b00367", - "standard_citation": "doi:10.1021/acscentsci.6b00367", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 20 - ] - ], - "date-time": "2017-09-20T17:42:08Z", - "timestamp": 1505929328749 - }, - "reference-count": 24, - "publisher": "American Chemical Society (ACS)", - "issue": "4", - "license": [ - { - "URL": "http://pubs.acs.org/page/policy/authorchoice_termsofuse.html", - "start": { - "date-parts": [ - [ - 2017, - 4, - 3 - ] - ], - "date-time": "2017-04-03T00:00:00Z", - "timestamp": 1491177600000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/100005883", - "name": "Hertz Foundation", - "doi-asserted-by": "publisher", - "award": [] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2017, - 4, - 26 - ] - ] - }, - "DOI": "10.1021/acscentsci.6b00367", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2017, - 4, - 3 - ] - ], - "date-time": "2017-04-03T12:00:27Z", - "timestamp": 1491220827000 - }, - "page": "283-293", - "source": "Crossref", - "is-referenced-by-count": 6, - "title": "Low Data Drug Discovery with One-Shot Learning", - "prefix": "10.1021", - "volume": "3", - "author": [ - { - "given": "Han", - "family": "Altae-Tran", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0001-8450-4262", - "authenticated-orcid": true, - "given": "Bharath", - "family": "Ramsundar", - "affiliation": [] - }, - { - "given": "Aneesh S.", - "family": "Pappu", - "affiliation": [] - }, - { - "given": "Vijay", - "family": "Pande", - "affiliation": [] - } - ], - "member": "316", - "published-online": { - "date-parts": [ - [ - 2017, - 4, - 3 - ] - ] - }, - "container-title": "ACS Central Science", - "original-title": [], - "link": [ - { - "URL": "http://pubs.acs.org/doi/pdf/10.1021/acscentsci.6b00367", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "unspecified" - }, - { - "URL": "http://pubs.acs.org/doi/pdf/10.1021/acscentsci.6b00367", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 4, - 26 - ] - ], - "date-time": "2017-04-26T08:25:34Z", - "timestamp": 1493195134000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 4, - 3 - ] - ] - }, - "references-count": 24, - "alternative-id": [ - "10.1021/acscentsci.6b00367" - ], - "URL": "https://doi.org/10.1021/acscentsci.6b00367", - "relation": {}, - "container-title-short": "ACS Cent. Sci.", - "id": "P4ixsM8i" - }, - "citation_id": "P4ixsM8i" - }, - "doi:10.1117/12.2249981": { - "source": "doi", - "identifer": "10.1117/12.2249981", - "standard_citation": "doi:10.1117/12.2249981", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 11 - ] - ], - "date-time": "2017-08-11T00:52:27Z", - "timestamp": 1502412747654 - }, - "reference-count": 0, - "publisher": "SPIE", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2017, - 3, - 3 - ] - ] - }, - "DOI": "10.1117/12.2249981", - "type": "paper-conference", - "created": { - "date-parts": [ - [ - 2017, - 3, - 3 - ] - ], - "date-time": "2017-03-03T20:31:36Z", - "timestamp": 1488573096000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "\nClassification of breast MRI lesions using small-size training sets: comparison of deep learning approaches\n", - "prefix": "10.1117", - "author": [ - { - "given": "Guy", - "family": "Amit", - "affiliation": [ - { - "name": "IBM Research - Haifa (Israel)" - } - ] - }, - { - "given": "Rami", - "family": "Ben-Ari", - "affiliation": [ - { - "name": "IBM Research - Haifa (Israel)" - } - ] - }, - { - "given": "Omer", - "family": "Hadad", - "affiliation": [ - { - "name": "IBM Research - Haifa (Israel)" - } - ] - }, - { - "given": "Einat", - "family": "Monovich", - "affiliation": [ - { - "name": "IBM Research - Haifa (Israel)" - } - ] - }, - { - "given": "Noa", - "family": "Granot", - "affiliation": [ - { - "name": "IBM Research - Haifa (Israel)" - } - ] - }, - { - "given": "Sharbell", - "family": "Hashoul", - "affiliation": [ - { - "name": "IBM Research - Haifa (Israel)" - } - ] - } - ], - "member": "189", - "container-title": "Medical Imaging 2017: Computer-Aided Diagnosis", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 5, - 30 - ] - ], - "date-time": "2017-05-30T16:35:25Z", - "timestamp": 1496162125000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Samuel G.", - "family": "Armato", - "affiliation": [ - { - "name": "The Univ. of Chicago (United States)" - } - ] - }, - { - "given": "Nicholas A.", - "family": "Petrick", - "affiliation": [ - { - "name": "U.S. Food and Drug Administration (United States)" - } - ] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 3, - 3 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1117/12.2249981", - "relation": {}, - "id": "SOi9mAC2" - }, - "citation_id": "SOi9mAC2" - }, - "doi:10.1186/s13059-017-1189-z": { - "source": "doi", - "identifer": "10.1186/s13059-017-1189-z", - "standard_citation": "doi:10.1186/s13059-017-1189-z", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 22 - ] - ], - "date-time": "2017-09-22T00:02:17Z", - "timestamp": 1506038537148 - }, - "reference-count": 72, - "publisher": "Springer Nature", - "issue": "1", - "content-domain": { - "domain": [ - "link.springer.com" - ], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2017, - 12 - ] - ] - }, - "DOI": "10.1186/s13059-017-1189-z", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2017, - 4, - 10 - ] - ], - "date-time": "2017-04-10T15:24:58Z", - "timestamp": 1491837898000 - }, - "update-policy": "http://dx.doi.org/10.1007/springer_crossmark_policy", - "source": "Crossref", - "is-referenced-by-count": 3, - "title": "DeepCpG: accurate prediction of single-cell DNA methylation states using deep learning", - "prefix": "10.1186", - "volume": "18", - "author": [ - { - "given": "Christof", - "family": "Angermueller", - "affiliation": [] - }, - { - "given": "Heather J.", - "family": "Lee", - "affiliation": [] - }, - { - "given": "Wolf", - "family": "Reik", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0003-0507-1839", - "authenticated-orcid": false, - "given": "Oliver", - "family": "Stegle", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2017, - 4, - 11 - ] - ] - }, - "container-title": "Genome Biology", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1186/s13059-017-1189-z.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 5, - 12 - ] - ], - "date-time": "2017-05-12T09:15:58Z", - "timestamp": 1494580558000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 4, - 11 - ] - ] - }, - "references-count": 72, - "alternative-id": [ - "1189" - ], - "URL": "https://doi.org/10.1186/s13059-017-1189-z", - "relation": { - "has-preprint": [ - { - "id-type": "doi", - "id": "10.1101/055715", - "asserted-by": "object" - } - ], - "cites": [] - }, - "container-title-short": "Genome Biol", - "article-number": "67", - "id": "19EJTHByG" - }, - "citation_id": "19EJTHByG" - }, - "doi:10.1101/095653": { - "source": "doi", - "identifer": "10.1101/095653", - "standard_citation": "doi:10.1101/095653", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T05:44:11Z", - "timestamp": 1502343851192 - }, - "posted": { - "date-parts": [ - [ - 2016, - 12, - 20 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2016, - 12, - 29 - ] - ] - }, - "abstract": "Despite many recent advances in systems biology and a marked increase in the availability of high-throughput biological data, the productivity of research and development in the pharmaceutical industry is on the decline. This is primarily due to clinical trial failure rates reaching up to 95% in oncology and other disease areas. We have developed a comprehensive analytical and computational pipeline utilizing deep learning techniques and novel systems biology analytical tools to predict the outcomes of phase I/II clinical trials. The pipeline predicts the side effects of a drug using deep neural networks and estimates drug-induced pathway activation. It then uses the predicted side effect probabilities and pathway activation scores as an input to train a classifier which predicts clinical trial outcomes. This classifier was trained on 577 transcriptomic datasets and has achieved a cross-validated accuracy of 0.83. When compared to a direct gene-based classifier, our multi-stage approach dramatically improves the accuracy of the predictions. The classifier was applied to a set of compounds currently present in the pipelines of several major pharmaceutical companies to highlight potential risks in their portfolios and estimate the fraction of clinical trials that were likely to fail in phase I and II.", - "DOI": "10.1101/095653", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2016, - 12, - 21 - ] - ], - "date-time": "2016-12-21T06:10:11Z", - "timestamp": 1482300611000 - }, - "source": "Crossref", - "is-referenced-by-count": 1, - "title": "Integrated deep learned transcriptomic and structure-based predictor of clinical trials outcomes", - "prefix": "10.1101", - "author": [ - { - "given": "Artem V", - "family": "Artemov", - "affiliation": [] - }, - { - "given": "Evgeny", - "family": "Putin", - "affiliation": [] - }, - { - "given": "Quentin", - "family": "Vanhaelen", - "affiliation": [] - }, - { - "given": "Alexander", - "family": "Aliper", - "affiliation": [] - }, - { - "given": "Ivan V", - "family": "Ozerov", - "affiliation": [] - }, - { - "given": "Alex", - "family": "Zhavoronkov", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/095653", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 1, - 4 - ] - ], - "date-time": "2017-01-04T06:18:59Z", - "timestamp": 1483510739000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 12, - 20 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/095653", - "relation": {}, - "id": "mo3GQwJj" - }, - "citation_id": "mo3GQwJj" - }, - "doi:10.1101/046508": { - "source": "doi", - "identifer": "10.1101/046508", - "standard_citation": "doi:10.1101/046508", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T06:11:29Z", - "timestamp": 1502259089826 - }, - "posted": { - "date-parts": [ - [ - 2016, - 3, - 31 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2016, - 3, - 31 - ] - ] - }, - "abstract": "Rare cell populations play a pivotal role in the initiation and progression of diseases like cancer. However, the identification of such subpopulations remains a difficult task. This work describes CellCnn, a representation learning approach to detect rare cell subsets associated with disease using high dimensional single cell measurements. Using CellCnn, we identify paracrine signaling and AIDS onset associated cell subsets in peripheral blood, and minimal residual disease associated populations in leukemia with frequencies as low as 0.005%.", - "DOI": "10.1101/046508", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2016, - 4, - 1 - ] - ], - "date-time": "2016-04-01T05:45:18Z", - "timestamp": 1459489518000 - }, - "source": "Crossref", - "is-referenced-by-count": 1, - "title": "Sensitive detection of rare disease-associated cell subsets via representation learning.", - "prefix": "10.1101", - "author": [ - { - "given": "Eirini", - "family": "Arvaniti", - "affiliation": [] - }, - { - "given": "Manfred", - "family": "Claassen", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/046508", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 7, - 20 - ] - ], - "date-time": "2017-07-20T18:56:23Z", - "timestamp": 1500576983000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 3, - 31 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/046508", - "relation": { - "is-preprint-of": [ - { - "id-type": "doi", - "id": "10.1038/ncomms14825", - "asserted-by": "subject" - } - ] - }, - "id": "r3Gbjksq" - }, - "citation_id": "r3Gbjksq" - }, - "doi:10.1371/journal.pone.0141287": { - "source": "doi", - "identifer": "10.1371/journal.pone.0141287", - "standard_citation": "doi:10.1371/journal.pone.0141287", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 15 - ] - ], - "date-time": "2017-09-15T14:42:17Z", - "timestamp": 1505486537943 - }, - "reference-count": 39, - "publisher": "Public Library of Science (PLoS)", - "issue": "11", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2015, - 11, - 10 - ] - ], - "date-time": "2015-11-10T00:00:00Z", - "timestamp": 1447113600000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "www.plosone.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pone.0141287", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 11, - 10 - ] - ], - "date-time": "2015-11-10T19:08:46Z", - "timestamp": 1447182526000 - }, - "page": "e0141287", - "update-policy": "http://dx.doi.org/10.1371/journal.pone.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 7, - "title": "Continuous Distributed Representation of Biological Sequences for Deep Proteomics and Genomics", - "prefix": "10.1371", - "volume": "10", - "author": [ - { - "given": "Ehsaneddin", - "family": "Asgari", - "affiliation": [] - }, - { - "given": "Mohammad R. K.", - "family": "Mofrad", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2015, - 11, - 10 - ] - ] - }, - "container-title": "PLOS ONE", - "original-title": [], - "link": [ - { - "URL": "http://dx.plos.org/10.1371/journal.pone.0141287", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 23 - ] - ], - "date-time": "2017-06-23T23:20:53Z", - "timestamp": 1498260053000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Firas H", - "family": "Kobeissy", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 11, - 10 - ] - ] - }, - "references-count": 39, - "URL": "https://doi.org/10.1371/journal.pone.0141287", - "relation": { - "cites": [] - }, - "subject": [ - "General Biochemistry, Genetics and Molecular Biology", - "General Agricultural and Biological Sciences", - "General Medicine" - ], - "container-title-short": "PLoS ONE", - "id": "1E1PWjqTm" - }, - "citation_id": "1E1PWjqTm" - }, - "doi:10.1371/journal.pone.0130140": { - "source": "doi", - "identifer": "10.1371/journal.pone.0130140", - "standard_citation": "doi:10.1371/journal.pone.0130140", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T02:42:17Z", - "timestamp": 1506998537418 - }, - "reference-count": 60, - "publisher": "Public Library of Science (PLoS)", - "issue": "7", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2015, - 7, - 10 - ] - ], - "date-time": "2015-07-10T00:00:00Z", - "timestamp": 1436486400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "www.plosone.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pone.0130140", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 7, - 10 - ] - ], - "date-time": "2015-07-10T18:18:19Z", - "timestamp": 1436552299000 - }, - "page": "e0130140", - "update-policy": "http://dx.doi.org/10.1371/journal.pone.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 29, - "title": "On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer-Wise Relevance Propagation", - "prefix": "10.1371", - "volume": "10", - "author": [ - { - "given": "Sebastian", - "family": "Bach", - "affiliation": [] - }, - { - "given": "Alexander", - "family": "Binder", - "affiliation": [] - }, - { - "given": "Grégoire", - "family": "Montavon", - "affiliation": [] - }, - { - "given": "Frederick", - "family": "Klauschen", - "affiliation": [] - }, - { - "given": "Klaus-Robert", - "family": "Müller", - "affiliation": [] - }, - { - "given": "Wojciech", - "family": "Samek", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2015, - 7, - 10 - ] - ] - }, - "container-title": "PLOS ONE", - "original-title": [], - "link": [ - { - "URL": "http://dx.plos.org/10.1371/journal.pone.0130140", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 23 - ] - ], - "date-time": "2017-06-23T15:28:58Z", - "timestamp": 1498231738000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Oscar Deniz", - "family": "Suarez", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 7, - 10 - ] - ] - }, - "references-count": 60, - "URL": "https://doi.org/10.1371/journal.pone.0130140", - "relation": { - "cites": [] - }, - "subject": [ - "General Biochemistry, Genetics and Molecular Biology", - "General Agricultural and Biological Sciences", - "General Medicine" - ], - "container-title-short": "PLoS ONE", - "id": "au5CLIOH" - }, - "citation_id": "au5CLIOH" - }, - "arxiv:1409.0473": { - "source": "arxiv", - "identifer": "1409.0473", - "standard_citation": "arxiv:1409.0473", - "bibtex": "@article{haHzVaaz,\n abstract = {Neural machine translation is a recently proposed approach to machine\ntranslation. Unlike the traditional statistical machine translation, the neural\nmachine translation aims at building a single neural network that can be\njointly tuned to maximize the translation performance. The models proposed\nrecently for neural machine translation often belong to a family of\nencoder-decoders and consists of an encoder that encodes a source sentence into\na fixed-length vector from which a decoder generates a translation. In this\npaper, we conjecture that the use of a fixed-length vector is a bottleneck in\nimproving the performance of this basic encoder-decoder architecture, and\npropose to extend this by allowing a model to automatically (soft-)search for\nparts of a source sentence that are relevant to predicting a target word, without having to form these parts as a hard segment explicitly. With this new\napproach, we achieve a translation performance comparable to the existing\nstate-of-the-art phrase-based system on the task of English-to-French\ntranslation. Furthermore, qualitative analysis reveals that the\n(soft-)alignments found by the model agree well with our intuition.},\n archiveprefix = {arXiv},\n author = {Dzmitry Bahdanau and Kyunghyun Cho and Yoshua Bengio},\n eprint = {1409.0473v7},\n file = {1409.0473v7.pdf},\n month = {Sep},\n primaryclass = {cs.CL},\n title = {Neural Machine Translation by Jointly Learning to Align and Translate},\n url = {https://arxiv.org/abs/1409.0473v7},\n year = {2014}\n}\n\n", - "citation_id": "haHzVaaz" - }, - "doi:10.1117/12.2083124": { - "source": "doi", - "identifer": "10.1117/12.2083124", - "standard_citation": "doi:10.1117/12.2083124", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T20:40:35Z", - "timestamp": 1502311235401 - }, - "reference-count": 0, - "publisher": "SPIE", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015, - 3, - 20 - ] - ] - }, - "DOI": "10.1117/12.2083124", - "type": "paper-conference", - "created": { - "date-parts": [ - [ - 2015, - 3, - 21 - ] - ], - "date-time": "2015-03-21T01:27:02Z", - "timestamp": 1426901222000 - }, - "source": "Crossref", - "is-referenced-by-count": 4, - "title": "Deep learning with non-medical training used for chest pathology identification", - "prefix": "10.1117", - "author": [ - { - "given": "Yaniv", - "family": "Bar", - "affiliation": [] - }, - { - "given": "Idit", - "family": "Diamant", - "affiliation": [] - }, - { - "given": "Lior", - "family": "Wolf", - "affiliation": [] - }, - { - "given": "Hayit", - "family": "Greenspan", - "affiliation": [] - } - ], - "member": "189", - "container-title": "Medical Imaging 2015: Computer-Aided Diagnosis", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2015, - 12, - 3 - ] - ], - "date-time": "2015-12-03T23:31:47Z", - "timestamp": 1449185507000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Lubomir M.", - "family": "Hadjiiski", - "affiliation": [] - }, - { - "given": "Georgia D.", - "family": "Tourassi", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 3, - 20 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1117/12.2083124", - "relation": {}, - "id": "1Fy5bcnCI" - }, - "citation_id": "1Fy5bcnCI" - }, - "doi:10.1038/nature09000": { - "source": "doi", - "identifer": "10.1038/nature09000", - "standard_citation": "doi:10.1038/nature09000", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T11:22:17Z", - "timestamp": 1507116137279 - }, - "reference-count": 50, - "publisher": "Springer Nature", - "issue": "7294", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2010, - 5, - 6 - ] - ] - }, - "DOI": "10.1038/nature09000", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2010, - 5, - 4 - ] - ], - "date-time": "2010-05-04T20:20:31Z", - "timestamp": 1273004431000 - }, - "page": "53-59", - "source": "Crossref", - "is-referenced-by-count": 341, - "title": "Deciphering the splicing code", - "prefix": "10.1038", - "volume": "465", - "author": [ - { - "given": "Yoseph", - "family": "Barash", - "affiliation": [] - }, - { - "given": "John A.", - "family": "Calarco", - "affiliation": [] - }, - { - "given": "Weijun", - "family": "Gao", - "affiliation": [] - }, - { - "given": "Qun", - "family": "Pan", - "affiliation": [] - }, - { - "given": "Xinchen", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Ofer", - "family": "Shai", - "affiliation": [] - }, - { - "given": "Benjamin J.", - "family": "Blencowe", - "affiliation": [] - }, - { - "given": "Brendan J.", - "family": "Frey", - "affiliation": [] - } - ], - "member": "339", - "container-title": "Nature", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 19 - ] - ], - "date-time": "2017-06-19T07:48:40Z", - "timestamp": 1497858520000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2010, - 5, - 6 - ] - ] - }, - "references-count": 50, - "alternative-id": [ - "nature09000" - ], - "URL": "https://doi.org/10.1038/nature09000", - "relation": { - "cites": [] - }, - "subject": [ - "Multidisciplinary" - ], - "container-title-short": "Nature", - "id": "11ETDdRKr" - }, - "citation_id": "11ETDdRKr" - }, - "doi:10.7326/0003-4819-115-11-843": { - "source": "doi", - "identifer": "10.7326/0003-4819-115-11-843", - "standard_citation": "doi:10.7326/0003-4819-115-11-843", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 20 - ] - ], - "date-time": "2017-09-20T20:42:15Z", - "timestamp": 1505940135196 - }, - "reference-count": 0, - "publisher": "American College of Physicians", - "issue": "11", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 1991, - 12, - 1 - ] - ] - }, - "DOI": "10.7326/0003-4819-115-11-843", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2013, - 4, - 12 - ] - ], - "date-time": "2013-04-12T16:21:18Z", - "timestamp": 1365783678000 - }, - "page": "843", - "source": "Crossref", - "is-referenced-by-count": 222, - "title": "Use of an Artificial Neural Network for the Diagnosis of Myocardial Infarction", - "prefix": "10.7326", - "volume": "115", - "author": [ - { - "given": "William G.", - "family": "Baxt", - "affiliation": [] - } - ], - "member": "4285", - "container-title": "Annals of Internal Medicine", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2013, - 4, - 12 - ] - ], - "date-time": "2013-04-12T16:21:18Z", - "timestamp": 1365783678000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 1991, - 12, - 1 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.7326/0003-4819-115-11-843", - "relation": {}, - "subject": [ - "Internal Medicine", - "General Medicine" - ], - "container-title-short": "Ann Intern Med", - "id": "jdg2u7bX" - }, - "citation_id": "jdg2u7bX" - }, - "arxiv:1412.7024": { - "source": "arxiv", - "identifer": "1412.7024", - "standard_citation": "arxiv:1412.7024", - "bibtex": "@article{1G3owNNps,\n abstract = {Multipliers are the most space and power-hungry arithmetic operators of the\ndigital implementation of deep neural networks. We train a set of\nstate-of-the-art neural networks (Maxout networks) on three benchmark datasets:\nMNIST, CIFAR-10 and SVHN. They are trained with three distinct formats:\nfloating point, fixed point and dynamic fixed point. For each of those datasets\nand for each of those formats, we assess the impact of the precision of the\nmultiplications on the final error after training. We find that very low\nprecision is sufficient not just for running trained networks but also for\ntraining them. For example, it is possible to train Maxout networks with 10\nbits multiplications.},\n archiveprefix = {arXiv},\n author = {Matthieu Courbariaux and Yoshua Bengio and Jean-Pierre David},\n eprint = {1412.7024v5},\n file = {1412.7024v5.pdf},\n month = {12},\n primaryclass = {cs.LG},\n title = {Training deep neural networks with low precision multiplications},\n url = {https://arxiv.org/abs/1412.7024v5},\n year = {2014}\n}\n\n", - "citation_id": "1G3owNNps" - }, - "doi:10.1038/nrg3079": { - "source": "doi", - "identifer": "10.1038/nrg3079", - "standard_citation": "doi:10.1038/nrg3079", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 19 - ] - ], - "date-time": "2017-09-19T16:02:28Z", - "timestamp": 1505836948900 - }, - "reference-count": 171, - "publisher": "Springer Nature", - "issue": "12", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/nrg3079", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2011, - 11, - 18 - ] - ], - "date-time": "2011-11-18T05:51:56Z", - "timestamp": 1321595516000 - }, - "page": "846-860", - "source": "Crossref", - "is-referenced-by-count": 280, - "title": "Evolution of microRNA diversity and regulation in animals", - "prefix": "10.1038", - "volume": "12", - "author": [ - { - "given": "Eugene", - "family": "Berezikov", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2011, - 11, - 18 - ] - ] - }, - "container-title": "Nature Reviews Genetics", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 20 - ] - ], - "date-time": "2017-06-20T09:57:14Z", - "timestamp": 1497952634000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2011, - 11, - 18 - ] - ] - }, - "references-count": 171, - "alternative-id": [ - "nrg3079" - ], - "URL": "https://doi.org/10.1038/nrg3079", - "relation": { - "cites": [] - }, - "subject": [ - "Genetics(clinical)", - "Genetics", - "Molecular Biology" - ], - "container-title-short": "Nat Rev Genet", - "id": "8lpCCppx" - }, - "citation_id": "8lpCCppx" - }, - "url:https://papers.nips.cc/paper/4443-algorithms-for-hyper-parameter-optimization.pdf": { - "source": "url", - "identifer": "https://papers.nips.cc/paper/4443-algorithms-for-hyper-parameter-optimization.pdf", - "standard_citation": "url:https://papers.nips.cc/paper/4443-algorithms-for-hyper-parameter-optimization.pdf", - "citeproc": { - "type": "paper-conference", - "title": "Algorithms for Hyper-parameter Optimization", - "container-title": "Proceedings of the 24th International Conference on Neural Information Processing Systems", - "collection-title": "NIPS'11", - "publisher": "Curran Associates Inc.", - "publisher-place": "USA", - "page": "2546–2554", - "source": "ACM Digital Library", - "event-place": "USA", - "abstract": "Several recent advances to the state of the art in image classification benchmarks have come from better configurations of existing techniques rather than novel approaches to feature learning. Traditionally, hyper-parameter optimization has been the job of humans because they can be very efficient in regimes where only a few trials are possible. Presently, computer clusters and GPU processors make it possible to run more trials and we show that algorithmic approaches can find better results. We present hyper-parameter optimization results on tasks of training neural networks and deep belief networks (DBNs). We optimize hyper-parameters using random search and two new greedy sequential methods based on the expected improvement criterion. Random search has been shown to be sufficiently efficient for learning neural networks for several datasets, but we show it is unreliable for training DBNs. The sequential algorithms are applied to the most difficult DBN learning problems from [1] and find significantly better results than the best previously reported. This work contributes novel techniques for making response surface models P(y|x) in which many elements of hyper-parameter assignment (x) are known to be irrelevant given particular values of other elements.", - "URL": "http://dl.acm.org/citation.cfm?id=2986459.2986743", - "author": [ - { - "family": "Bergstra", - "given": "James" - }, - { - "family": "Bardenet", - "given": "Rémi" - }, - { - "family": "Bengio", - "given": "Yoshua" - }, - { - "family": "Kégl", - "given": "Balázs" - } - ], - "issued": { - "date-parts": [ - [ - "2011" - ] - ] - }, - "accessed": { - "date-parts": [ - [ - "2017", - 5, - 23 - ] - ] - }, - "id": "wz83yfHF" - }, - "citation_id": "wz83yfHF" - }, - "url:http://www.jmlr.org/papers/v13/bergstra12a.html": { - "source": "url", - "identifer": "http://www.jmlr.org/papers/v13/bergstra12a.html", - "standard_citation": "url:http://www.jmlr.org/papers/v13/bergstra12a.html", - "citeproc": { - "URL": "http://www.jmlr.org/papers/v13/bergstra12a.html", - "title": "Random Search for Hyper-Parameter Optimization", - "container-title": "Journal of Machine Learning Research", - "issued": { - "date-parts": [ - [ - 2012 - ] - ] - }, - "author": [ - { - "family": "Bergstra", - "given": "James" - }, - { - "family": "Bengio", - "given": "Yoshua" - } - ], - "greycite-status": "Scanned", - "greycite-scanned": "2017-05-17 02:21:09", - "archives": [ - "http://wayback.archive.org/web/http://www.jmlr.org/papers/v13/bergstra12a.html" - ], - "type": "webpage", - "id": "1FSwIjR9s" - }, - "citation_id": "1FSwIjR9s" - }, - "arxiv:1603.09195": { - "source": "arxiv", - "identifer": "1603.09195", - "standard_citation": "arxiv:1603.09195", - "bibtex": "@article{1BTJ1KqRa,\n abstract = {Motivation: The MinION device by Oxford Nanopore is the first portable\nsequencing device. MinION is able to produce very long reads (reads over\n100~kBp were reported), however it suffers from high sequencing error rate. In\nthis paper, we show that the error rate can be reduced by improving the base\ncalling process.\nResults: We present the first open-source DNA base caller for the MinION\nsequencing platform by Oxford Nanopore. By employing carefully crafted\nrecurrent neural networks, our tool improves the base calling accuracy compared\nto the default base caller supplied by the manufacturer. This advance may\nfurther enhance applicability of MinION for genome sequencing and various\nclinical applications.\nAvailability: DeepNano can be downloaded at\nhttp://compbio.fmph.uniba.sk/deepnano/.\nContact: boza@fmph.uniba.sk},\n archiveprefix = {arXiv},\n author = {Vladimír Boža and Broňa Brejová and Tomáš Vinař},\n doi = {10.1371/journal.pone.0178751},\n eprint = {1603.09195v1},\n file = {1603.09195v1.pdf},\n month = {Mar},\n note = {PLoS ONE 12(6): e0178751},\n primaryclass = {q-bio.GN},\n title = {DeepNano: Deep Recurrent Neural Networks for Base Calling in MinION\nNanopore Reads},\n url = {https://arxiv.org/abs/1603.09195v1},\n year = {2016}\n}\n\n", - "citation_id": "1BTJ1KqRa" - }, - "doi:10.1038/nrg.2016.134": { - "source": "doi", - "identifer": "10.1038/nrg.2016.134", - "standard_citation": "doi:10.1038/nrg.2016.134", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 30 - ] - ], - "date-time": "2017-09-30T05:02:18Z", - "timestamp": 1506747738252 - }, - "reference-count": 233, - "publisher": "Springer Nature", - "issue": "12", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/nrg.2016.134", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 10, - 31 - ] - ], - "date-time": "2016-10-31T04:49:33Z", - "timestamp": 1477889373000 - }, - "page": "719-732", - "source": "Crossref", - "is-referenced-by-count": 22, - "title": "A network-biology perspective of microRNA function and dysfunction in cancer", - "prefix": "10.1038", - "volume": "17", - "author": [ - { - "given": "Cameron P.", - "family": "Bracken", - "affiliation": [] - }, - { - "given": "Hamish S.", - "family": "Scott", - "affiliation": [] - }, - { - "given": "Gregory J.", - "family": "Goodall", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2016, - 10, - 31 - ] - ] - }, - "container-title": "Nature Reviews Genetics", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T02:25:13Z", - "timestamp": 1498357513000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 10, - 31 - ] - ] - }, - "references-count": 233, - "alternative-id": [ - "nrg.2016.134" - ], - "URL": "https://doi.org/10.1038/nrg.2016.134", - "relation": { - "cites": [] - }, - "subject": [ - "Genetics(clinical)", - "Genetics", - "Molecular Biology" - ], - "container-title-short": "Nat Rev Genet", - "id": "yVKIhIAf" - }, - "citation_id": "yVKIhIAf" - }, - "doi:10.1109/isbi.2016.7493240": { - "source": "doi", - "identifer": "10.1109/isbi.2016.7493240", - "standard_citation": "doi:10.1109/isbi.2016.7493240", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 30 - ] - ], - "date-time": "2017-08-30T18:02:09Z", - "timestamp": 1504116129503 - }, - "reference-count": 11, - "publisher": "IEEE", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 4 - ] - ] - }, - "DOI": "10.1109/isbi.2016.7493240", - "type": "paper-conference", - "created": { - "date-parts": [ - [ - 2016, - 7, - 1 - ] - ], - "date-time": "2016-07-01T04:38:22Z", - "timestamp": 1467347902000 - }, - "source": "Crossref", - "is-referenced-by-count": 2, - "title": "Detection of age-related macular degeneration via deep learning", - "prefix": "10.1109", - "author": [ - { - "given": "P.", - "family": "Burlina", - "affiliation": [] - }, - { - "given": "D. E.", - "family": "Freund", - "affiliation": [] - }, - { - "given": "N.", - "family": "Joshi", - "affiliation": [] - }, - { - "given": "Y.", - "family": "Wolfson", - "affiliation": [] - }, - { - "given": "N. M.", - "family": "Bressler", - "affiliation": [] - } - ], - "member": "263", - "container-title": "2016 IEEE 13th International Symposium on Biomedical Imaging (ISBI)", - "original-title": [], - "link": [ - { - "URL": "http://xplorestaging.ieee.org/ielx7/7486633/7493185/07493240.pdf?arnumber=7493240", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T17:34:40Z", - "timestamp": 1498325680000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 4 - ] - ] - }, - "references-count": 11, - "URL": "https://doi.org/10.1109/isbi.2016.7493240", - "relation": {}, - "id": "iBPOt78R" - }, - "citation_id": "iBPOt78R" - }, - "arxiv:1312.6184": { - "source": "arxiv", - "identifer": "1312.6184", - "standard_citation": "arxiv:1312.6184", - "bibtex": "@article{1AhGoHZP9,\n abstract = {Currently, deep neural networks are the state of the art on problems such as\nspeech recognition and computer vision. In this extended abstract, we show that\nshallow feed-forward networks can learn the complex functions previously\nlearned by deep nets and achieve accuracies previously only achievable with\ndeep models. Moreover, in some cases the shallow neural nets can learn these\ndeep functions using a total number of parameters similar to the original deep\nmodel. We evaluate our method on the TIMIT phoneme recognition task and are\nable to train shallow fully-connected nets that perform similarly to complex, well-engineered, deep convolutional architectures. Our success in training\nshallow neural nets to mimic deeper models suggests that there probably exist\nbetter algorithms for training shallow feed-forward nets than those currently\navailable.},\n archiveprefix = {arXiv},\n author = {Lei Jimmy Ba and Rich Caruana},\n eprint = {1312.6184v7},\n file = {1312.6184v7.pdf},\n month = {12},\n primaryclass = {cs.LG},\n title = {Do Deep Nets Really Need to be Deep?},\n url = {https://arxiv.org/abs/1312.6184v7},\n year = {2013}\n}\n\n", - "citation_id": "1AhGoHZP9" - }, - "doi:10.1101/114892": { - "source": "doi", - "identifer": "10.1101/114892", - "standard_citation": "doi:10.1101/114892", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 19 - ] - ], - "date-time": "2017-09-19T05:40:04Z", - "timestamp": 1505799604209 - }, - "posted": { - "date-parts": [ - [ - 2017, - 3, - 8 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2017, - 9, - 18 - ] - ] - }, - "abstract": "Identifying robust survival subgroups of hepatocellular carcinoma (HCC) will significantly improve patient care. Currently, endeavor of integrating multi-omics data to explicitly predict HCC survival from multiple patient cohorts is lacking. To fill in this gap, we present a deep learning (DL) based model on HCC that robustly differentiates survival subpopulations of patients in six cohorts. We build the DL based, survival-sensitive model on 360 HCC patients' data using RNA-seq, miRNA-seq and methylation data from TCGA, which predicts prognosis as good as an alternative model where genomics and clinical data are both considered. This DL based model provides two optimal subgroups of patients with significant survival differences (P=7.13e-6) and good model fitness (C-index=0.68). More aggressive subtype is associated with frequent TP53 inactivation mutations, higher expression of stemness markers (KRT19, EPCAM) and tumor marker BIRC5, and activated Wnt and Akt signaling pathways. We validated this multi-omics model on five external datasets of various omics types: LIRI-JP cohort (n=230, C-index=0.75), NCI cohort (n=221, C-index=0.67), Chinese cohort (n=166, C-index=0.69), E-TABM-36 cohort (n=40, C-index=0.77), and Hawaiian cohort (n=27, C-index=0.82). This is the first study to employ deep learning to identify multi-omics features linked to the differential survival of HCC patients. Given its robustness over multiple cohorts, we expect this workflow to be useful at predicting HCC prognosis prediction.", - "DOI": "10.1101/114892", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2017, - 3, - 9 - ] - ], - "date-time": "2017-03-09T06:10:48Z", - "timestamp": 1489039848000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Deep Learning based multi-omics integration robustly predicts survival in liver cancer", - "prefix": "10.1101", - "author": [ - { - "ORCID": "http://orcid.org/0000-0002-4117-6403", - "authenticated-orcid": false, - "given": "Kumardeep", - "family": "Chaudhary", - "affiliation": [] - }, - { - "given": "Olivier B.", - "family": "Poirion", - "affiliation": [] - }, - { - "given": "Liangqun", - "family": "Lu", - "affiliation": [] - }, - { - "given": "Lana X.", - "family": "Garmire", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/114892", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 9, - 19 - ] - ], - "date-time": "2017-09-19T05:10:47Z", - "timestamp": 1505797847000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 3, - 8 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/114892", - "relation": {}, - "id": "obeRVckH" - }, - "citation_id": "obeRVckH" - }, - "arxiv:1512.03542": { - "source": "arxiv", - "identifer": "1512.03542", - "standard_citation": "arxiv:1512.03542", - "bibtex": "@article{14DAmZTDg,\n abstract = {Exponential growth in Electronic Healthcare Records (EHR) has resulted in new\nopportunities and urgent needs for discovery of meaningful data-driven\nrepresentations and patterns of diseases in Computational Phenotyping research.\nDeep Learning models have shown superior performance for robust prediction in\ncomputational phenotyping tasks, but suffer from the issue of model\ninterpretability which is crucial for clinicians involved in decision-making.\nIn this paper, we introduce a novel knowledge-distillation approach called\nInterpretable Mimic Learning, to learn interpretable phenotype features for\nmaking robust prediction while mimicking the performance of deep learning\nmodels. Our framework uses Gradient Boosting Trees to learn interpretable\nfeatures from deep learning models such as Stacked Denoising Autoencoder and\nLong Short-Term Memory. Exhaustive experiments on a real-world clinical\ntime-series dataset show that our method obtains similar or better performance\nthan the deep learning models, and it provides interpretable phenotypes for\nclinical decision making.},\n archiveprefix = {arXiv},\n author = {Zhengping Che and Sanjay Purushotham and Robinder Khemani and Yan Liu},\n eprint = {1512.03542v1},\n file = {1512.03542v1.pdf},\n month = {12},\n primaryclass = {stat.ML},\n title = {Distilling Knowledge from Deep Networks with Applications to Healthcare\nDomain},\n url = {https://arxiv.org/abs/1512.03542v1},\n year = {2015}\n}\n\n", - "citation_id": "14DAmZTDg" - }, - "arxiv:1606.01865": { - "source": "arxiv", - "identifer": "1606.01865", - "standard_citation": "arxiv:1606.01865", - "bibtex": "@article{O7Vbecm2,\n abstract = {Multivariate time series data in practical applications, such as health care, geoscience, and biology, are characterized by a variety of missing values. In\ntime series prediction and other related tasks, it has been noted that missing\nvalues and their missing patterns are often correlated with the target labels, a.k.a., informative missingness. There is very limited work on exploiting the\nmissing patterns for effective imputation and improving prediction performance.\nIn this paper, we develop novel deep learning models, namely GRU-D, as one of\nthe early attempts. GRU-D is based on Gated Recurrent Unit (GRU), a\nstate-of-the-art recurrent neural network. It takes two representations of\nmissing patterns, i.e., masking and time interval, and effectively incorporates\nthem into a deep model architecture so that it not only captures the long-term\ntemporal dependencies in time series, but also utilizes the missing patterns to\nachieve better prediction results. Experiments of time series classification\ntasks on real-world clinical datasets (MIMIC-III, PhysioNet) and synthetic\ndatasets demonstrate that our models achieve state-of-the-art performance and\nprovides useful insights for better understanding and utilization of missing\nvalues in time series analysis.},\n archiveprefix = {arXiv},\n author = {Zhengping Che and Sanjay Purushotham and Kyunghyun Cho and David Sontag and Yan Liu},\n eprint = {1606.01865v2},\n file = {1606.01865v2.pdf},\n month = {Jun},\n primaryclass = {cs.LG},\n title = {Recurrent Neural Networks for Multivariate Time Series with Missing\nValues},\n url = {https://arxiv.org/abs/1606.01865v2},\n year = {2016}\n}\n\n", - "citation_id": "O7Vbecm2" - }, - "arxiv:1504.04788": { - "source": "arxiv", - "identifer": "1504.04788", - "standard_citation": "arxiv:1504.04788", - "bibtex": "@article{15lYGmZpY,\n abstract = {As deep nets are increasingly used in applications suited for mobile devices, a fundamental dilemma becomes apparent: the trend in deep learning is to grow\nmodels to absorb ever-increasing data set sizes; however mobile devices are\ndesigned with very little memory and cannot store such large models. We present\na novel network architecture, HashedNets, that exploits inherent redundancy in\nneural networks to achieve drastic reductions in model sizes. HashedNets uses a\nlow-cost hash function to randomly group connection weights into hash buckets, and all connections within the same hash bucket share a single parameter value.\nThese parameters are tuned to adjust to the HashedNets weight sharing\narchitecture with standard backprop during training. Our hashing procedure\nintroduces no additional memory overhead, and we demonstrate on several\nbenchmark data sets that HashedNets shrink the storage requirements of neural\nnetworks substantially while mostly preserving generalization performance.},\n archiveprefix = {arXiv},\n author = {Wenlin Chen and James T. Wilson and Stephen Tyree and Kilian Q. Weinberger and Yixin Chen},\n eprint = {1504.04788v1},\n file = {1504.04788v1.pdf},\n month = {Apr},\n primaryclass = {cs.LG},\n title = {Compressing Neural Networks with the Hashing Trick},\n url = {https://arxiv.org/abs/1504.04788v1},\n year = {2015}\n}\n\n", - "citation_id": "15lYGmZpY" - }, - "doi:10.1093/bioinformatics/btv315": { - "source": "doi", - "identifer": "10.1093/bioinformatics/btv315", - "standard_citation": "doi:10.1093/bioinformatics/btv315", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T15:22:17Z", - "timestamp": 1507044137986 - }, - "reference-count": 29, - "publisher": "Oxford University Press (OUP)", - "issue": "18", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015, - 9, - 15 - ] - ] - }, - "DOI": "10.1093/bioinformatics/btv315", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 5, - 21 - ] - ], - "date-time": "2015-05-21T02:30:54Z", - "timestamp": 1432175454000 - }, - "page": "3008-3015", - "source": "Crossref", - "is-referenced-by-count": 7, - "title": "Trans-species learning of cellular signaling systems with bimodal deep belief networks", - "prefix": "10.1093", - "volume": "31", - "author": [ - { - "given": "Lujia", - "family": "Chen", - "affiliation": [] - }, - { - "given": "Chunhui", - "family": "Cai", - "affiliation": [] - }, - { - "given": "Vicky", - "family": "Chen", - "affiliation": [] - }, - { - "given": "Xinghua", - "family": "Lu", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2015, - 5, - 20 - ] - ] - }, - "container-title": "Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/bioinformatics/article-pdf/31/18/3008/17089876/btv315.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 22 - ] - ], - "date-time": "2017-08-22T15:00:23Z", - "timestamp": 1503414023000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 5, - 20 - ] - ] - }, - "references-count": 29, - "alternative-id": [ - "10.1093/bioinformatics/btv315" - ], - "URL": "https://doi.org/10.1093/bioinformatics/btv315", - "relation": {}, - "subject": [ - "Statistics and Probability", - "Computational Theory and Mathematics", - "Biochemistry", - "Molecular Biology", - "Computational Mathematics", - "Computer Science Applications" - ], - "container-title-short": "Bioinformatics", - "id": "rmjDc5rm" - }, - "citation_id": "rmjDc5rm" - }, - "doi:10.1186/s12859-015-0852-1": { - "source": "doi", - "identifer": "10.1186/s12859-015-0852-1", - "standard_citation": "doi:10.1186/s12859-015-0852-1", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T07:22:12Z", - "timestamp": 1507015332880 - }, - "reference-count": 29, - "publisher": "Springer Nature", - "issue": "S1", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2016, - 1, - 11 - ] - ], - "date-time": "2016-01-11T00:00:00Z", - "timestamp": 1452470400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "link.springer.com" - ], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 12 - ] - ] - }, - "DOI": "10.1186/s12859-015-0852-1", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 1, - 11 - ] - ], - "date-time": "2016-01-11T10:00:45Z", - "timestamp": 1452506445000 - }, - "update-policy": "http://dx.doi.org/10.1007/springer_crossmark_policy", - "source": "Crossref", - "is-referenced-by-count": 2, - "title": "Learning a hierarchical representation of the yeast transcriptomic machinery using an autoencoder model", - "prefix": "10.1186", - "volume": "17", - "author": [ - { - "given": "Lujia", - "family": "Chen", - "affiliation": [] - }, - { - "given": "Chunhui", - "family": "Cai", - "affiliation": [] - }, - { - "given": "Vicky", - "family": "Chen", - "affiliation": [] - }, - { - "given": "Xinghua", - "family": "Lu", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2016, - 1, - 11 - ] - ] - }, - "container-title": "BMC Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1186/s12859-015-0852-1.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://link.springer.com/article/10.1186/s12859-015-0852-1/fulltext.html", - "content-type": "text/html", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://link.springer.com/content/pdf/10.1186/s12859-015-0852-1", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T03:26:04Z", - "timestamp": 1498274764000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 1, - 11 - ] - ] - }, - "references-count": 29, - "alternative-id": [ - "852" - ], - "URL": "https://doi.org/10.1186/s12859-015-0852-1", - "relation": { - "cites": [] - }, - "subject": [ - "Biochemistry", - "Applied Mathematics", - "Molecular Biology", - "Structural Biology", - "Computer Science Applications" - ], - "container-title-short": "BMC Bioinformatics", - "article-number": "S9", - "id": "yVBx9Qx4" - }, - "citation_id": "yVBx9Qx4" - }, - "doi:10.1093/bioinformatics/btw074": { - "source": "doi", - "identifer": "10.1093/bioinformatics/btw074", - "standard_citation": "doi:10.1093/bioinformatics/btw074", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 25 - ] - ], - "date-time": "2017-08-25T04:17:39Z", - "timestamp": 1503634659572 - }, - "reference-count": 35, - "publisher": "Oxford University Press (OUP)", - "issue": "12", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 6, - 15 - ] - ] - }, - "DOI": "10.1093/bioinformatics/btw074", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 2, - 15 - ] - ], - "date-time": "2016-02-15T01:09:07Z", - "timestamp": 1455498547000 - }, - "page": "1832-1839", - "source": "Crossref", - "is-referenced-by-count": 16, - "title": "Gene expression inference with deep learning", - "prefix": "10.1093", - "volume": "32", - "author": [ - { - "given": "Yifei", - "family": "Chen", - "affiliation": [] - }, - { - "given": "Yi", - "family": "Li", - "affiliation": [] - }, - { - "given": "Rajiv", - "family": "Narayan", - "affiliation": [] - }, - { - "given": "Aravind", - "family": "Subramanian", - "affiliation": [] - }, - { - "given": "Xiaohui", - "family": "Xie", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2016, - 2, - 11 - ] - ] - }, - "container-title": "Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/bioinformatics/article-pdf/32/12/1832/", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 24 - ] - ], - "date-time": "2017-08-24T02:29:38Z", - "timestamp": 1503541778000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 2, - 11 - ] - ] - }, - "references-count": 35, - "alternative-id": [ - "10.1093/bioinformatics/btw074" - ], - "URL": "https://doi.org/10.1093/bioinformatics/btw074", - "relation": {}, - "subject": [ - "Statistics and Probability", - "Computational Theory and Mathematics", - "Biochemistry", - "Molecular Biology", - "Computational Mathematics", - "Computer Science Applications" - ], - "container-title-short": "Bioinformatics", - "id": "12QQw9p7v" - }, - "citation_id": "12QQw9p7v" - }, - "arxiv:1611.07012": { - "source": "arxiv", - "identifer": "1611.07012", - "standard_citation": "arxiv:1611.07012", - "bibtex": "@article{10nDTiETi,\n abstract = {Deep learning methods exhibit promising performance for predictive modeling\nin healthcare, but two important challenges remain: -Data insufficiency:Often\nin healthcare predictive modeling, the sample size is insufficient for deep\nlearning methods to achieve satisfactory results. -Interpretation:The\nrepresentations learned by deep learning methods should align with medical\nknowledge. To address these challenges, we propose a GRaph-based Attention\nModel, GRAM that supplements electronic health records (EHR) with hierarchical\ninformation inherent to medical ontologies. Based on the data volume and the\nontology structure, GRAM represents a medical concept as a combination of its\nancestors in the ontology via an attention mechanism. We compared predictive\nperformance (i.e. accuracy, data needs, interpretability) of GRAM to various\nmethods including the recurrent neural network (RNN) in two sequential\ndiagnoses prediction tasks and one heart failure prediction task. Compared to\nthe basic RNN, GRAM achieved 10% higher accuracy for predicting diseases rarely\nobserved in the training data and 3% improved area under the ROC curve for\npredicting heart failure using an order of magnitude less training data.\nAdditionally, unlike other methods, the medical concept representations learned\nby GRAM are well aligned with the medical ontology. Finally, GRAM exhibits\nintuitive attention behaviors by adaptively generalizing to higher level\nconcepts when facing data insufficiency at the lower level concepts.},\n archiveprefix = {arXiv},\n author = {Edward Choi and Mohammad Taha Bahadori and Le Song and Walter F. Stewart and Jimeng Sun},\n eprint = {1611.07012v3},\n file = {1611.07012v3.pdf},\n month = {Dec},\n primaryclass = {cs.LG},\n title = {GRAM: Graph-based Attention Model for Healthcare Representation Learning},\n url = {https://arxiv.org/abs/1611.07012v3},\n year = {2016}\n}\n\n", - "citation_id": "10nDTiETi" - }, - "arxiv:1608.05745": { - "source": "arxiv", - "identifer": "1608.05745", - "standard_citation": "arxiv:1608.05745", - "bibtex": "@article{UcRbawKo,\n abstract = {Accuracy and interpretability are two dominant features of successful\npredictive models. Typically, a choice must be made in favor of complex black\nbox models such as recurrent neural networks (RNN) for accuracy versus less\naccurate but more interpretable traditional models such as logistic regression.\nThis tradeoff poses challenges in medicine where both accuracy and\ninterpretability are important. We addressed this challenge by developing the\nREverse Time AttentIoN model (RETAIN) for application to Electronic Health\nRecords (EHR) data. RETAIN achieves high accuracy while remaining clinically\ninterpretable and is based on a two-level neural attention model that detects\ninfluential past visits and significant clinical variables within those visits\n(e.g. key diagnoses). RETAIN mimics physician practice by attending the EHR\ndata in a reverse time order so that recent clinical visits are likely to\nreceive higher attention. RETAIN was tested on a large health system EHR\ndataset with 14 million visits completed by 263K patients over an 8 year period\nand demonstrated predictive accuracy and computational scalability comparable\nto state-of-the-art methods such as RNN, and ease of interpretability\ncomparable to traditional models.},\n archiveprefix = {arXiv},\n author = {Edward Choi and Mohammad Taha Bahadori and Joshua A. Kulas and Andy Schuetz and Walter F. Stewart and Jimeng Sun},\n eprint = {1608.05745v4},\n file = {1608.05745v4.pdf},\n month = {Aug},\n primaryclass = {cs.LG},\n title = {RETAIN: An Interpretable Predictive Model for Healthcare using Reverse\nTime Attention Mechanism},\n url = {https://arxiv.org/abs/1608.05745v4},\n year = {2016}\n}\n\n", - "citation_id": "UcRbawKo" - }, - "arxiv:1610.02357": { - "source": "arxiv", - "identifer": "1610.02357", - "standard_citation": "arxiv:1610.02357", - "bibtex": "@article{VMkPJjVk,\n abstract = {We present an interpretation of Inception modules in convolutional neural\nnetworks as being an intermediate step in-between regular convolution and the\ndepthwise separable convolution operation (a depthwise convolution followed by\na pointwise convolution). In this light, a depthwise separable convolution can\nbe understood as an Inception module with a maximally large number of towers.\nThis observation leads us to propose a novel deep convolutional neural network\narchitecture inspired by Inception, where Inception modules have been replaced\nwith depthwise separable convolutions. We show that this architecture, dubbed\nXception, slightly outperforms Inception V3 on the ImageNet dataset (which\nInception V3 was designed for), and significantly outperforms Inception V3 on a\nlarger image classification dataset comprising 350 million images and 17,000\nclasses. Since the Xception architecture has the same number of parameters as\nInception V3, the performance gains are not due to increased capacity but\nrather to a more efficient use of model parameters.},\n archiveprefix = {arXiv},\n author = {François Chollet},\n eprint = {1610.02357v3},\n file = {1610.02357v3.pdf},\n month = {Nov},\n primaryclass = {cs.CV},\n title = {Xception: Deep Learning with Depthwise Separable Convolutions},\n url = {https://arxiv.org/abs/1610.02357v3},\n year = {2016}\n}\n\n", - "citation_id": "VMkPJjVk" - }, - "doi:10.1109/72.478409": { - "source": "doi", - "identifer": "10.1109/72.478409", - "standard_citation": "doi:10.1109/72.478409", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 11 - ] - ], - "date-time": "2017-09-11T23:22:19Z", - "timestamp": 1505172139209 - }, - "reference-count": 7, - "publisher": "Institute of Electrical and Electronics Engineers (IEEE)", - "issue": "1", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 1996 - ] - ] - }, - "DOI": "10.1109/72.478409", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2002, - 8, - 24 - ] - ], - "date-time": "2002-08-24T19:16:32Z", - "timestamp": 1030216592000 - }, - "page": "229-232", - "source": "Crossref", - "is-referenced-by-count": 98, - "title": "Confidence interval prediction for neural network models", - "prefix": "10.1109", - "volume": "7", - "author": [ - { - "given": "G.", - "family": "Chryssolouris", - "affiliation": [] - }, - { - "given": "M.", - "family": "Lee", - "affiliation": [] - }, - { - "given": "A.", - "family": "Ramsey", - "affiliation": [] - } - ], - "member": "263", - "container-title": "IEEE Transactions on Neural Networks", - "original-title": [], - "link": [ - { - "URL": "http://xplorestaging.ieee.org/ielx4/72/10170/00478409.pdf?arnumber=478409", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 3, - 9 - ] - ], - "date-time": "2017-03-09T11:16:32Z", - "timestamp": 1489058192000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 1996 - ] - ] - }, - "references-count": 7, - "URL": "https://doi.org/10.1109/72.478409", - "relation": {}, - "subject": [ - "Computer Networks and Communications", - "Software", - "Artificial Intelligence", - "General Medicine", - "Computer Science Applications" - ], - "container-title-short": "IEEE Trans. Neural Netw.", - "id": "9SnNyc8Y" - }, - "citation_id": "9SnNyc8Y" - }, - "url:http://www.jmlr.org/proceedings/papers/v28/coates13.html": { - "source": "url", - "identifer": "http://www.jmlr.org/proceedings/papers/v28/coates13.html", - "standard_citation": "url:http://www.jmlr.org/proceedings/papers/v28/coates13.html", - "citeproc": { - "URL": "http://www.jmlr.org/proceedings/papers/v28/coates13.html", - "greycite-canonical-uri": "http://proceedings.mlr.press/v28/coates13.html", - "title": "Deep learning with COTS HPC systems", - "issued": { - "date-parts": [ - [ - 2013, - 2, - 13 - ] - ] - }, - "author": [ - { - "family": "Coates", - "given": "Adam" - }, - { - "family": "Huval", - "given": "Brody" - }, - { - "family": "Wang", - "given": "Tao" - }, - { - "family": "Wu", - "given": "David" - }, - { - "family": "Catanzaro", - "given": "Bryan" - }, - { - "family": "Andrew", - "given": "Ng" - } - ], - "greycite-status": "Scanned", - "greycite-scanned": "2017-05-17 02:09:12", - "type": "webpage", - "id": "4MZ2tmZ8" - }, - "citation_id": "4MZ2tmZ8" - }, - "arxiv:1610.04662": { - "source": "arxiv", - "identifer": "1610.04662", - "standard_citation": "arxiv:1610.04662", - "bibtex": "@article{sLPsrfbl,\n abstract = {Melanoma is the deadliest form of skin cancer. While curable with early\ndetection, only highly trained specialists are capable of accurately\nrecognizing the disease. As expertise is in limited supply, automated systems\ncapable of identifying disease could save lives, reduce unnecessary biopsies, and reduce costs. Toward this goal, we propose a system that combines recent\ndevelopments in deep learning with established machine learning approaches, creating ensembles of methods that are capable of segmenting skin lesions, as\nwell as analyzing the detected area and surrounding tissue for melanoma\ndetection. The system is evaluated using the largest publicly available\nbenchmark dataset of dermoscopic images, containing 900 training and 379\ntesting images. New state-of-the-art performance levels are demonstrated, leading to an improvement in the area under receiver operating characteristic\ncurve of 7.5% (0.843 vs. 0.783), in average precision of 4% (0.649 vs. 0.624), and in specificity measured at the clinically relevant 95% sensitivity\noperating point 2.9 times higher than the previous state-of-the-art (36.8%\nspecificity compared to 12.5%). Compared to the average of 8 expert\ndermatologists on a subset of 100 test images, the proposed system produces a\nhigher accuracy (76% vs. 70.5%), and specificity (62% vs. 59%) evaluated at an\nequivalent sensitivity (82%).},\n archiveprefix = {arXiv},\n author = {Noel Codella and Quoc-Bao Nguyen and Sharath Pankanti and David Gutman and Brian Helba and Allan Halpern and John R. Smith},\n eprint = {1610.04662v2},\n file = {1610.04662v2.pdf},\n month = {Nov},\n note = {IBM Journal of Research and Development, vol. 61, no. 4/5, 2017},\n primaryclass = {cs.CV},\n title = {Deep Learning Ensembles for Melanoma Recognition in Dermoscopy Images},\n url = {https://arxiv.org/abs/1610.04662v2},\n year = {2016}\n}\n\n", - "citation_id": "sLPsrfbl" - }, - "doi:10.1038/nature11247": { - "source": "doi", - "identifer": "10.1038/nature11247", - "standard_citation": "doi:10.1038/nature11247", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T11:22:17Z", - "timestamp": 1507116137274 - }, - "reference-count": 80, - "publisher": "Springer Nature", - "issue": "7414", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/nature11247", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2012, - 9, - 4 - ] - ], - "date-time": "2012-09-04T14:12:59Z", - "timestamp": 1346767979000 - }, - "page": "57-74", - "source": "Crossref", - "is-referenced-by-count": 4414, - "title": "An integrated encyclopedia of DNA elements in the human genome", - "prefix": "10.1038", - "volume": "489", - "author": [ - { - "given": "Ian", - "family": "Dunham", - "affiliation": [] - }, - { - "given": "Anshul", - "family": "Kundaje", - "affiliation": [] - }, - { - "given": "Shelley F.", - "family": "Aldred", - "affiliation": [] - }, - { - "given": "Patrick J.", - "family": "Collins", - "affiliation": [] - }, - { - "given": "Carrie A.", - "family": "Davis", - "affiliation": [] - }, - { - "given": "Francis", - "family": "Doyle", - "affiliation": [] - }, - { - "given": "Charles B.", - "family": "Epstein", - "affiliation": [] - }, - { - "given": "Seth", - "family": "Frietze", - "affiliation": [] - }, - { - "given": "Jennifer", - "family": "Harrow", - "affiliation": [] - }, - { - "given": "Rajinder", - "family": "Kaul", - "affiliation": [] - }, - { - "given": "Jainab", - "family": "Khatun", - "affiliation": [] - }, - { - "given": "Bryan R.", - "family": "Lajoie", - "affiliation": [] - }, - { - "given": "Stephen G.", - "family": "Landt", - "affiliation": [] - }, - { - "given": "Bum-Kyu", - "family": "Lee", - "affiliation": [] - }, - { - "given": "Florencia", - "family": "Pauli", - "affiliation": [] - }, - { - "given": "Kate R.", - "family": "Rosenbloom", - "affiliation": [] - }, - { - "given": "Peter", - "family": "Sabo", - "affiliation": [] - }, - { - "given": "Alexias", - "family": "Safi", - "affiliation": [] - }, - { - "given": "Amartya", - "family": "Sanyal", - "affiliation": [] - }, - { - "given": "Noam", - "family": "Shoresh", - "affiliation": [] - }, - { - "given": "Jeremy M.", - "family": "Simon", - "affiliation": [] - }, - { - "given": "Lingyun", - "family": "Song", - "affiliation": [] - }, - { - "given": "Nathan D.", - "family": "Trinklein", - "affiliation": [] - }, - { - "given": "Robert C.", - "family": "Altshuler", - "affiliation": [] - }, - { - "given": "Ewan", - "family": "Birney", - "affiliation": [] - }, - { - "given": "James B.", - "family": "Brown", - "affiliation": [] - }, - { - "given": "Chao", - "family": "Cheng", - "affiliation": [] - }, - { - "given": "Sarah", - "family": "Djebali", - "affiliation": [] - }, - { - "given": "Xianjun", - "family": "Dong", - "affiliation": [] - }, - { - "given": "Ian", - "family": "Dunham", - "affiliation": [] - }, - { - "given": "Jason", - "family": "Ernst", - "affiliation": [] - }, - { - "given": "Terrence S.", - "family": "Furey", - "affiliation": [] - }, - { - "given": "Mark", - "family": "Gerstein", - "affiliation": [] - }, - { - "given": "Belinda", - "family": "Giardine", - "affiliation": [] - }, - { - "given": "Melissa", - "family": "Greven", - "affiliation": [] - }, - { - "given": "Ross C.", - "family": "Hardison", - "affiliation": [] - }, - { - "given": "Robert S.", - "family": "Harris", - "affiliation": [] - }, - { - "given": "Javier", - "family": "Herrero", - "affiliation": [] - }, - { - "given": "Michael M.", - "family": "Hoffman", - "affiliation": [] - }, - { - "given": "Sowmya", - "family": "Iyer", - "affiliation": [] - }, - { - "given": "Manolis", - "family": "Kellis", - "affiliation": [] - }, - { - "given": "Jainab", - "family": "Khatun", - "affiliation": [] - }, - { - "given": "Pouya", - "family": "Kheradpour", - "affiliation": [] - }, - { - "given": "Anshul", - "family": "Kundaje", - "affiliation": [] - }, - { - "given": "Timo", - "family": "Lassmann", - "affiliation": [] - }, - { - "given": "Qunhua", - "family": "Li", - "affiliation": [] - }, - { - "given": "Xinying", - "family": "Lin", - "affiliation": [] - }, - { - "given": "Georgi K.", - "family": "Marinov", - "affiliation": [] - }, - { - "given": "Angelika", - "family": "Merkel", - "affiliation": [] - }, - { - "given": "Ali", - "family": "Mortazavi", - "affiliation": [] - }, - { - "given": "Stephen C. J.", - "family": "Parker", - "affiliation": [] - }, - { - "given": "Timothy E.", - "family": "Reddy", - "affiliation": [] - }, - { - "given": "Joel", - "family": "Rozowsky", - "affiliation": [] - }, - { - "given": "Felix", - "family": "Schlesinger", - "affiliation": [] - }, - { - "given": "Robert E.", - "family": "Thurman", - "affiliation": [] - }, - { - "given": "Jie", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Lucas D.", - "family": "Ward", - "affiliation": [] - }, - { - "given": "Troy W.", - "family": "Whitfield", - "affiliation": [] - }, - { - "given": "Steven P.", - "family": "Wilder", - "affiliation": [] - }, - { - "given": "Weisheng", - "family": "Wu", - "affiliation": [] - }, - { - "given": "Hualin S.", - "family": "Xi", - "affiliation": [] - }, - { - "given": "Kevin Y.", - "family": "Yip", - "affiliation": [] - }, - { - "given": "Jiali", - "family": "Zhuang", - "affiliation": [] - }, - { - "given": "Bradley E.", - "family": "Bernstein", - "affiliation": [] - }, - { - "given": "Ewan", - "family": "Birney", - "affiliation": [] - }, - { - "given": "Ian", - "family": "Dunham", - "affiliation": [] - }, - { - "given": "Eric D.", - "family": "Green", - "affiliation": [] - }, - { - "given": "Chris", - "family": "Gunter", - "affiliation": [] - }, - { - "given": "Michael", - "family": "Snyder", - "affiliation": [] - }, - { - "given": "Michael J.", - "family": "Pazin", - "affiliation": [] - }, - { - "given": "Rebecca F.", - "family": "Lowdon", - "affiliation": [] - }, - { - "given": "Laura A. L.", - "family": "Dillon", - "affiliation": [] - }, - { - "given": "Leslie B.", - "family": "Adams", - "affiliation": [] - }, - { - "given": "Caroline J.", - "family": "Kelly", - "affiliation": [] - }, - { - "given": "Julia", - "family": "Zhang", - "affiliation": [] - }, - { - "given": "Judith R.", - "family": "Wexler", - "affiliation": [] - }, - { - "given": "Eric D.", - "family": "Green", - "affiliation": [] - }, - { - "given": "Peter J.", - "family": "Good", - "affiliation": [] - }, - { - "given": "Elise A.", - "family": "Feingold", - "affiliation": [] - }, - { - "given": "Bradley E.", - "family": "Bernstein", - "affiliation": [] - }, - { - "given": "Ewan", - "family": "Birney", - "affiliation": [] - }, - { - "given": "Gregory E.", - "family": "Crawford", - "affiliation": [] - }, - { - "given": "Job", - "family": "Dekker", - "affiliation": [] - }, - { - "given": "Laura", - "family": "Elnitski", - "affiliation": [] - }, - { - "given": "Peggy J.", - "family": "Farnham", - "affiliation": [] - }, - { - "given": "Mark", - "family": "Gerstein", - "affiliation": [] - }, - { - "given": "Morgan C.", - "family": "Giddings", - "affiliation": [] - }, - { - "given": "Thomas R.", - "family": "Gingeras", - "affiliation": [] - }, - { - "given": "Eric D.", - "family": "Green", - "affiliation": [] - }, - { - "given": "Roderic", - "family": "Guigó", - "affiliation": [] - }, - { - "given": "Ross C.", - "family": "Hardison", - "affiliation": [] - }, - { - "given": "Timothy J.", - "family": "Hubbard", - "affiliation": [] - }, - { - "given": "Manolis", - "family": "Kellis", - "affiliation": [] - }, - { - "given": "W. James", - "family": "Kent", - "affiliation": [] - }, - { - "given": "Jason D.", - "family": "Lieb", - "affiliation": [] - }, - { - "given": "Elliott H.", - "family": "Margulies", - "affiliation": [] - }, - { - "given": "Richard M.", - "family": "Myers", - "affiliation": [] - }, - { - "given": "Michael", - "family": "Snyder", - "affiliation": [] - }, - { - "given": "John A.", - "family": "Stamatoyannopoulos", - "affiliation": [] - }, - { - "given": "Scott A.", - "family": "Tenenbaum", - "affiliation": [] - }, - { - "given": "Zhiping", - "family": "Weng", - "affiliation": [] - }, - { - "given": "Kevin P.", - "family": "White", - "affiliation": [] - }, - { - "given": "Barbara", - "family": "Wold", - "affiliation": [] - }, - { - "given": "Jainab", - "family": "Khatun", - "affiliation": [] - }, - { - "given": "Yanbao", - "family": "Yu", - "affiliation": [] - }, - { - "given": "John", - "family": "Wrobel", - "affiliation": [] - }, - { - "given": "Brian A.", - "family": "Risk", - "affiliation": [] - }, - { - "given": "Harsha P.", - "family": "Gunawardena", - "affiliation": [] - }, - { - "given": "Heather C.", - "family": "Kuiper", - "affiliation": [] - }, - { - "given": "Christopher W.", - "family": "Maier", - "affiliation": [] - }, - { - "given": "Ling", - "family": "Xie", - "affiliation": [] - }, - { - "given": "Xian", - "family": "Chen", - "affiliation": [] - }, - { - "given": "Morgan C.", - "family": "Giddings", - "affiliation": [] - }, - { - "given": "Bradley E.", - "family": "Bernstein", - "affiliation": [] - }, - { - "given": "Charles B.", - "family": "Epstein", - "affiliation": [] - }, - { - "given": "Noam", - "family": "Shoresh", - "affiliation": [] - }, - { - "given": "Jason", - "family": "Ernst", - "affiliation": [] - }, - { - "given": "Pouya", - "family": "Kheradpour", - "affiliation": [] - }, - { - "given": "Tarjei S.", - "family": "Mikkelsen", - "affiliation": [] - }, - { - "given": "Shawn", - "family": "Gillespie", - "affiliation": [] - }, - { - "given": "Alon", - "family": "Goren", - "affiliation": [] - }, - { - "given": "Oren", - "family": "Ram", - "affiliation": [] - }, - { - "given": "Xiaolan", - "family": "Zhang", - "affiliation": [] - }, - { - "given": "Li", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Robbyn", - "family": "Issner", - "affiliation": [] - }, - { - "given": "Michael J.", - "family": "Coyne", - "affiliation": [] - }, - { - "given": "Timothy", - "family": "Durham", - "affiliation": [] - }, - { - "given": "Manching", - "family": "Ku", - "affiliation": [] - }, - { - "given": "Thanh", - "family": "Truong", - "affiliation": [] - }, - { - "given": "Lucas D.", - "family": "Ward", - "affiliation": [] - }, - { - "given": "Robert C.", - "family": "Altshuler", - "affiliation": [] - }, - { - "given": "Matthew L.", - "family": "Eaton", - "affiliation": [] - }, - { - "given": "Manolis", - "family": "Kellis", - "affiliation": [] - }, - { - "given": "Sarah", - "family": "Djebali", - "affiliation": [] - }, - { - "given": "Carrie A.", - "family": "Davis", - "affiliation": [] - }, - { - "given": "Angelika", - "family": "Merkel", - "affiliation": [] - }, - { - "given": "Alex", - "family": "Dobin", - "affiliation": [] - }, - { - "given": "Timo", - "family": "Lassmann", - "affiliation": [] - }, - { - "given": "Ali", - "family": "Mortazavi", - "affiliation": [] - }, - { - "given": "Andrea", - "family": "Tanzer", - "affiliation": [] - }, - { - "given": "Julien", - "family": "Lagarde", - "affiliation": [] - }, - { - "given": "Wei", - "family": "Lin", - "affiliation": [] - }, - { - "given": "Felix", - "family": "Schlesinger", - "affiliation": [] - }, - { - "given": "Chenghai", - "family": "Xue", - "affiliation": [] - }, - { - "given": "Georgi K.", - "family": "Marinov", - "affiliation": [] - }, - { - "given": "Jainab", - "family": "Khatun", - "affiliation": [] - }, - { - "given": "Brian A.", - "family": "Williams", - "affiliation": [] - }, - { - "given": "Chris", - "family": "Zaleski", - "affiliation": [] - }, - { - "given": "Joel", - "family": "Rozowsky", - "affiliation": [] - }, - { - "given": "Maik", - "family": "Röder", - "affiliation": [] - }, - { - "given": "Felix", - "family": "Kokocinski", - "affiliation": [] - }, - { - "given": "Rehab F.", - "family": "Abdelhamid", - "affiliation": [] - }, - { - "given": "Tyler", - "family": "Alioto", - "affiliation": [] - }, - { - "given": "Igor", - "family": "Antoshechkin", - "affiliation": [] - }, - { - "given": "Michael T.", - "family": "Baer", - "affiliation": [] - }, - { - "given": "Philippe", - "family": "Batut", - "affiliation": [] - }, - { - "given": "Ian", - "family": "Bell", - "affiliation": [] - }, - { - "given": "Kimberly", - "family": "Bell", - "affiliation": [] - }, - { - "given": "Sudipto", - "family": "Chakrabortty", - "affiliation": [] - }, - { - "given": "Xian", - "family": "Chen", - "affiliation": [] - }, - { - "given": "Jacqueline", - "family": "Chrast", - "affiliation": [] - }, - { - "given": "Joao", - "family": "Curado", - "affiliation": [] - }, - { - "given": "Thomas", - "family": "Derrien", - "affiliation": [] - }, - { - "given": "Jorg", - "family": "Drenkow", - "affiliation": [] - }, - { - "given": "Erica", - "family": "Dumais", - "affiliation": [] - }, - { - "given": "Jackie", - "family": "Dumais", - "affiliation": [] - }, - { - "given": "Radha", - "family": "Duttagupta", - "affiliation": [] - }, - { - "given": "Megan", - "family": "Fastuca", - "affiliation": [] - }, - { - "given": "Kata", - "family": "Fejes-Toth", - "affiliation": [] - }, - { - "given": "Pedro", - "family": "Ferreira", - "affiliation": [] - }, - { - "given": "Sylvain", - "family": "Foissac", - "affiliation": [] - }, - { - "given": "Melissa J.", - "family": "Fullwood", - "affiliation": [] - }, - { - "given": "Hui", - "family": "Gao", - "affiliation": [] - }, - { - "given": "David", - "family": "Gonzalez", - "affiliation": [] - }, - { - "given": "Assaf", - "family": "Gordon", - "affiliation": [] - }, - { - "given": "Harsha P.", - "family": "Gunawardena", - "affiliation": [] - }, - { - "given": "Cédric", - "family": "Howald", - "affiliation": [] - }, - { - "given": "Sonali", - "family": "Jha", - "affiliation": [] - }, - { - "given": "Rory", - "family": "Johnson", - "affiliation": [] - }, - { - "given": "Philipp", - "family": "Kapranov", - "affiliation": [] - }, - { - "given": "Brandon", - "family": "King", - "affiliation": [] - }, - { - "given": "Colin", - "family": "Kingswood", - "affiliation": [] - }, - { - "given": "Guoliang", - "family": "Li", - "affiliation": [] - }, - { - "given": "Oscar J.", - "family": "Luo", - "affiliation": [] - }, - { - "given": "Eddie", - "family": "Park", - "affiliation": [] - }, - { - "given": "Jonathan B.", - "family": "Preall", - "affiliation": [] - }, - { - "given": "Kimberly", - "family": "Presaud", - "affiliation": [] - }, - { - "given": "Paolo", - "family": "Ribeca", - "affiliation": [] - }, - { - "given": "Brian A.", - "family": "Risk", - "affiliation": [] - }, - { - "given": "Daniel", - "family": "Robyr", - "affiliation": [] - }, - { - "given": "Xiaoan", - "family": "Ruan", - "affiliation": [] - }, - { - "given": "Michael", - "family": "Sammeth", - "affiliation": [] - }, - { - "given": "Kuljeet Singh", - "family": "Sandhu", - "affiliation": [] - }, - { - "given": "Lorain", - "family": "Schaeffer", - "affiliation": [] - }, - { - "given": "Lei-Hoon", - "family": "See", - "affiliation": [] - }, - { - "given": "Atif", - "family": "Shahab", - "affiliation": [] - }, - { - "given": "Jorgen", - "family": "Skancke", - "affiliation": [] - }, - { - "given": "Ana Maria", - "family": "Suzuki", - "affiliation": [] - }, - { - "given": "Hazuki", - "family": "Takahashi", - "affiliation": [] - }, - { - "given": "Hagen", - "family": "Tilgner", - "affiliation": [] - }, - { - "given": "Diane", - "family": "Trout", - "affiliation": [] - }, - { - "given": "Nathalie", - "family": "Walters", - "affiliation": [] - }, - { - "given": "Huaien", - "family": "Wang", - "affiliation": [] - }, - { - "given": "John", - "family": "Wrobel", - "affiliation": [] - }, - { - "given": "Yanbao", - "family": "Yu", - "affiliation": [] - }, - { - "given": "Yoshihide", - "family": "Hayashizaki", - "affiliation": [] - }, - { - "given": "Jennifer", - "family": "Harrow", - "affiliation": [] - }, - { - "given": "Mark", - "family": "Gerstein", - "affiliation": [] - }, - { - "given": "Timothy J.", - "family": "Hubbard", - "affiliation": [] - }, - { - "given": "Alexandre", - "family": "Reymond", - "affiliation": [] - }, - { - "given": "Stylianos E.", - "family": "Antonarakis", - "affiliation": [] - }, - { - "given": "Gregory J.", - "family": "Hannon", - "affiliation": [] - }, - { - "given": "Morgan C.", - "family": "Giddings", - "affiliation": [] - }, - { - "given": "Yijun", - "family": "Ruan", - "affiliation": [] - }, - { - "given": "Barbara", - "family": "Wold", - "affiliation": [] - }, - { - "given": "Piero", - "family": "Carninci", - "affiliation": [] - }, - { - "given": "Roderic", - "family": "Guigó", - "affiliation": [] - }, - { - "given": "Thomas R.", - "family": "Gingeras", - "affiliation": [] - }, - { - "given": "Kate R.", - "family": "Rosenbloom", - "affiliation": [] - }, - { - "given": "Cricket A.", - "family": "Sloan", - "affiliation": [] - }, - { - "given": "Katrina", - "family": "Learned", - "affiliation": [] - }, - { - "given": "Venkat S.", - "family": "Malladi", - "affiliation": [] - }, - { - "given": "Matthew C.", - "family": "Wong", - "affiliation": [] - }, - { - "given": "Galt P.", - "family": "Barber", - "affiliation": [] - }, - { - "given": "Melissa S.", - "family": "Cline", - "affiliation": [] - }, - { - "given": "Timothy R.", - "family": "Dreszer", - "affiliation": [] - }, - { - "given": "Steven G.", - "family": "Heitner", - "affiliation": [] - }, - { - "given": "Donna", - "family": "Karolchik", - "affiliation": [] - }, - { - "given": "W. James", - "family": "Kent", - "affiliation": [] - }, - { - "given": "Vanessa M.", - "family": "Kirkup", - "affiliation": [] - }, - { - "given": "Laurence R.", - "family": "Meyer", - "affiliation": [] - }, - { - "given": "Jeffrey C.", - "family": "Long", - "affiliation": [] - }, - { - "given": "Morgan", - "family": "Maddren", - "affiliation": [] - }, - { - "given": "Brian J.", - "family": "Raney", - "affiliation": [] - }, - { - "given": "Terrence S.", - "family": "Furey", - "affiliation": [] - }, - { - "given": "Lingyun", - "family": "Song", - "affiliation": [] - }, - { - "given": "Linda L.", - "family": "Grasfeder", - "affiliation": [] - }, - { - "given": "Paul G.", - "family": "Giresi", - "affiliation": [] - }, - { - "given": "Bum-Kyu", - "family": "Lee", - "affiliation": [] - }, - { - "given": "Anna", - "family": "Battenhouse", - "affiliation": [] - }, - { - "given": "Nathan C.", - "family": "Sheffield", - "affiliation": [] - }, - { - "given": "Jeremy M.", - "family": "Simon", - "affiliation": [] - }, - { - "given": "Kimberly A.", - "family": "Showers", - "affiliation": [] - }, - { - "given": "Alexias", - "family": "Safi", - "affiliation": [] - }, - { - "given": "Darin", - "family": "London", - "affiliation": [] - }, - { - "given": "Akshay A.", - "family": "Bhinge", - "affiliation": [] - }, - { - "given": "Christopher", - "family": "Shestak", - "affiliation": [] - }, - { - "given": "Matthew R.", - "family": "Schaner", - "affiliation": [] - }, - { - "given": "Seul", - "family": "Ki Kim", - "affiliation": [] - }, - { - "given": "Zhuzhu Z.", - "family": "Zhang", - "affiliation": [] - }, - { - "given": "Piotr A.", - "family": "Mieczkowski", - "affiliation": [] - }, - { - "given": "Joanna O.", - "family": "Mieczkowska", - "affiliation": [] - }, - { - "given": "Zheng", - "family": "Liu", - "affiliation": [] - }, - { - "given": "Ryan M.", - "family": "McDaniell", - "affiliation": [] - }, - { - "given": "Yunyun", - "family": "Ni", - "affiliation": [] - }, - { - "given": "Naim U.", - "family": "Rashid", - "affiliation": [] - }, - { - "given": "Min Jae", - "family": "Kim", - "affiliation": [] - }, - { - "given": "Sheera", - "family": "Adar", - "affiliation": [] - }, - { - "given": "Zhancheng", - "family": "Zhang", - "affiliation": [] - }, - { - "given": "Tianyuan", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Deborah", - "family": "Winter", - "affiliation": [] - }, - { - "given": "Damian", - "family": "Keefe", - "affiliation": [] - }, - { - "given": "Ewan", - "family": "Birney", - "affiliation": [] - }, - { - "given": "Vishwanath R.", - "family": "Iyer", - "affiliation": [] - }, - { - "given": "Jason D.", - "family": "Lieb", - "affiliation": [] - }, - { - "given": "Gregory E.", - "family": "Crawford", - "affiliation": [] - }, - { - "given": "Guoliang", - "family": "Li", - "affiliation": [] - }, - { - "given": "Kuljeet Singh", - "family": "Sandhu", - "affiliation": [] - }, - { - "given": "Meizhen", - "family": "Zheng", - "affiliation": [] - }, - { - "given": "Ping", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Oscar J.", - "family": "Luo", - "affiliation": [] - }, - { - "given": "Atif", - "family": "Shahab", - "affiliation": [] - }, - { - "given": "Melissa J.", - "family": "Fullwood", - "affiliation": [] - }, - { - "given": "Xiaoan", - "family": "Ruan", - "affiliation": [] - }, - { - "given": "Yijun", - "family": "Ruan", - "affiliation": [] - }, - { - "given": "Richard M.", - "family": "Myers", - "affiliation": [] - }, - { - "given": "Florencia", - "family": "Pauli", - "affiliation": [] - }, - { - "given": "Brian A.", - "family": "Williams", - "affiliation": [] - }, - { - "given": "Jason", - "family": "Gertz", - "affiliation": [] - }, - { - "given": "Georgi K.", - "family": "Marinov", - "affiliation": [] - }, - { - "given": "Timothy E.", - "family": "Reddy", - "affiliation": [] - }, - { - "given": "Jost", - "family": "Vielmetter", - "affiliation": [] - }, - { - "given": "E.", - "family": "Partridge", - "affiliation": [] - }, - { - "given": "Diane", - "family": "Trout", - "affiliation": [] - }, - { - "given": "Katherine E.", - "family": "Varley", - "affiliation": [] - }, - { - "given": "Clarke", - "family": "Gasper", - "affiliation": [] - }, - { - "given": "Anita", - "family": "Bansal", - "affiliation": [] - }, - { - "given": "Shirley", - "family": "Pepke", - "affiliation": [] - }, - { - "given": "Preti", - "family": "Jain", - "affiliation": [] - }, - { - "given": "Henry", - "family": "Amrhein", - "affiliation": [] - }, - { - "given": "Kevin M.", - "family": "Bowling", - "affiliation": [] - }, - { - "given": "Michael", - "family": "Anaya", - "affiliation": [] - }, - { - "given": "Marie K.", - "family": "Cross", - "affiliation": [] - }, - { - "given": "Brandon", - "family": "King", - "affiliation": [] - }, - { - "given": "Michael A.", - "family": "Muratet", - "affiliation": [] - }, - { - "given": "Igor", - "family": "Antoshechkin", - "affiliation": [] - }, - { - "given": "Kimberly M.", - "family": "Newberry", - "affiliation": [] - }, - { - "given": "Kenneth", - "family": "McCue", - "affiliation": [] - }, - { - "given": "Amy S.", - "family": "Nesmith", - "affiliation": [] - }, - { - "given": "Katherine I.", - "family": "Fisher-Aylor", - "affiliation": [] - }, - { - "given": "Barbara", - "family": "Pusey", - "affiliation": [] - }, - { - "given": "Gilberto", - "family": "DeSalvo", - "affiliation": [] - }, - { - "given": "Stephanie L.", - "family": "Parker", - "affiliation": [] - }, - { - "given": "Sreeram", - "family": "Balasubramanian", - "affiliation": [] - }, - { - "given": "Nicholas S.", - "family": "Davis", - "affiliation": [] - }, - { - "given": "Sarah K.", - "family": "Meadows", - "affiliation": [] - }, - { - "given": "Tracy", - "family": "Eggleston", - "affiliation": [] - }, - { - "given": "Chris", - "family": "Gunter", - "affiliation": [] - }, - { - "given": "J. Scott", - "family": "Newberry", - "affiliation": [] - }, - { - "given": "Shawn E.", - "family": "Levy", - "affiliation": [] - }, - { - "given": "Devin M.", - "family": "Absher", - "affiliation": [] - }, - { - "given": "Ali", - "family": "Mortazavi", - "affiliation": [] - }, - { - "given": "Wing H.", - "family": "Wong", - "affiliation": [] - }, - { - "given": "Barbara", - "family": "Wold", - "affiliation": [] - }, - { - "given": "Matthew J.", - "family": "Blow", - "affiliation": [] - }, - { - "given": "Axel", - "family": "Visel", - "affiliation": [] - }, - { - "given": "Len A.", - "family": "Pennachio", - "affiliation": [] - }, - { - "given": "Laura", - "family": "Elnitski", - "affiliation": [] - }, - { - "given": "Elliott H.", - "family": "Margulies", - "affiliation": [] - }, - { - "given": "Stephen C. J.", - "family": "Parker", - "affiliation": [] - }, - { - "given": "Hanna M.", - "family": "Petrykowska", - "affiliation": [] - }, - { - "given": "Alexej", - "family": "Abyzov", - "affiliation": [] - }, - { - "given": "Bronwen", - "family": "Aken", - "affiliation": [] - }, - { - "given": "Daniel", - "family": "Barrell", - "affiliation": [] - }, - { - "given": "Gemma", - "family": "Barson", - "affiliation": [] - }, - { - "given": "Andrew", - "family": "Berry", - "affiliation": [] - }, - { - "given": "Alexandra", - "family": "Bignell", - "affiliation": [] - }, - { - "given": "Veronika", - "family": "Boychenko", - "affiliation": [] - }, - { - "given": "Giovanni", - "family": "Bussotti", - "affiliation": [] - }, - { - "given": "Jacqueline", - "family": "Chrast", - "affiliation": [] - }, - { - "given": "Claire", - "family": "Davidson", - "affiliation": [] - }, - { - "given": "Thomas", - "family": "Derrien", - "affiliation": [] - }, - { - "given": "Gloria", - "family": "Despacio-Reyes", - "affiliation": [] - }, - { - "given": "Mark", - "family": "Diekhans", - "affiliation": [] - }, - { - "given": "Iakes", - "family": "Ezkurdia", - "affiliation": [] - }, - { - "given": "Adam", - "family": "Frankish", - "affiliation": [] - }, - { - "given": "James", - "family": "Gilbert", - "affiliation": [] - }, - { - "given": "Jose Manuel", - "family": "Gonzalez", - "affiliation": [] - }, - { - "given": "Ed", - "family": "Griffiths", - "affiliation": [] - }, - { - "given": "Rachel", - "family": "Harte", - "affiliation": [] - }, - { - "given": "David A.", - "family": "Hendrix", - "affiliation": [] - }, - { - "given": "Cédric", - "family": "Howald", - "affiliation": [] - }, - { - "given": "Toby", - "family": "Hunt", - "affiliation": [] - }, - { - "given": "Irwin", - "family": "Jungreis", - "affiliation": [] - }, - { - "given": "Mike", - "family": "Kay", - "affiliation": [] - }, - { - "given": "Ekta", - "family": "Khurana", - "affiliation": [] - }, - { - "given": "Felix", - "family": "Kokocinski", - "affiliation": [] - }, - { - "given": "Jing", - "family": "Leng", - "affiliation": [] - }, - { - "given": "Michael F.", - "family": "Lin", - "affiliation": [] - }, - { - "given": "Jane", - "family": "Loveland", - "affiliation": [] - }, - { - "given": "Zhi", - "family": "Lu", - "affiliation": [] - }, - { - "given": "Deepa", - "family": "Manthravadi", - "affiliation": [] - }, - { - "given": "Marco", - "family": "Mariotti", - "affiliation": [] - }, - { - "given": "Jonathan", - "family": "Mudge", - "affiliation": [] - }, - { - "given": "Gaurab", - "family": "Mukherjee", - "affiliation": [] - }, - { - "given": "Cedric", - "family": "Notredame", - "affiliation": [] - }, - { - "given": "Baikang", - "family": "Pei", - "affiliation": [] - }, - { - "given": "Jose Manuel", - "family": "Rodriguez", - "affiliation": [] - }, - { - "given": "Gary", - "family": "Saunders", - "affiliation": [] - }, - { - "given": "Andrea", - "family": "Sboner", - "affiliation": [] - }, - { - "given": "Stephen", - "family": "Searle", - "affiliation": [] - }, - { - "given": "Cristina", - "family": "Sisu", - "affiliation": [] - }, - { - "given": "Catherine", - "family": "Snow", - "affiliation": [] - }, - { - "given": "Charlie", - "family": "Steward", - "affiliation": [] - }, - { - "given": "Andrea", - "family": "Tanzer", - "affiliation": [] - }, - { - "given": "Electra", - "family": "Tapanari", - "affiliation": [] - }, - { - "given": "Michael L.", - "family": "Tress", - "affiliation": [] - }, - { - "given": "Marijke J.", - "family": "van Baren", - "affiliation": [] - }, - { - "given": "Nathalie", - "family": "Walters", - "affiliation": [] - }, - { - "given": "Stefan", - "family": "Washietl", - "affiliation": [] - }, - { - "given": "Laurens", - "family": "Wilming", - "affiliation": [] - }, - { - "given": "Amonida", - "family": "Zadissa", - "affiliation": [] - }, - { - "given": "Zhengdong", - "family": "Zhang", - "affiliation": [] - }, - { - "given": "Michael", - "family": "Brent", - "affiliation": [] - }, - { - "given": "David", - "family": "Haussler", - "affiliation": [] - }, - { - "given": "Manolis", - "family": "Kellis", - "affiliation": [] - }, - { - "given": "Alfonso", - "family": "Valencia", - "affiliation": [] - }, - { - "given": "Mark", - "family": "Gerstein", - "affiliation": [] - }, - { - "given": "Alexandre", - "family": "Reymond", - "affiliation": [] - }, - { - "given": "Roderic", - "family": "Guigó", - "affiliation": [] - }, - { - "given": "Jennifer", - "family": "Harrow", - "affiliation": [] - }, - { - "given": "Timothy J.", - "family": "Hubbard", - "affiliation": [] - }, - { - "given": "Stephen G.", - "family": "Landt", - "affiliation": [] - }, - { - "given": "Seth", - "family": "Frietze", - "affiliation": [] - }, - { - "given": "Alexej", - "family": "Abyzov", - "affiliation": [] - }, - { - "given": "Nick", - "family": "Addleman", - "affiliation": [] - }, - { - "given": "Roger P.", - "family": "Alexander", - "affiliation": [] - }, - { - "given": "Raymond K.", - "family": "Auerbach", - "affiliation": [] - }, - { - "given": "Suganthi", - "family": "Balasubramanian", - "affiliation": [] - }, - { - "given": "Keith", - "family": "Bettinger", - "affiliation": [] - }, - { - "given": "Nitin", - "family": "Bhardwaj", - "affiliation": [] - }, - { - "given": "Alan P.", - "family": "Boyle", - "affiliation": [] - }, - { - "given": "Alina R.", - "family": "Cao", - "affiliation": [] - }, - { - "given": "Philip", - "family": "Cayting", - "affiliation": [] - }, - { - "given": "Alexandra", - "family": "Charos", - "affiliation": [] - }, - { - "given": "Yong", - "family": "Cheng", - "affiliation": [] - }, - { - "given": "Chao", - "family": "Cheng", - "affiliation": [] - }, - { - "given": "Catharine", - "family": "Eastman", - "affiliation": [] - }, - { - "given": "Ghia", - "family": "Euskirchen", - "affiliation": [] - }, - { - "given": "Joseph D.", - "family": "Fleming", - "affiliation": [] - }, - { - "given": "Fabian", - "family": "Grubert", - "affiliation": [] - }, - { - "given": "Lukas", - "family": "Habegger", - "affiliation": [] - }, - { - "given": "Manoj", - "family": "Hariharan", - "affiliation": [] - }, - { - "given": "Arif", - "family": "Harmanci", - "affiliation": [] - }, - { - "given": "Sushma", - "family": "Iyengar", - "affiliation": [] - }, - { - "given": "Victor X.", - "family": "Jin", - "affiliation": [] - }, - { - "given": "Konrad J.", - "family": "Karczewski", - "affiliation": [] - }, - { - "given": "Maya", - "family": "Kasowski", - "affiliation": [] - }, - { - "given": "Phil", - "family": "Lacroute", - "affiliation": [] - }, - { - "given": "Hugo", - "family": "Lam", - "affiliation": [] - }, - { - "given": "Nathan", - "family": "Lamarre-Vincent", - "affiliation": [] - }, - { - "given": "Jing", - "family": "Leng", - "affiliation": [] - }, - { - "given": "Jin", - "family": "Lian", - "affiliation": [] - }, - { - "given": "Marianne", - "family": "Lindahl-Allen", - "affiliation": [] - }, - { - "given": "Renqiang", - "family": "Min", - "affiliation": [] - }, - { - "given": "Benoit", - "family": "Miotto", - "affiliation": [] - }, - { - "given": "Hannah", - "family": "Monahan", - "affiliation": [] - }, - { - "given": "Zarmik", - "family": "Moqtaderi", - "affiliation": [] - }, - { - "given": "Xinmeng J.", - "family": "Mu", - "affiliation": [] - }, - { - "given": "Henriette", - "family": "O’Geen", - "affiliation": [] - }, - { - "given": "Zhengqing", - "family": "Ouyang", - "affiliation": [] - }, - { - "given": "Dorrelyn", - "family": "Patacsil", - "affiliation": [] - }, - { - "given": "Baikang", - "family": "Pei", - "affiliation": [] - }, - { - "given": "Debasish", - "family": "Raha", - "affiliation": [] - }, - { - "given": "Lucia", - "family": "Ramirez", - "affiliation": [] - }, - { - "given": "Brian", - "family": "Reed", - "affiliation": [] - }, - { - "given": "Joel", - "family": "Rozowsky", - "affiliation": [] - }, - { - "given": "Andrea", - "family": "Sboner", - "affiliation": [] - }, - { - "given": "Minyi", - "family": "Shi", - "affiliation": [] - }, - { - "given": "Cristina", - "family": "Sisu", - "affiliation": [] - }, - { - "given": "Teri", - "family": "Slifer", - "affiliation": [] - }, - { - "given": "Heather", - "family": "Witt", - "affiliation": [] - }, - { - "given": "Linfeng", - "family": "Wu", - "affiliation": [] - }, - { - "given": "Xiaoqin", - "family": "Xu", - "affiliation": [] - }, - { - "given": "Koon-Kiu", - "family": "Yan", - "affiliation": [] - }, - { - "given": "Xinqiong", - "family": "Yang", - "affiliation": [] - }, - { - "given": "Kevin Y.", - "family": "Yip", - "affiliation": [] - }, - { - "given": "Zhengdong", - "family": "Zhang", - "affiliation": [] - }, - { - "given": "Kevin", - "family": "Struhl", - "affiliation": [] - }, - { - "given": "Sherman M.", - "family": "Weissman", - "affiliation": [] - }, - { - "given": "Mark", - "family": "Gerstein", - "affiliation": [] - }, - { - "given": "Peggy J.", - "family": "Farnham", - "affiliation": [] - }, - { - "given": "Michael", - "family": "Snyder", - "affiliation": [] - }, - { - "given": "Scott A.", - "family": "Tenenbaum", - "affiliation": [] - }, - { - "given": "Luiz O.", - "family": "Penalva", - "affiliation": [] - }, - { - "given": "Francis", - "family": "Doyle", - "affiliation": [] - }, - { - "given": "Subhradip", - "family": "Karmakar", - "affiliation": [] - }, - { - "given": "Stephen G.", - "family": "Landt", - "affiliation": [] - }, - { - "given": "Raj R.", - "family": "Bhanvadia", - "affiliation": [] - }, - { - "given": "Alina", - "family": "Choudhury", - "affiliation": [] - }, - { - "given": "Marc", - "family": "Domanus", - "affiliation": [] - }, - { - "given": "Lijia", - "family": "Ma", - "affiliation": [] - }, - { - "given": "Jennifer", - "family": "Moran", - "affiliation": [] - }, - { - "given": "Dorrelyn", - "family": "Patacsil", - "affiliation": [] - }, - { - "given": "Teri", - "family": "Slifer", - "affiliation": [] - }, - { - "given": "Alec", - "family": "Victorsen", - "affiliation": [] - }, - { - "given": "Xinqiong", - "family": "Yang", - "affiliation": [] - }, - { - "given": "Michael", - "family": "Snyder", - "affiliation": [] - }, - { - "given": "Kevin P.", - "family": "White", - "affiliation": [] - }, - { - "given": "Thomas", - "family": "Auer", - "affiliation": [] - }, - { - "given": "Lazaro", - "family": "Centanin", - "affiliation": [] - }, - { - "given": "Michael", - "family": "Eichenlaub", - "affiliation": [] - }, - { - "given": "Franziska", - "family": "Gruhl", - "affiliation": [] - }, - { - "given": "Stephan", - "family": "Heermann", - "affiliation": [] - }, - { - "given": "Burkhard", - "family": "Hoeckendorf", - "affiliation": [] - }, - { - "given": "Daigo", - "family": "Inoue", - "affiliation": [] - }, - { - "given": "Tanja", - "family": "Kellner", - "affiliation": [] - }, - { - "given": "Stephan", - "family": "Kirchmaier", - "affiliation": [] - }, - { - "given": "Claudia", - "family": "Mueller", - "affiliation": [] - }, - { - "given": "Robert", - "family": "Reinhardt", - "affiliation": [] - }, - { - "given": "Lea", - "family": "Schertel", - "affiliation": [] - }, - { - "given": "Stephanie", - "family": "Schneider", - "affiliation": [] - }, - { - "given": "Rebecca", - "family": "Sinn", - "affiliation": [] - }, - { - "given": "Beate", - "family": "Wittbrodt", - "affiliation": [] - }, - { - "given": "Jochen", - "family": "Wittbrodt", - "affiliation": [] - }, - { - "given": "Zhiping", - "family": "Weng", - "affiliation": [] - }, - { - "given": "Troy W.", - "family": "Whitfield", - "affiliation": [] - }, - { - "given": "Jie", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Patrick J.", - "family": "Collins", - "affiliation": [] - }, - { - "given": "Shelley F.", - "family": "Aldred", - "affiliation": [] - }, - { - "given": "Nathan D.", - "family": "Trinklein", - "affiliation": [] - }, - { - "given": "E. Christopher", - "family": "Partridge", - "affiliation": [] - }, - { - "given": "Richard M.", - "family": "Myers", - "affiliation": [] - }, - { - "given": "Job", - "family": "Dekker", - "affiliation": [] - }, - { - "given": "Gaurav", - "family": "Jain", - "affiliation": [] - }, - { - "given": "Bryan R.", - "family": "Lajoie", - "affiliation": [] - }, - { - "given": "Amartya", - "family": "Sanyal", - "affiliation": [] - }, - { - "given": "Gayathri", - "family": "Balasundaram", - "affiliation": [] - }, - { - "given": "Daniel L.", - "family": "Bates", - "affiliation": [] - }, - { - "given": "Rachel", - "family": "Byron", - "affiliation": [] - }, - { - "given": "Theresa K.", - "family": "Canfield", - "affiliation": [] - }, - { - "given": "Morgan J.", - "family": "Diegel", - "affiliation": [] - }, - { - "given": "Douglas", - "family": "Dunn", - "affiliation": [] - }, - { - "given": "Abigail K.", - "family": "Ebersol", - "affiliation": [] - }, - { - "given": "Tristan", - "family": "Frum", - "affiliation": [] - }, - { - "given": "Kavita", - "family": "Garg", - "affiliation": [] - }, - { - "given": "Erica", - "family": "Gist", - "affiliation": [] - }, - { - "given": "R. Scott", - "family": "Hansen", - "affiliation": [] - }, - { - "given": "Lisa", - "family": "Boatman", - "affiliation": [] - }, - { - "given": "Eric", - "family": "Haugen", - "affiliation": [] - }, - { - "given": "Richard", - "family": "Humbert", - "affiliation": [] - }, - { - "given": "Gaurav", - "family": "Jain", - "affiliation": [] - }, - { - "given": "Audra K.", - "family": "Johnson", - "affiliation": [] - }, - { - "given": "Ericka M.", - "family": "Johnson", - "affiliation": [] - }, - { - "given": "Tattyana V.", - "family": "Kutyavin", - "affiliation": [] - }, - { - "given": "Bryan R.", - "family": "Lajoie", - "affiliation": [] - }, - { - "given": "Kristen", - "family": "Lee", - "affiliation": [] - }, - { - "given": "Dimitra", - "family": "Lotakis", - "affiliation": [] - }, - { - "given": "Matthew T.", - "family": "Maurano", - "affiliation": [] - }, - { - "given": "Shane J.", - "family": "Neph", - "affiliation": [] - }, - { - "given": "Fiedencio V.", - "family": "Neri", - "affiliation": [] - }, - { - "given": "Eric D.", - "family": "Nguyen", - "affiliation": [] - }, - { - "given": "Hongzhu", - "family": "Qu", - "affiliation": [] - }, - { - "given": "Alex P.", - "family": "Reynolds", - "affiliation": [] - }, - { - "given": "Vaughn", - "family": "Roach", - "affiliation": [] - }, - { - "given": "Eric", - "family": "Rynes", - "affiliation": [] - }, - { - "given": "Peter", - "family": "Sabo", - "affiliation": [] - }, - { - "given": "Minerva E.", - "family": "Sanchez", - "affiliation": [] - }, - { - "given": "Richard S.", - "family": "Sandstrom", - "affiliation": [] - }, - { - "given": "Amartya", - "family": "Sanyal", - "affiliation": [] - }, - { - "given": "Anthony O.", - "family": "Shafer", - "affiliation": [] - }, - { - "given": "Andrew B.", - "family": "Stergachis", - "affiliation": [] - }, - { - "given": "Sean", - "family": "Thomas", - "affiliation": [] - }, - { - "given": "Robert E.", - "family": "Thurman", - "affiliation": [] - }, - { - "given": "Benjamin", - "family": "Vernot", - "affiliation": [] - }, - { - "given": "Jeff", - "family": "Vierstra", - "affiliation": [] - }, - { - "given": "Shinny", - "family": "Vong", - "affiliation": [] - }, - { - "given": "Hao", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Molly A.", - "family": "Weaver", - "affiliation": [] - }, - { - "given": "Yongqi", - "family": "Yan", - "affiliation": [] - }, - { - "given": "Miaohua", - "family": "Zhang", - "affiliation": [] - }, - { - "given": "Joshua M.", - "family": "Akey", - "affiliation": [] - }, - { - "given": "Michael", - "family": "Bender", - "affiliation": [] - }, - { - "given": "Michael O.", - "family": "Dorschner", - "affiliation": [] - }, - { - "given": "Mark", - "family": "Groudine", - "affiliation": [] - }, - { - "given": "Michael J.", - "family": "MacCoss", - "affiliation": [] - }, - { - "given": "Patrick", - "family": "Navas", - "affiliation": [] - }, - { - "given": "George", - "family": "Stamatoyannopoulos", - "affiliation": [] - }, - { - "given": "Rajinder", - "family": "Kaul", - "affiliation": [] - }, - { - "given": "Job", - "family": "Dekker", - "affiliation": [] - }, - { - "given": "John A.", - "family": "Stamatoyannopoulos", - "affiliation": [] - }, - { - "given": "Ian", - "family": "Dunham", - "affiliation": [] - }, - { - "given": "Kathryn", - "family": "Beal", - "affiliation": [] - }, - { - "given": "Alvis", - "family": "Brazma", - "affiliation": [] - }, - { - "given": "Paul", - "family": "Flicek", - "affiliation": [] - }, - { - "given": "Javier", - "family": "Herrero", - "affiliation": [] - }, - { - "given": "Nathan", - "family": "Johnson", - "affiliation": [] - }, - { - "given": "Damian", - "family": "Keefe", - "affiliation": [] - }, - { - "given": "Margus", - "family": "Lukk", - "affiliation": [] - }, - { - "given": "Nicholas M.", - "family": "Luscombe", - "affiliation": [] - }, - { - "given": "Daniel", - "family": "Sobral", - "affiliation": [] - }, - { - "given": "Juan M.", - "family": "Vaquerizas", - "affiliation": [] - }, - { - "given": "Steven P.", - "family": "Wilder", - "affiliation": [] - }, - { - "given": "Serafim", - "family": "Batzoglou", - "affiliation": [] - }, - { - "given": "Arend", - "family": "Sidow", - "affiliation": [] - }, - { - "given": "Nadine", - "family": "Hussami", - "affiliation": [] - }, - { - "given": "Sofia", - "family": "Kyriazopoulou-Panagiotopoulou", - "affiliation": [] - }, - { - "given": "Max W.", - "family": "Libbrecht", - "affiliation": [] - }, - { - "given": "Marc A.", - "family": "Schaub", - "affiliation": [] - }, - { - "given": "Anshul", - "family": "Kundaje", - "affiliation": [] - }, - { - "given": "Ross C.", - "family": "Hardison", - "affiliation": [] - }, - { - "given": "Webb", - "family": "Miller", - "affiliation": [] - }, - { - "given": "Belinda", - "family": "Giardine", - "affiliation": [] - }, - { - "given": "Robert S.", - "family": "Harris", - "affiliation": [] - }, - { - "given": "Weisheng", - "family": "Wu", - "affiliation": [] - }, - { - "given": "Peter J.", - "family": "Bickel", - "affiliation": [] - }, - { - "given": "Balazs", - "family": "Banfai", - "affiliation": [] - }, - { - "given": "Nathan P.", - "family": "Boley", - "affiliation": [] - }, - { - "given": "James B.", - "family": "Brown", - "affiliation": [] - }, - { - "given": "Haiyan", - "family": "Huang", - "affiliation": [] - }, - { - "given": "Qunhua", - "family": "Li", - "affiliation": [] - }, - { - "given": "Jingyi Jessica", - "family": "Li", - "affiliation": [] - }, - { - "given": "William Stafford", - "family": "Noble", - "affiliation": [] - }, - { - "given": "Jeffrey A.", - "family": "Bilmes", - "affiliation": [] - }, - { - "given": "Orion J.", - "family": "Buske", - "affiliation": [] - }, - { - "given": "Michael M.", - "family": "Hoffman", - "affiliation": [] - }, - { - "given": "Avinash D.", - "family": "Sahu", - "affiliation": [] - }, - { - "given": "Peter V.", - "family": "Kharchenko", - "affiliation": [] - }, - { - "given": "Peter J.", - "family": "Park", - "affiliation": [] - }, - { - "given": "Dannon", - "family": "Baker", - "affiliation": [] - }, - { - "given": "James", - "family": "Taylor", - "affiliation": [] - }, - { - "given": "Zhiping", - "family": "Weng", - "affiliation": [] - }, - { - "given": "Sowmya", - "family": "Iyer", - "affiliation": [] - }, - { - "given": "Xianjun", - "family": "Dong", - "affiliation": [] - }, - { - "given": "Melissa", - "family": "Greven", - "affiliation": [] - }, - { - "given": "Xinying", - "family": "Lin", - "affiliation": [] - }, - { - "given": "Jie", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Hualin S.", - "family": "Xi", - "affiliation": [] - }, - { - "given": "Jiali", - "family": "Zhuang", - "affiliation": [] - }, - { - "given": "Mark", - "family": "Gerstein", - "affiliation": [] - }, - { - "given": "Roger P.", - "family": "Alexander", - "affiliation": [] - }, - { - "given": "Suganthi", - "family": "Balasubramanian", - "affiliation": [] - }, - { - "given": "Chao", - "family": "Cheng", - "affiliation": [] - }, - { - "given": "Arif", - "family": "Harmanci", - "affiliation": [] - }, - { - "given": "Lucas", - "family": "Lochovsky", - "affiliation": [] - }, - { - "given": "Renqiang", - "family": "Min", - "affiliation": [] - }, - { - "given": "Xinmeng J.", - "family": "Mu", - "affiliation": [] - }, - { - "given": "Joel", - "family": "Rozowsky", - "affiliation": [] - }, - { - "given": "Koon-Kiu", - "family": "Yan", - "affiliation": [] - }, - { - "given": "Kevin Y.", - "family": "Yip", - "affiliation": [] - }, - { - "given": "Ewan", - "family": "Birney", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2012, - 9, - 5 - ] - ] - }, - "container-title": "Nature", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 20 - ] - ], - "date-time": "2017-06-20T23:38:33Z", - "timestamp": 1498001913000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2012, - 9, - 5 - ] - ] - }, - "references-count": 80, - "alternative-id": [ - "nature11247" - ], - "URL": "https://doi.org/10.1038/nature11247", - "relation": { - "cites": [] - }, - "subject": [ - "Multidisciplinary" - ], - "container-title-short": "Nature", - "id": "15J98V2qM" - }, - "citation_id": "15J98V2qM" - }, - "arxiv:1410.0759": { - "source": "arxiv", - "identifer": "1410.0759", - "standard_citation": "arxiv:1410.0759", - "bibtex": "@article{YwdqeYZi,\n abstract = {We present a library of efficient implementations of deep learning\nprimitives. Deep learning workloads are computationally intensive, and\noptimizing their kernels is difficult and time-consuming. As parallel\narchitectures evolve, kernels must be reoptimized, which makes maintaining\ncodebases difficult over time. Similar issues have long been addressed in the\nHPC community by libraries such as the Basic Linear Algebra Subroutines (BLAS).\nHowever, there is no analogous library for deep learning. Without such a\nlibrary, researchers implementing deep learning workloads on parallel\nprocessors must create and optimize their own implementations of the main\ncomputational kernels, and this work must be repeated as new parallel\nprocessors emerge. To address this problem, we have created a library similar\nin intent to BLAS, with optimized routines for deep learning workloads. Our\nimplementation contains routines for GPUs, although similarly to the BLAS\nlibrary, these routines could be implemented for other platforms. The library\nis easy to integrate into existing frameworks, and provides optimized\nperformance and memory usage. For example, integrating cuDNN into Caffe, a\npopular framework for convolutional networks, improves performance by 36% on a\nstandard model while also reducing memory consumption.},\n archiveprefix = {arXiv},\n author = {Sharan Chetlur and Cliff Woolley and Philippe Vandermersch and Jonathan Cohen and John Tran and Bryan Catanzaro and Evan Shelhamer},\n eprint = {1410.0759v3},\n file = {1410.0759v3.pdf},\n month = {Nov},\n primaryclass = {cs.NE},\n title = {cuDNN: Efficient Primitives for Deep Learning},\n url = {https://arxiv.org/abs/1410.0759v3},\n year = {2014}\n}\n\n", - "citation_id": "YwdqeYZi" - }, - "arxiv:1406.1231": { - "source": "arxiv", - "identifer": "1406.1231", - "standard_citation": "arxiv:1406.1231", - "bibtex": "@article{1Dzz0P0qr,\n abstract = {Although artificial neural networks have occasionally been used for\nQuantitative Structure-Activity/Property Relationship (QSAR/QSPR) studies in\nthe past, the literature has of late been dominated by other machine learning\ntechniques such as random forests. However, a variety of new neural net\ntechniques along with successful applications in other domains have renewed\ninterest in network approaches. In this work, inspired by the winning team's\nuse of neural networks in a recent QSAR competition, we used an artificial\nneural network to learn a function that predicts activities of compounds for\nmultiple assays at the same time. We conducted experiments leveraging recent\nmethods for dealing with overfitting in neural networks as well as other tricks\nfrom the neural networks literature. We compared our methods to alternative\nmethods reported to perform well on these tasks and found that our neural net\nmethods provided superior performance.},\n archiveprefix = {arXiv},\n author = {George E. Dahl and Navdeep Jaitly and Ruslan Salakhutdinov},\n eprint = {1406.1231v1},\n file = {1406.1231v1.pdf},\n month = {Jun},\n primaryclass = {stat.ML},\n title = {Multi-task Neural Networks for QSAR Predictions},\n url = {https://arxiv.org/abs/1406.1231v1},\n year = {2014}\n}\n\n", - "citation_id": "1Dzz0P0qr" - }, - "url:http://research.google.com/archive/large_deep_networks_nips2012.html": { - "source": "url", - "identifer": "http://research.google.com/archive/large_deep_networks_nips2012.html", - "standard_citation": "url:http://research.google.com/archive/large_deep_networks_nips2012.html", - "citeproc": { - "URL": "http://research.google.com/archive/large_deep_networks_nips2012.html", - "type": "webpage", - "title": "Large Scale Distributed Deep Networks", - "container-title": "Neural Information Processing Systems 2012", - "issued": { - "date-parts": [ - [ - 2012, - 12 - ] - ] - }, - "author": [ - { - "given": "Jeffrey", - "family": "Dean" - }, - { - "given": "Greg S", - "family": "Corrado" - }, - { - "given": "Rajat", - "family": "Monga" - }, - { - "given": "Kai", - "family": "Chen" - }, - { - "given": "Matthieu", - "family": "Devin" - }, - { - "given": "Quoc V", - "family": "Le" - }, - { - "given": "Mark Z", - "family": "Mao" - }, - { - "given": "Marc’Aurelio", - "family": "Ranzato" - }, - { - "given": "Andrew", - "family": "Senior" - }, - { - "given": "Paul", - "family": "Tucker" - }, - { - "given": "Ke", - "family": "Yang" - }, - { - "given": "Andrew Y", - "family": "Ng" - } - ], - "redirects-to": "https://research.google.com/archive/large_deep_networks_nips2012.html", - "id": "17cBimWgp" - }, - "citation_id": "17cBimWgp" - }, - "url:https://github.com/deepchem/deepchem": { - "source": "url", - "identifer": "https://github.com/deepchem/deepchem", - "standard_citation": "url:https://github.com/deepchem/deepchem", - "citeproc": { - "URL": "https://github.com/deepchem/deepchem", - "title": "deepchem/deepchem", - "container-title": "GitHub", - "issued": { - "date-parts": [ - [ - 2017 - ] - ] - }, - "type": "webpage", - "id": "Ytvk62dX" - }, - "citation_id": "Ytvk62dX" - }, - "arxiv:1605.07156": { - "source": "arxiv", - "identifer": "1605.07156", - "standard_citation": "arxiv:1605.07156", - "bibtex": "@article{SAvEOARL,\n abstract = {Each human genome is a 3 billion base pair set of encoding instructions.\nDecoding the genome using deep learning fundamentally differs from most tasks, as we do not know the full structure of the data and therefore cannot design\narchitectures to suit it. As such, architectures that fit the structure of\ngenomics should be learned not prescribed. Here, we develop a novel search\nalgorithm, applicable across domains, that discovers an optimal architecture\nwhich simultaneously learns general genomic patterns and identifies the most\nimportant sequence motifs in predicting functional genomic outcomes. The\narchitectures we find using this algorithm succeed at using only RNA expression\ndata to predict gene regulatory structure, learn human-interpretable\nvisualizations of key sequence motifs, and surpass state-of-the-art results on\nbenchmark genomics challenges.},\n archiveprefix = {arXiv},\n author = {Laura Deming and Sasha Targ and Nate Sauder and Diogo Almeida and Chun Jimmie Ye},\n eprint = {1605.07156v1},\n file = {1605.07156v1.pdf},\n month = {May},\n primaryclass = {cs.LG},\n title = {Genetic Architect: Discovering Genomic Structure with Learned Neural\nArchitectures},\n url = {https://arxiv.org/abs/1605.07156v1},\n year = {2016}\n}\n\n", - "citation_id": "SAvEOARL" - }, - "doi:10.1007/978-3-319-24553-9_74": { - "source": "doi", - "identifer": "10.1007/978-3-319-24553-9_74", - "standard_citation": "doi:10.1007/978-3-319-24553-9_74", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T02:02:22Z", - "timestamp": 1506996142345 - }, - "publisher-location": "Cham", - "reference-count": 18, - "publisher": "Springer International Publishing", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2015, - 1, - 1 - ] - ], - "date-time": "2015-01-01T00:00:00Z", - "timestamp": 1420070400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015 - ] - ] - }, - "DOI": "10.1007/978-3-319-24553-9_74", - "type": "chapter", - "created": { - "date-parts": [ - [ - 2015, - 9, - 24 - ] - ], - "date-time": "2015-09-24T06:01:29Z", - "timestamp": 1443074489000 - }, - "page": "605-612", - "source": "Crossref", - "is-referenced-by-count": 6, - "title": "Deep Learning and Structured Prediction for the Segmentation of Mass in Mammograms", - "prefix": "10.1007", - "author": [ - { - "given": "Neeraj", - "family": "Dhungel", - "affiliation": [] - }, - { - "given": "Gustavo", - "family": "Carneiro", - "affiliation": [] - }, - { - "given": "Andrew P.", - "family": "Bradley", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2015, - 11, - 18 - ] - ] - }, - "container-title": "Lecture Notes in Computer Science", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1007/978-3-319-24553-9_74", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 5, - 19 - ] - ], - "date-time": "2017-05-19T14:38:34Z", - "timestamp": 1495204714000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015 - ] - ] - }, - "references-count": 18, - "URL": "https://doi.org/10.1007/978-3-319-24553-9_74", - "relation": { - "cites": [] - }, - "id": "VFw1VXDP" - }, - "citation_id": "VFw1VXDP" - }, - "doi:10.1007/978-3-319-46723-8_13": { - "source": "doi", - "identifer": "10.1007/978-3-319-46723-8_13", - "standard_citation": "doi:10.1007/978-3-319-46723-8_13", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 7 - ] - ], - "date-time": "2017-09-07T12:22:14Z", - "timestamp": 1504786934120 - }, - "publisher-location": "Cham", - "reference-count": 16, - "publisher": "Springer International Publishing", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2016, - 1, - 1 - ] - ], - "date-time": "2016-01-01T00:00:00Z", - "timestamp": 1451606400000 - }, - "delay-in-days": 0, - "content-version": "vor" - }, - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2016, - 1, - 1 - ] - ], - "date-time": "2016-01-01T00:00:00Z", - "timestamp": 1451606400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016 - ] - ] - }, - "DOI": "10.1007/978-3-319-46723-8_13", - "type": "chapter", - "created": { - "date-parts": [ - [ - 2016, - 10, - 1 - ] - ], - "date-time": "2016-10-01T03:01:21Z", - "timestamp": 1475290881000 - }, - "page": "106-114", - "source": "Crossref", - "is-referenced-by-count": 4, - "title": "The Automated Learning of Deep Features for Breast Mass Classification from Mammograms", - "prefix": "10.1007", - "author": [ - { - "given": "Neeraj", - "family": "Dhungel", - "affiliation": [] - }, - { - "given": "Gustavo", - "family": "Carneiro", - "affiliation": [] - }, - { - "given": "Andrew P.", - "family": "Bradley", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2016, - 10, - 2 - ] - ] - }, - "container-title": "Medical Image Computing and Computer-Assisted Intervention – MICCAI 2016", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1007/978-3-319-46723-8_13", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T00:08:43Z", - "timestamp": 1498349323000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016 - ] - ] - }, - "references-count": 16, - "URL": "https://doi.org/10.1007/978-3-319-46723-8_13", - "relation": { - "cites": [] - }, - "id": "JK8NuXy3" - }, - "citation_id": "JK8NuXy3" - }, - "doi:10.1016/j.media.2017.01.009": { - "source": "doi", - "identifer": "10.1016/j.media.2017.01.009", - "standard_citation": "doi:10.1016/j.media.2017.01.009", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T21:28:46Z", - "timestamp": 1502400526760 - }, - "reference-count": 62, - "publisher": "Elsevier BV", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2017, - 4, - 1 - ] - ], - "date-time": "2017-04-01T00:00:00Z", - "timestamp": 1491004800000 - }, - "delay-in-days": 0, - "content-version": "tdm" - } - ], - "content-domain": { - "domain": [ - "medicalimageanalysisjournal.com", - "elsevier.com", - "sciencedirect.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2017, - 4 - ] - ] - }, - "DOI": "10.1016/j.media.2017.01.009", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2017, - 1, - 28 - ] - ], - "date-time": "2017-01-28T15:20:55Z", - "timestamp": 1485616855000 - }, - "page": "114-128", - "update-policy": "http://dx.doi.org/10.1016/elsevier_cm_policy", - "source": "Crossref", - "is-referenced-by-count": 3, - "title": "A deep learning approach for the analysis of masses in mammograms with minimal user intervention", - "prefix": "10.1016", - "volume": "37", - "author": [ - { - "ORCID": "http://orcid.org/0000-0001-9048-2397", - "authenticated-orcid": false, - "given": "Neeraj", - "family": "Dhungel", - "affiliation": [] - }, - { - "given": "Gustavo", - "family": "Carneiro", - "affiliation": [] - }, - { - "given": "Andrew P.", - "family": "Bradley", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Medical Image Analysis", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S136184151730018X?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S136184151730018X?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T09:15:17Z", - "timestamp": 1498382117000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 4 - ] - ] - }, - "references-count": 62, - "alternative-id": [ - "S136184151730018X" - ], - "URL": "https://doi.org/10.1016/j.media.2017.01.009", - "relation": {}, - "subject": [ - "Radiological and Ultrasound Technology", - "Health Informatics", - "Radiology Nuclear Medicine and imaging", - "Computer Vision and Pattern Recognition", - "Computer Graphics and Computer-Aided Design" - ], - "container-title-short": "Medical Image Analysis", - "assertion": [ - { - "value": "Elsevier", - "name": "publisher", - "label": "This article is maintained by" - }, - { - "value": "A deep learning approach for the analysis of masses in mammograms with minimal user intervention", - "name": "articletitle", - "label": "Article Title" - }, - { - "value": "Medical Image Analysis", - "name": "journaltitle", - "label": "Journal Title" - }, - { - "value": "http://dx.doi.org/10.1016/j.media.2017.01.009", - "name": "articlelink", - "label": "CrossRef DOI link to publisher maintained version" - }, - { - "value": "article", - "name": "content_type", - "label": "Content Type" - }, - { - "value": "© 2017 Elsevier B.V. All rights reserved.", - "name": "copyright", - "label": "Copyright" - } - ], - "id": "5kfDbGhA" - }, - "citation_id": "5kfDbGhA" - }, - "doi:10.1186/s12859-015-0753-3": { - "source": "doi", - "identifer": "10.1186/s12859-015-0753-3", - "standard_citation": "doi:10.1186/s12859-015-0753-3", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 11 - ] - ], - "date-time": "2017-08-11T00:22:05Z", - "timestamp": 1502410925950 - }, - "reference-count": 45, - "publisher": "Springer Nature", - "issue": "1", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2015, - 10, - 7 - ] - ], - "date-time": "2015-10-07T00:00:00Z", - "timestamp": 1444176000000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/501100001809", - "name": "National Natural Science Foundation of China", - "doi-asserted-by": "crossref", - "award": [ - "61472078" - ] - }, - { - "name": "the National Basic Research Program of China", - "award": [ - "No. 2012CB316501" - ] - } - ], - "content-domain": { - "domain": [ - "link.springer.com" - ], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015, - 12 - ] - ] - }, - "DOI": "10.1186/s12859-015-0753-3", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 10, - 7 - ] - ], - "date-time": "2015-10-07T09:23:32Z", - "timestamp": 1444209812000 - }, - "update-policy": "http://dx.doi.org/10.1007/springer_crossmark_policy", - "source": "Crossref", - "is-referenced-by-count": 1, - "title": "DectICO: an alignment-free supervised metagenomic classification method based on feature extraction and dynamic selection", - "prefix": "10.1186", - "volume": "16", - "author": [ - { - "given": "Xiao", - "family": "Ding", - "affiliation": [] - }, - { - "given": "Fudong", - "family": "Cheng", - "affiliation": [] - }, - { - "given": "Changchang", - "family": "Cao", - "affiliation": [] - }, - { - "given": "Xiao", - "family": "Sun", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2015, - 10, - 7 - ] - ] - }, - "container-title": "BMC Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1186/s12859-015-0753-3.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://link.springer.com/article/10.1186/s12859-015-0753-3/fulltext.html", - "content-type": "text/html", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://link.springer.com/content/pdf/10.1186/s12859-015-0753-3.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 23 - ] - ], - "date-time": "2017-06-23T20:54:17Z", - "timestamp": 1498251257000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 10, - 7 - ] - ] - }, - "references-count": 45, - "alternative-id": [ - "753" - ], - "URL": "https://doi.org/10.1186/s12859-015-0753-3", - "relation": { - "cites": [] - }, - "subject": [ - "Biochemistry", - "Applied Mathematics", - "Molecular Biology", - "Structural Biology", - "Computer Science Applications" - ], - "container-title-short": "BMC Bioinformatics", - "article-number": "323", - "id": "5W4KMSdT" - }, - "citation_id": "5W4KMSdT" - }, - "doi:10.1186/s12859-015-0793-8": { - "source": "doi", - "identifer": "10.1186/s12859-015-0793-8", - "standard_citation": "doi:10.1186/s12859-015-0793-8", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 11 - ] - ], - "date-time": "2017-08-11T02:07:05Z", - "timestamp": 1502417225763 - }, - "reference-count": 31, - "publisher": "Springer Nature", - "issue": "1", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2015, - 11, - 4 - ] - ], - "date-time": "2015-11-04T00:00:00Z", - "timestamp": 1446595200000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/100000001", - "name": "National Science Foundation", - "doi-asserted-by": "crossref", - "award": [ - "1120622" - ] - }, - { - "DOI": "10.13039/100000015", - "name": "U.S. Department of Energy", - "doi-asserted-by": "publisher", - "award": [ - "SC004335" - ] - } - ], - "content-domain": { - "domain": [ - "link.springer.com" - ], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015, - 12 - ] - ] - }, - "DOI": "10.1186/s12859-015-0793-8", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 11, - 4 - ] - ], - "date-time": "2015-11-04T14:13:32Z", - "timestamp": 1446646412000 - }, - "update-policy": "http://dx.doi.org/10.1007/springer_crossmark_policy", - "source": "Crossref", - "is-referenced-by-count": 3, - "title": "Fizzy: feature subset selection for metagenomics", - "prefix": "10.1186", - "volume": "16", - "author": [ - { - "given": "Gregory", - "family": "Ditzler", - "affiliation": [] - }, - { - "given": "J. Calvin", - "family": "Morrison", - "affiliation": [] - }, - { - "given": "Yemin", - "family": "Lan", - "affiliation": [] - }, - { - "given": "Gail L.", - "family": "Rosen", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2015, - 11, - 4 - ] - ] - }, - "container-title": "BMC Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1186/s12859-015-0793-8.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://link.springer.com/article/10.1186/s12859-015-0793-8/fulltext.html", - "content-type": "text/html", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://link.springer.com/content/pdf/10.1186/s12859-015-0793-8.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 23 - ] - ], - "date-time": "2017-06-23T22:53:14Z", - "timestamp": 1498258394000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 11, - 4 - ] - ] - }, - "references-count": 31, - "alternative-id": [ - "793" - ], - "URL": "https://doi.org/10.1186/s12859-015-0793-8", - "relation": { - "cites": [] - }, - "subject": [ - "Biochemistry", - "Applied Mathematics", - "Molecular Biology", - "Structural Biology", - "Computer Science Applications" - ], - "container-title-short": "BMC Bioinformatics", - "article-number": "358", - "id": "1AN5UPfb1" - }, - "citation_id": "1AN5UPfb1" - }, - "doi:10.1109/tnnls.2014.2320415": { - "source": "doi", - "identifer": "10.1109/tnnls.2014.2320415", - "standard_citation": "doi:10.1109/tnnls.2014.2320415", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 11 - ] - ], - "date-time": "2017-08-11T02:43:22Z", - "timestamp": 1502419402107 - }, - "reference-count": 15, - "publisher": "Institute of Electrical and Electronics Engineers (IEEE)", - "issue": "4", - "license": [ - { - "URL": "http://www.ieee.org/publications_standards/publications/rights/ieeecopyrightform.pdf", - "start": { - "date-parts": [ - [ - 2015, - 4, - 1 - ] - ], - "date-time": "2015-04-01T00:00:00Z", - "timestamp": 1427846400000 - }, - "delay-in-days": 0, - "content-version": "am" - }, - { - "URL": "http://www.ieee.org/publications_standards/publications/rights/ieeecopyrightform.pdf", - "start": { - "date-parts": [ - [ - 2015, - 4, - 1 - ] - ], - "date-time": "2015-04-01T00:00:00Z", - "timestamp": 1427846400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/100000001", - "name": "National Science Foundation", - "doi-asserted-by": "publisher", - "award": [ - "ECCS-1310496", - "1120622", - "0845827", - "ECCS-0926159" - ] - }, - { - "name": "Department of Energy", - "award": [ - "SC004335" - ] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015, - 4 - ] - ] - }, - "DOI": "10.1109/tnnls.2014.2320415", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 5, - 29 - ] - ], - "date-time": "2014-05-29T18:04:03Z", - "timestamp": 1401386643000 - }, - "page": "880-886", - "source": "Crossref", - "is-referenced-by-count": 3, - "title": "A Bootstrap Based Neyman-Pearson Test for Identifying Variable Importance", - "prefix": "10.1109", - "volume": "26", - "author": [ - { - "given": "Gregory", - "family": "Ditzler", - "affiliation": [] - }, - { - "given": "Robi", - "family": "Polikar", - "affiliation": [] - }, - { - "given": "Gail", - "family": "Rosen", - "affiliation": [] - } - ], - "member": "263", - "container-title": "IEEE Transactions on Neural Networks and Learning Systems", - "original-title": [], - "link": [ - { - "URL": "http://ieeexplore.ieee.org/ielam/5962385/7061550/6823119-aam.pdf", - "content-type": "application/pdf", - "content-version": "am", - "intended-application": "syndication" - }, - { - "URL": "http://xplorestaging.ieee.org/ielx7/5962385/7061550/06823119.pdf?arnumber=6823119", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 4, - 3 - ] - ], - "date-time": "2017-04-03T21:53:25Z", - "timestamp": 1491256405000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 4 - ] - ] - }, - "references-count": 15, - "URL": "https://doi.org/10.1109/tnnls.2014.2320415", - "relation": {}, - "subject": [ - "Computer Networks and Communications", - "Software", - "Artificial Intelligence", - "Computer Science Applications" - ], - "container-title-short": "IEEE Trans. Neural Netw. Learning Syst.", - "id": "O9D66oYa" - }, - "citation_id": "O9D66oYa" - }, - "doi:10.1109/tnb.2015.2461219": { - "source": "doi", - "identifer": "10.1109/tnb.2015.2461219", - "standard_citation": "doi:10.1109/tnb.2015.2461219", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T17:42:13Z", - "timestamp": 1507052533515 - }, - "reference-count": 44, - "publisher": "Institute of Electrical and Electronics Engineers (IEEE)", - "issue": "6", - "license": [ - { - "URL": "http://www.ieee.org/publications_standards/publications/rights/ieeecopyrightform.pdf", - "start": { - "date-parts": [ - [ - 2015, - 9, - 1 - ] - ], - "date-time": "2015-09-01T00:00:00Z", - "timestamp": 1441065600000 - }, - "delay-in-days": 0, - "content-version": "am" - }, - { - "URL": "http://www.ieee.org/publications_standards/publications/rights/ieeecopyrightform.pdf", - "start": { - "date-parts": [ - [ - 2015, - 9, - 1 - ] - ], - "date-time": "2015-09-01T00:00:00Z", - "timestamp": 1441065600000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/100000001", - "name": "National Science Foundation", - "doi-asserted-by": "publisher", - "award": [ - "0926159", - "1120622", - "0845827" - ] - }, - { - "name": "Department of Energy", - "award": [] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015, - 9 - ] - ] - }, - "DOI": "10.1109/tnb.2015.2461219", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 8, - 24 - ] - ], - "date-time": "2015-08-24T18:25:26Z", - "timestamp": 1440440726000 - }, - "page": "608-616", - "source": "Crossref", - "is-referenced-by-count": 8, - "title": "Multi-Layer and Recursive Neural Networks for Metagenomic Classification", - "prefix": "10.1109", - "volume": "14", - "author": [ - { - "given": "Gregory", - "family": "Ditzler", - "affiliation": [] - }, - { - "given": "Robi", - "family": "Polikar", - "affiliation": [] - }, - { - "given": "Gail", - "family": "Rosen", - "affiliation": [] - } - ], - "member": "263", - "container-title": "IEEE Transactions on NanoBioscience", - "original-title": [], - "link": [ - { - "URL": "http://ieeexplore.ieee.org/ielam/7728/7229393/7219432-aam.pdf", - "content-type": "application/pdf", - "content-version": "am", - "intended-application": "syndication" - }, - { - "URL": "http://xplorestaging.ieee.org/ielx7/7728/7229393/07219432.pdf?arnumber=7219432", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 29 - ] - ], - "date-time": "2017-06-29T20:39:33Z", - "timestamp": 1498768773000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 9 - ] - ] - }, - "references-count": 44, - "URL": "https://doi.org/10.1109/tnb.2015.2461219", - "relation": {}, - "subject": [ - "Biotechnology", - "Electrical and Electronic Engineering", - "Medicine (miscellaneous)", - "Bioengineering", - "Pharmaceutical Science", - "Biomedical Engineering", - "Computer Science Applications" - ], - "container-title-short": "IEEE Trans.on Nanobioscience", - "id": "i38A0beL" - }, - "citation_id": "i38A0beL" - }, - "pmid:27330264": { - "source": "pmid", - "identifer": "27330264", - "standard_citation": "pmid:27330264", - "citeproc": { - "source": "PMC", - "accessed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ] - }, - "id": "188FA7whS", - "title": "Taming the Wild: A Unified Analysis of Hogwild!-Style Algorithms", - "author": [ - { - "family": "De Sa", - "given": "Christopher" - }, - { - "family": "Zhang", - "given": "Ce" - }, - { - "family": "Olukotun", - "given": "Kunle" - }, - { - "family": "Ré", - "given": "Christopher" - } - ], - "container-title-short": "Adv Neural Inf Process Syst", - "container-title": "Advances in neural information processing systems", - "issued": { - "date-parts": [ - [ - 2015, - 12 - ] - ] - }, - "page": "2656-2664", - "volume": "28", - "PMID": "27330264", - "PMCID": "PMC4907892", - "type": "article-journal", - "URL": "https://www.ncbi.nlm.nih.gov/pubmed/27330264" - }, - "citation_id": "188FA7whS" - }, - "url:http://kundajelab.github.io/dragonn/": { - "source": "url", - "identifer": "http://kundajelab.github.io/dragonn/", - "standard_citation": "url:http://kundajelab.github.io/dragonn/", - "citeproc": { - "URL": "http://kundajelab.github.io/dragonn/", - "title": "DragoNN", - "issued": { - "date-parts": [ - [ - 2016, - 11, - 6 - ] - ] - }, - "greycite-status": "Scanned", - "greycite-scanned": "2017-05-17 02:09:29", - "type": "webpage", - "id": "117PEpTMe" - }, - "citation_id": "117PEpTMe" - }, - "url:https://www.synapse.org/#!Synapse:syn6131484/wiki/402026": { - "source": "url", - "identifer": "https://www.synapse.org/#!Synapse:syn6131484/wiki/402026", - "standard_citation": "url:https://www.synapse.org/#!Synapse:syn6131484/wiki/402026", - "citeproc": { - "URL": "https://www.synapse.org/#!Synapse:syn6131484/wiki/402026", - "title": "ENCODE-DREAM in vivo Transcription Factor Binding Site Prediction Challenge", - "issued": { - "date-parts": [ - [ - 2017 - ] - ] - }, - "type": "webpage", - "id": "wW6QbBXz" - }, - "citation_id": "wW6QbBXz" - }, - "url:http://papers.nips.cc/paper/5954-convolutional-networks-on-graphs-for-learning-molecular-fingerprints": { - "source": "url", - "identifer": "http://papers.nips.cc/paper/5954-convolutional-networks-on-graphs-for-learning-molecular-fingerprints", - "standard_citation": "url:http://papers.nips.cc/paper/5954-convolutional-networks-on-graphs-for-learning-molecular-fingerprints", - "citeproc": { - "URL": "http://papers.nips.cc/paper/5954-convolutional-networks-on-graphs-for-learning-molecular-fingerprints", - "title": "Convolutional Networks on Graphs for Learning Molecular Fingerprints", - "issued": { - "date-parts": [ - [ - 2015 - ] - ] - }, - "author": [ - { - "family": "Duvenaud", - "given": "David K." - }, - { - "family": "Maclaurin", - "given": "Dougal" - }, - { - "family": "Iparraguirre", - "given": "Jorge" - }, - { - "family": "Bombarell", - "given": "Rafael" - }, - { - "family": "Hirzel", - "given": "Timothy" - }, - { - "family": "Aspuru-Guzik", - "given": "Alan" - }, - { - "family": "Adams", - "given": "Ryan P." - } - ], - "greycite-status": "Scanned", - "greycite-scanned": "2017-05-09 00:03:57", - "type": "webpage", - "id": "Oe573FaL" - }, - "citation_id": "Oe573FaL" - }, - "doi:10.1145/2771283": { - "source": "doi", - "identifer": "10.1145/2771283", - "standard_citation": "doi:10.1145/2771283", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T01:42:17Z", - "timestamp": 1506994937351 - }, - "reference-count": 0, - "publisher": "Association for Computing Machinery (ACM)", - "issue": "7", - "license": [ - { - "URL": "http://www.acm.org/publications/policies/copyright_policy#Background", - "start": { - "date-parts": [ - [ - 2015, - 6, - 25 - ] - ], - "date-time": "2015-06-25T00:00:00Z", - "timestamp": 1435190400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015, - 6, - 25 - ] - ] - }, - "DOI": "10.1145/2771283", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 6, - 25 - ] - ], - "date-time": "2015-06-25T14:36:19Z", - "timestamp": 1435242979000 - }, - "page": "14-16", - "source": "Crossref", - "is-referenced-by-count": 8, - "title": "Growing pains for deep learning", - "prefix": "10.1145", - "volume": "58", - "author": [ - { - "given": "Chris", - "family": "Edwards", - "affiliation": [] - } - ], - "member": "320", - "container-title": "Communications of the ACM", - "original-title": [], - "link": [ - { - "URL": "http://dl.acm.org/ft_gateway.cfm?id=2771283&ftid=1596834&dwn=1", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 12, - 6 - ] - ], - "date-time": "2016-12-06T21:05:22Z", - "timestamp": 1481058322000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 6, - 25 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1145/2771283", - "relation": {}, - "subject": [ - "General Computer Science" - ], - "container-title-short": "Commun. ACM", - "id": "1FocAi7N0" - }, - "citation_id": "1FocAi7N0" - }, - "url:http://www.iro.umontreal.ca/~lisa/publications2/index.php/publications/show/247": { - "source": "url", - "identifer": "http://www.iro.umontreal.ca/~lisa/publications2/index.php/publications/show/247", - "standard_citation": "url:http://www.iro.umontreal.ca/~lisa/publications2/index.php/publications/show/247", - "citeproc": { - "type": "report", - "title": "Visualizing Higher-Layer Features of a Deep Network", - "publisher": "University of Montreal", - "abstract": "Deep architectures have demonstrated state-of-the-art results in a variety of settings, especially with vision datasets. Beyond the model definitions and the quantitative analyses, there is a need for qualitative comparisons of the solutions learned by various deep architectures. The goal of this paper is to find good qualitative interpretations of high level features represented by such models. To this end, we contrast and compare several techniques applied on Stacked Denoising Autoencoders and Deep Belief Networks, trained on several vision datasets. We show that, perhaps counter-intuitively, such interpretation is possible at the unit level, that it is simple to accomplish and that the results are consistent across various techniques. We hope that such techniques will allow researchers in deep architectures to understand more of how and why deep architectures work", - "URL": "http://www.iro.umontreal.ca/~lisa/publications2/index.php/publications/show/247", - "number": "1341", - "author": [ - { - "family": "Erhan", - "given": "Dumitru" - }, - { - "family": "Bengio", - "given": "Yoshua" - }, - { - "family": "Courville", - "given": "Aaron" - }, - { - "family": "Vincent", - "given": "Pascal" - } - ], - "issued": { - "date-parts": [ - [ - "2009", - 6 - ] - ] - }, - "id": "UAAd9Uez" - }, - "citation_id": "UAAd9Uez" - }, - "url:https://github.com/maxpumperla/elephas": { - "source": "url", - "identifer": "https://github.com/maxpumperla/elephas", - "standard_citation": "url:https://github.com/maxpumperla/elephas", - "citeproc": { - "URL": "https://github.com/maxpumperla/elephas", - "title": "maxpumperla/elephas", - "container-title": "GitHub", - "issued": { - "date-parts": [ - [ - 2017 - ] - ] - }, - "type": "webpage", - "id": "y9IoEy4r" - }, - "citation_id": "y9IoEy4r" - }, - "doi:10.1101/081380": { - "source": "doi", - "identifer": "10.1101/081380", - "standard_citation": "doi:10.1101/081380", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T01:35:48Z", - "timestamp": 1502328948137 - }, - "posted": { - "date-parts": [ - [ - 2016, - 10, - 17 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2016, - 10, - 17 - ] - ] - }, - "abstract": "Numerous advances in sequencing technologies have revolutionized genomics through generating many types of genomic functional data. Statistical tools have been developed to analyze individual data types, but there lack strategies to integrate disparate datasets under a unified framework. Moreover, most analysis techniques heavily rely on feature selection and data preprocessing which increase the difficulty of addressing biological questions through the integration of multiple datasets. Here, we introduce FIDDLE (Flexible Integration of Data with Deep LEarning) an open source data-agnostic flexible integrative framework that learns a unified representation from multiple data types to infer another data type. As a case study, we use multiple Saccharomyces cerevisiae genomic datasets to predict global transcription start sites (TSS) through the simulation of TSS-seq data. We demonstrate that a type of data can be inferred from other sources of data types without manually specifying the relevant features and preprocessing. We show that models built from multiple genome-wide datasets perform profoundly better than models built from individual datasets. Thus FIDDLE learns the complex synergistic relationship within individual datasets and, importantly, across datasets.", - "DOI": "10.1101/081380", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2016, - 10, - 19 - ] - ], - "date-time": "2016-10-19T05:13:03Z", - "timestamp": 1476853983000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "FIDDLE: An integrative deep learning framework for functional genomic data inference", - "prefix": "10.1101", - "author": [ - { - "given": "Umut", - "family": "Eser", - "affiliation": [] - }, - { - "given": "L. Stirling", - "family": "Churchman", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/081380", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 1, - 4 - ] - ], - "date-time": "2017-01-04T06:17:26Z", - "timestamp": 1483510646000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 10, - 17 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/081380", - "relation": {}, - "id": "yOz8Ybj2" - }, - "citation_id": "yOz8Ybj2" - }, - "doi:10.1109/embc.2016.7590963": { - "source": "doi", - "identifer": "10.1109/embc.2016.7590963", - "standard_citation": "doi:10.1109/embc.2016.7590963", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T01:46:42Z", - "timestamp": 1502329602790 - }, - "reference-count": 24, - "publisher": "IEEE", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 8 - ] - ] - }, - "DOI": "10.1109/embc.2016.7590963", - "type": "paper-conference", - "created": { - "date-parts": [ - [ - 2016, - 10, - 20 - ] - ], - "date-time": "2016-10-20T20:57:43Z", - "timestamp": 1476997063000 - }, - "source": "Crossref", - "is-referenced-by-count": 1, - "title": "Melanoma detection by analysis of clinical images using convolutional neural network", - "prefix": "10.1109", - "author": [ - { - "given": "E.", - "family": "Nasr-Esfahani", - "affiliation": [] - }, - { - "given": "S.", - "family": "Samavi", - "affiliation": [] - }, - { - "given": "N.", - "family": "Karimi", - "affiliation": [] - }, - { - "given": "S.M.R.", - "family": "Soroushmehr", - "affiliation": [] - }, - { - "given": "M.H.", - "family": "Jafari", - "affiliation": [] - }, - { - "given": "K.", - "family": "Ward", - "affiliation": [] - }, - { - "given": "K.", - "family": "Najarian", - "affiliation": [] - } - ], - "member": "263", - "container-title": "2016 38th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC)", - "original-title": [], - "link": [ - { - "URL": "http://xplorestaging.ieee.org/ielx7/7580725/7590615/07590963.pdf?arnumber=7590963", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 11, - 16 - ] - ], - "date-time": "2016-11-16T14:31:40Z", - "timestamp": 1479306700000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 8 - ] - ] - }, - "references-count": 24, - "URL": "https://doi.org/10.1109/embc.2016.7590963", - "relation": {}, - "id": "O39LDkX" - }, - "citation_id": "O39LDkX" - }, - "doi:10.1109/ijcnn.2010.5596644": { - "source": "doi", - "identifer": "10.1109/ijcnn.2010.5596644", - "standard_citation": "doi:10.1109/ijcnn.2010.5596644", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T20:52:54Z", - "timestamp": 1502398374851 - }, - "reference-count": 23, - "publisher": "IEEE", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2010, - 7 - ] - ] - }, - "DOI": "10.1109/ijcnn.2010.5596644", - "type": "paper-conference", - "created": { - "date-parts": [ - [ - 2010, - 10, - 19 - ] - ], - "date-time": "2010-10-19T18:58:15Z", - "timestamp": 1287514695000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Neural network-based taxonomic clustering for metagenomics", - "prefix": "10.1109", - "author": [ - { - "given": "Steven D.", - "family": "Essinger", - "affiliation": [] - }, - { - "given": "Robi", - "family": "Polikar", - "affiliation": [] - }, - { - "given": "Gail L.", - "family": "Rosen", - "affiliation": [] - } - ], - "member": "263", - "container-title": "The 2010 International Joint Conference on Neural Networks (IJCNN)", - "original-title": [], - "link": [ - { - "URL": "http://xplorestaging.ieee.org/ielx5/5581822/5595732/05596644.pdf?arnumber=5596644", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 19 - ] - ], - "date-time": "2017-06-19T15:15:42Z", - "timestamp": 1497885342000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2010, - 7 - ] - ] - }, - "references-count": 23, - "URL": "https://doi.org/10.1109/ijcnn.2010.5596644", - "relation": {}, - "id": "11wVLI2Hn" - }, - "citation_id": "11wVLI2Hn" - }, - "url:http://alifar76.github.io/sklearn-metrics/": { - "source": "url", - "identifer": "http://alifar76.github.io/sklearn-metrics/", - "standard_citation": "url:http://alifar76.github.io/sklearn-metrics/", - "citeproc": { - "URL": "http://alifar76.github.io/sklearn-metrics/", - "title": "TensorFlow vs. scikit-learn : The Microbiome Challenge", - "container-title": "Ali A. Faruqi", - "issued": { - "date-parts": [ - [ - 2016, - 7, - 9 - ] - ] - }, - "author": [ - { - "family": "Faruqi", - "given": "Ali A." - } - ], - "greycite-status": "Scanned", - "greycite-scanned": "2017-05-17 02:09:46", - "archives": [ - "http://wayback.archive.org/web/http://alifar76.github.io/sklearn-metrics/" - ], - "type": "webpage", - "id": "NQ5jiN7B" - }, - "citation_id": "NQ5jiN7B" - }, - "doi:10.1101/105957": { - "source": "doi", - "identifer": "10.1101/105957", - "standard_citation": "doi:10.1101/105957", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T22:39:12Z", - "timestamp": 1502404752126 - }, - "posted": { - "date-parts": [ - [ - 2017, - 2, - 3 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2017, - 2, - 3 - ] - ] - }, - "abstract": "New architectures of multilayer artificial neural networks and new methods for training them are rapidly revolutionizing the application of machine learning in diverse fields, including business, social science, physical sciences, and biology. Interpreting deep neural networks, however, currently remains elusive, and a critical challenge lies in understanding which meaningful features a network is actually learning. We present a general method for interpreting deep neural networks and extracting network-learned features from input data. We describe our algorithm in the context of biological sequence analysis. Our approach, based on ideas from statistical physics, samples from the maximum entropy distribution over possible sequences, anchored at an input sequence and subject to constraints implied by the empirical function learned by a network. Using our framework, we demonstrate that local transcription factor binding motifs can be identified from a network trained on ChIP-seq data and that nucleosome positioning signals are indeed learned by a network trained on chemical cleavage nucleosome maps. Imposing a further constraint on the maximum entropy distribution, similar to the grand canonical ensemble in statistical physics, also allows us to probe whether a network is learning global sequence features, such as the high GC content in nucleosome-rich regions. This work thus provides valuable mathematical tools for interpreting and extracting learned features from feed-forward neural networks.", - "DOI": "10.1101/105957", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2017, - 2, - 4 - ] - ], - "date-time": "2017-02-04T06:10:16Z", - "timestamp": 1486188616000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Maximum Entropy Methods for Extracting the Learned Features of Deep Neural Networks", - "prefix": "10.1101", - "author": [ - { - "given": "Alex I", - "family": "Finnegan", - "affiliation": [] - }, - { - "given": "Jun S", - "family": "Song", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/105957", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 2, - 4 - ] - ], - "date-time": "2017-02-04T06:10:20Z", - "timestamp": 1486188620000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 2, - 3 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/105957", - "relation": {}, - "id": "VjsZbMSz" - }, - "citation_id": "VjsZbMSz" - }, - "arxiv:1704.03296": { - "source": "arxiv", - "identifer": "1704.03296", - "standard_citation": "arxiv:1704.03296", - "bibtex": "@article{y4t9EzPn,\n abstract = {As machine learning algorithms are increasingly applied to high impact yet\nhigh risk tasks, e.g. problems in health, it is critical that researchers can\nexplain how such algorithms arrived at their predictions. In recent years, a\nnumber of image saliency methods have been developed to summarize where highly\ncomplex neural networks \"look\" in an image for evidence for their predictions.\nHowever, these techniques are limited by their heuristic nature and\narchitectural constraints.\nIn this paper, we make two main contributions: First, we propose a general\nframework for learning different kinds of explanations for any black box\nalgorithm. Second, we introduce a paradigm that learns the minimally salient\npart of an image by directly editing it and learning from the corresponding\nchanges to its output. Unlike previous works, our method is model-agnostic and\ntestable because it is grounded in replicable image perturbations.},\n archiveprefix = {arXiv},\n author = {Ruth Fong and Andrea Vedaldi},\n eprint = {1704.03296v1},\n file = {1704.03296v1.pdf},\n month = {Apr},\n primaryclass = {cs.CV},\n title = {Interpretable Explanations of Black Boxes by Meaningful Perturbation},\n url = {https://arxiv.org/abs/1704.03296v1},\n year = {2017}\n}\n\n", - "citation_id": "y4t9EzPn" - }, - "arxiv:1506.02142": { - "source": "arxiv", - "identifer": "1506.02142", - "standard_citation": "arxiv:1506.02142", - "bibtex": "@article{1FDihfnM,\n abstract = {Deep learning tools have gained tremendous attention in applied machine\nlearning. However such tools for regression and classification do not capture\nmodel uncertainty. In comparison, Bayesian models offer a mathematically\ngrounded framework to reason about model uncertainty, but usually come with a\nprohibitive computational cost. In this paper we develop a new theoretical\nframework casting dropout training in deep neural networks (NNs) as approximate\nBayesian inference in deep Gaussian processes. A direct result of this theory\ngives us tools to model uncertainty with dropout NNs -- extracting information\nfrom existing models that has been thrown away so far. This mitigates the\nproblem of representing uncertainty in deep learning without sacrificing either\ncomputational complexity or test accuracy. We perform an extensive study of the\nproperties of dropout's uncertainty. Various network architectures and\nnon-linearities are assessed on tasks of regression and classification, using\nMNIST as an example. We show a considerable improvement in predictive\nlog-likelihood and RMSE compared to existing state-of-the-art methods, and\nfinish by using dropout's uncertainty in deep reinforcement learning.},\n archiveprefix = {arXiv},\n author = {Yarin Gal and Zoubin Ghahramani},\n eprint = {1506.02142v6},\n file = {1506.02142v6.pdf},\n month = {Jun},\n primaryclass = {stat.ML},\n title = {Dropout as a Bayesian Approximation: Representing Model Uncertainty in\nDeep Learning},\n url = {https://arxiv.org/abs/1506.02142v6},\n year = {2015}\n}\n\n", - "citation_id": "1FDihfnM" - }, - "doi:10.1016/j.cell.2015.11.009": { - "source": "doi", - "identifer": "10.1016/j.cell.2015.11.009", - "standard_citation": "doi:10.1016/j.cell.2015.11.009", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 24 - ] - ], - "date-time": "2017-09-24T22:22:15Z", - "timestamp": 1506291735023 - }, - "reference-count": 55, - "publisher": "Elsevier BV", - "issue": "6", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2015, - 12, - 1 - ] - ], - "date-time": "2015-12-01T00:00:00Z", - "timestamp": 1448928000000 - }, - "delay-in-days": 0, - "content-version": "tdm" - }, - { - "URL": "http://www.elsevier.com/open-access/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2016, - 12, - 3 - ] - ], - "date-time": "2016-12-03T00:00:00Z", - "timestamp": 1480723200000 - }, - "delay-in-days": 368, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "cell.com", - "elsevier.com", - "sciencedirect.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2015, - 12 - ] - ] - }, - "DOI": "10.1016/j.cell.2015.11.009", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 11, - 19 - ] - ], - "date-time": "2015-11-19T18:49:41Z", - "timestamp": 1447958981000 - }, - "page": "1400-1412", - "update-policy": "http://dx.doi.org/10.1016/elsevier_cm_policy", - "source": "Crossref", - "is-referenced-by-count": 77, - "title": "Single-Cell Genomics Unveils Critical Regulators of Th17 Cell Pathogenicity", - "prefix": "10.1016", - "volume": "163", - "author": [ - { - "given": "Jellert T.", - "family": "Gaublomme", - "affiliation": [] - }, - { - "given": "Nir", - "family": "Yosef", - "affiliation": [] - }, - { - "given": "Youjin", - "family": "Lee", - "affiliation": [] - }, - { - "given": "Rona S.", - "family": "Gertner", - "affiliation": [] - }, - { - "given": "Li V.", - "family": "Yang", - "affiliation": [] - }, - { - "given": "Chuan", - "family": "Wu", - "affiliation": [] - }, - { - "given": "Pier Paolo", - "family": "Pandolfi", - "affiliation": [] - }, - { - "given": "Tak", - "family": "Mak", - "affiliation": [] - }, - { - "given": "Rahul", - "family": "Satija", - "affiliation": [] - }, - { - "given": "Alex K.", - "family": "Shalek", - "affiliation": [] - }, - { - "given": "Vijay K.", - "family": "Kuchroo", - "affiliation": [] - }, - { - "given": "Hongkun", - "family": "Park", - "affiliation": [] - }, - { - "given": "Aviv", - "family": "Regev", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Cell", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S0092867415014890?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S0092867415014890?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T00:00:58Z", - "timestamp": 1498262458000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 12 - ] - ] - }, - "references-count": 55, - "alternative-id": [ - "S0092867415014890" - ], - "URL": "https://doi.org/10.1016/j.cell.2015.11.009", - "relation": {}, - "subject": [ - "General Biochemistry, Genetics and Molecular Biology" - ], - "container-title-short": "Cell", - "assertion": [ - { - "value": "Elsevier", - "name": "publisher", - "label": "This article is maintained by" - }, - { - "value": "Single-Cell Genomics Unveils Critical Regulators of Th17 Cell Pathogenicity", - "name": "articletitle", - "label": "Article Title" - }, - { - "value": "Cell", - "name": "journaltitle", - "label": "Journal Title" - }, - { - "value": "http://dx.doi.org/10.1016/j.cell.2015.11.009", - "name": "articlelink", - "label": "CrossRef DOI link to publisher maintained version" - }, - { - "value": "http://dx.doi.org/10.1016/j.cell.2015.10.068", - "name": "associatedlink", - "label": "CrossRef DOI link to the associated document" - }, - { - "value": "article", - "name": "content_type", - "label": "Content Type" - }, - { - "value": "Copyright © 2015 Elsevier Inc. All rights reserved.", - "name": "copyright", - "label": "Copyright" - } - ], - "id": "1HPu3R2B4" - }, - "citation_id": "1HPu3R2B4" - }, - "doi:10.1038/nrg.2015.16": { - "source": "doi", - "identifer": "10.1038/nrg.2015.16", - "standard_citation": "doi:10.1038/nrg.2015.16", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 30 - ] - ], - "date-time": "2017-09-30T00:22:12Z", - "timestamp": 1506730932764 - }, - "reference-count": 119, - "publisher": "Springer Nature", - "issue": "3", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/nrg.2015.16", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 1, - 25 - ] - ], - "date-time": "2016-01-25T09:18:09Z", - "timestamp": 1453713489000 - }, - "page": "175-188", - "source": "Crossref", - "is-referenced-by-count": 167, - "title": "Single-cell genome sequencing: current state of the science", - "prefix": "10.1038", - "volume": "17", - "author": [ - { - "given": "Charles", - "family": "Gawad", - "affiliation": [] - }, - { - "given": "Winston", - "family": "Koh", - "affiliation": [] - }, - { - "given": "Stephen R.", - "family": "Quake", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2016, - 1, - 25 - ] - ] - }, - "container-title": "Nature Reviews Genetics", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T04:26:35Z", - "timestamp": 1498278395000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 1, - 25 - ] - ] - }, - "references-count": 119, - "alternative-id": [ - "nrg.2015.16" - ], - "URL": "https://doi.org/10.1038/nrg.2015.16", - "relation": { - "cites": [] - }, - "subject": [ - "Genetics(clinical)", - "Genetics", - "Molecular Biology" - ], - "container-title-short": "Nat Rev Genet", - "id": "1AWC7HsO0" - }, - "citation_id": "1AWC7HsO0" - }, - "doi:10.1186/s13059-016-0917-0": { - "source": "doi", - "identifer": "10.1186/s13059-016-0917-0", - "standard_citation": "doi:10.1186/s13059-016-0917-0", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 8 - ] - ], - "date-time": "2017-09-08T21:42:10Z", - "timestamp": 1504906930621 - }, - "reference-count": 55, - "publisher": "Springer Nature", - "issue": "1", - "content-domain": { - "domain": [ - "link.springer.com" - ], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 12 - ] - ] - }, - "DOI": "10.1186/s13059-016-0917-0", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 3, - 23 - ] - ], - "date-time": "2016-03-23T15:59:14Z", - "timestamp": 1458748754000 - }, - "update-policy": "http://dx.doi.org/10.1007/springer_crossmark_policy", - "source": "Crossref", - "is-referenced-by-count": 31, - "title": "The real cost of sequencing: scaling computation to keep pace with data generation", - "prefix": "10.1186", - "volume": "17", - "author": [ - { - "given": "Paul", - "family": "Muir", - "affiliation": [] - }, - { - "given": "Shantao", - "family": "Li", - "affiliation": [] - }, - { - "given": "Shaoke", - "family": "Lou", - "affiliation": [] - }, - { - "given": "Daifeng", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Daniel J", - "family": "Spakowicz", - "affiliation": [] - }, - { - "given": "Leonidas", - "family": "Salichos", - "affiliation": [] - }, - { - "given": "Jing", - "family": "Zhang", - "affiliation": [] - }, - { - "given": "George M.", - "family": "Weinstock", - "affiliation": [] - }, - { - "given": "Farren", - "family": "Isaacs", - "affiliation": [] - }, - { - "given": "Joel", - "family": "Rozowsky", - "affiliation": [] - }, - { - "given": "Mark", - "family": "Gerstein", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2016, - 3, - 23 - ] - ] - }, - "container-title": "Genome Biology", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1186/s13059-016-0917-0", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T09:09:29Z", - "timestamp": 1498295369000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 3, - 23 - ] - ] - }, - "references-count": 55, - "alternative-id": [ - "917" - ], - "URL": "https://doi.org/10.1186/s13059-016-0917-0", - "relation": { - "cites": [] - }, - "container-title-short": "Genome Biol", - "article-number": "53", - "id": "1E7bFCRV4" - }, - "citation_id": "1E7bFCRV4" - }, - "doi:10.1117/12.140112": { - "source": "doi", - "identifer": "10.1117/12.140112", - "standard_citation": "doi:10.1117/12.140112", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 11 - ] - ], - "date-time": "2017-09-11T08:22:11Z", - "timestamp": 1505118131763 - }, - "reference-count": 0, - "publisher": "SPIE", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 1992, - 7, - 1 - ] - ] - }, - "DOI": "10.1117/12.140112", - "type": "paper-conference", - "created": { - "date-parts": [ - [ - 2005, - 3, - 4 - ] - ], - "date-time": "2005-03-04T23:04:16Z", - "timestamp": 1109977456000 - }, - "source": "Crossref", - "is-referenced-by-count": 1, - "title": "Sequence learning with recurrent networks: analysis of internal representations", - "prefix": "10.1117", - "author": [ - { - "given": "Joydeep", - "family": "Ghosh", - "affiliation": [] - }, - { - "given": "Vijay", - "family": "Karamcheti", - "affiliation": [] - } - ], - "member": "189", - "container-title": "Science of Artificial Neural Networks", - "original-title": [], - "link": [ - { - "URL": "http://proceedings.spiedigitallibrary.org/proceeding.aspx?articleid=993054", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 3, - 7 - ] - ], - "date-time": "2017-03-07T02:25:12Z", - "timestamp": 1488853512000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Dennis W.", - "family": "Ruck", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 1992, - 7, - 1 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1117/12.140112", - "relation": {}, - "id": "oc44yBj0" - }, - "citation_id": "oc44yBj0" - }, - "arxiv:1610.02415": { - "source": "arxiv", - "identifer": "1610.02415", - "standard_citation": "arxiv:1610.02415", - "bibtex": "@article{2dU8f4XJ,\n abstract = {We report a method to convert discrete representations of molecules to and\nfrom a multidimensional continuous representation. This generative model allows\nefficient search and optimization through open-ended spaces of chemical\ncompounds. We train deep neural networks on hundreds of thousands of existing\nchemical structures to construct two coupled functions: an encoder and a\ndecoder. The encoder converts the discrete representation of a molecule into a\nreal-valued continuous vector, and the decoder converts these continuous\nvectors back to the discrete representation from this latent space. Continuous\nrepresentations allow us to automatically generate novel chemical structures by\nperforming simple operations in the latent space, such as decoding random\nvectors, perturbing known chemical structures, or interpolating between\nmolecules. Continuous representations also allow the use of powerful\ngradient-based optimization to efficiently guide the search for optimized\nfunctional compounds. We demonstrate our method in the design of drug-like\nmolecules as well as organic light-emitting diodes.},\n archiveprefix = {arXiv},\n author = {Rafael Gómez-Bombarelli and David Duvenaud and José Miguel Hernández-Lobato and Jorge Aguilera-Iparraguirre and Timothy D. Hirzel and Ryan P. Adams and Alán Aspuru-Guzik},\n eprint = {1610.02415v2},\n file = {1610.02415v2.pdf},\n month = {Nov},\n primaryclass = {cs.LG},\n title = {Automatic chemical design using a data-driven continuous representation\nof molecules},\n url = {https://arxiv.org/abs/1610.02415v2},\n year = {2016}\n}\n\n", - "citation_id": "2dU8f4XJ" - }, - "doi:10.14778/2212351.2212354": { - "source": "doi", - "identifer": "10.14778/2212351.2212354", - "standard_citation": "doi:10.14778/2212351.2212354", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T04:22:16Z", - "timestamp": 1507004536513 - }, - "reference-count": 0, - "publisher": "VLDB Endowment", - "issue": "8", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2012, - 4, - 1 - ] - ] - }, - "DOI": "10.14778/2212351.2212354", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 6, - 24 - ] - ], - "date-time": "2014-06-24T12:17:57Z", - "timestamp": 1403612277000 - }, - "page": "716-727", - "source": "Crossref", - "is-referenced-by-count": 425, - "title": "Distributed GraphLab", - "prefix": "10.14778", - "volume": "5", - "author": [ - { - "given": "Yucheng", - "family": "Low", - "affiliation": [] - }, - { - "given": "Danny", - "family": "Bickson", - "affiliation": [] - }, - { - "given": "Joseph", - "family": "Gonzalez", - "affiliation": [] - }, - { - "given": "Carlos", - "family": "Guestrin", - "affiliation": [] - }, - { - "given": "Aapo", - "family": "Kyrola", - "affiliation": [] - }, - { - "given": "Joseph M.", - "family": "Hellerstein", - "affiliation": [] - } - ], - "member": "5777", - "container-title": "Proceedings of the VLDB Endowment", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2014, - 6, - 24 - ] - ], - "date-time": "2014-06-24T12:28:24Z", - "timestamp": 1403612904000 - }, - "score": 1.0, - "subtitle": [ - "a framework for machine learning and data mining in the cloud" - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2012, - 4, - 1 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.14778/2212351.2212354", - "relation": {}, - "container-title-short": "Proc. VLDB Endow.", - "id": "1XcexUAV" - }, - "citation_id": "1XcexUAV" - }, - "url:http://www.fasebj.org/content/30/1_Supplement/406.3": { - "source": "url", - "identifer": "http://www.fasebj.org/content/30/1_Supplement/406.3", - "standard_citation": "url:http://www.fasebj.org/content/30/1_Supplement/406.3", - "citeproc": { - "URL": "http://www.fasebj.org/content/30/1_Supplement/406.3", - "title": "Utilizing Machine Learning Approaches to Understand the Interrelationship of Diet, the Human Gastrointestinal Microbiome, and Health", - "container-title": "The FASEB Journal", - "issued": { - "date-parts": [ - [ - 2016, - 4, - 1 - ] - ] - }, - "author": [ - { - "family": "Guetterman", - "given": "Heather" - }, - { - "family": "Auvil", - "given": "Loretta" - }, - { - "family": "Russell", - "given": "Nate" - }, - { - "family": "Welge", - "given": "Michael" - }, - { - "family": "Berry", - "given": "Matt" - }, - { - "family": "Gatzke", - "given": "Lisa" - }, - { - "family": "Bushell", - "given": "Colleen" - }, - { - "family": "Holscher", - "given": "Hannah" - } - ], - "greycite-status": "Scanned", - "greycite-scanned": "2017-05-17 02:09:57", - "archives": [ - "http://wayback.archive.org/web/http://www.fasebj.org/content/30/1_Supplement/406.3" - ], - "type": "webpage", - "id": "W0cYSf89" - }, - "citation_id": "W0cYSf89" - }, - "doi:10.1136/amiajnl-2013-001815": { - "source": "doi", - "identifer": "10.1136/amiajnl-2013-001815", - "standard_citation": "doi:10.1136/amiajnl-2013-001815", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 15 - ] - ], - "date-time": "2017-09-15T20:42:09Z", - "timestamp": 1505508129525 - }, - "reference-count": 54, - "publisher": "Oxford University Press (OUP)", - "issue": "2", - "content-domain": { - "domain": [ - "bmj.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2014, - 3 - ] - ] - }, - "DOI": "10.1136/amiajnl-2013-001815", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2013, - 8, - 20 - ] - ], - "date-time": "2013-08-20T04:22:03Z", - "timestamp": 1376972523000 - }, - "page": "315-325", - "update-policy": "http://dx.doi.org/10.1136/crossmarkpolicy", - "source": "Crossref", - "is-referenced-by-count": 20, - "title": "From vital signs to clinical outcomes for patients with sepsis: a machine learning basis for a clinical decision support system", - "prefix": "10.1093", - "volume": "21", - "author": [ - { - "given": "Eren", - "family": "Gultepe", - "affiliation": [] - }, - { - "given": "Jeffrey P", - "family": "Green", - "affiliation": [] - }, - { - "given": "Hien", - "family": "Nguyen", - "affiliation": [] - }, - { - "given": "Jason", - "family": "Adams", - "affiliation": [] - }, - { - "given": "Timothy", - "family": "Albertson", - "affiliation": [] - }, - { - "given": "Ilias", - "family": "Tagkopoulos", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2014, - 3, - 1 - ] - ] - }, - "container-title": "Journal of the American Medical Informatics Association", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/jamia/article-pdf/doi/10.1136/amiajnl-2013-001815/5964958/21-2-315.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 23 - ] - ], - "date-time": "2017-08-23T06:03:17Z", - "timestamp": 1503468197000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 3 - ] - ] - }, - "references-count": 54, - "alternative-id": [ - "10.1136/amiajnl-2013-001815" - ], - "URL": "https://doi.org/10.1136/amiajnl-2013-001815", - "relation": {}, - "subject": [ - "Health Informatics" - ], - "container-title-short": "J Am Med Inform Assoc", - "id": "eCrLGgiX" - }, - "citation_id": "eCrLGgiX" - }, - "doi:10.1101/031906": { - "source": "doi", - "identifer": "10.1101/031906", - "standard_citation": "doi:10.1101/031906", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 11 - ] - ], - "date-time": "2017-08-11T02:48:50Z", - "timestamp": 1502419730323 - }, - "posted": { - "date-parts": [ - [ - 2015, - 11, - 16 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2015, - 11, - 16 - ] - ] - }, - "abstract": "Genes play a central role in all biological processes. DNA microarray technology has made it possible to study the expression behavior of thousands of genes in one go. Often, gene expression data is used to generate features for supervised and unsupervised learning tasks. At the same time, advances in the field of deep learning have made available a plethora of architectures. In this paper, we use deep architectures pre-trained in an unsupervised manner using denoising autoencoders as a preprocessing step for a popular unsupervised learning task. Denoising autoencoders (DA) can be used to learn a compact representation of input, and have been used to generate features for further supervised learning tasks. We propose that our deep architectures can be treated as empirical versions of Deep Belief Networks (DBNs). We use our deep architectures to regenerate gene expression time series data for two different data sets. We test our hypothesis on two popular datasets for the unsupervised learning task of clustering and find promising improvements in performance.", - "DOI": "10.1101/031906", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2015, - 11, - 17 - ] - ], - "date-time": "2015-11-17T06:10:42Z", - "timestamp": 1447740642000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Learning structure in gene expression data using deep architectures, with an application to gene clustering", - "prefix": "10.1101", - "author": [ - { - "given": "Aman", - "family": "Gupta", - "affiliation": [] - }, - { - "given": "Haohan", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Madhavi", - "family": "Ganapathiraju", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/031906", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 1, - 4 - ] - ], - "date-time": "2017-01-04T01:58:36Z", - "timestamp": 1483495116000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 11, - 16 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/031906", - "relation": {}, - "id": "AnenJOuU" - }, - "citation_id": "AnenJOuU" - }, - "arxiv:1502.02551": { - "source": "arxiv", - "identifer": "1502.02551", - "standard_citation": "arxiv:1502.02551", - "bibtex": "@article{CKcJuj03,\n abstract = {Training of large-scale deep neural networks is often constrained by the\navailable computational resources. We study the effect of limited precision\ndata representation and computation on neural network training. Within the\ncontext of low-precision fixed-point computations, we observe the rounding\nscheme to play a crucial role in determining the network's behavior during\ntraining. Our results show that deep networks can be trained using only 16-bit\nwide fixed-point number representation when using stochastic rounding, and\nincur little to no degradation in the classification accuracy. We also\ndemonstrate an energy-efficient hardware accelerator that implements\nlow-precision fixed-point arithmetic with stochastic rounding.},\n archiveprefix = {arXiv},\n author = {Suyog Gupta and Ankur Agrawal and Kailash Gopalakrishnan and Pritish Narayanan},\n eprint = {1502.02551v1},\n file = {1502.02551v1.pdf},\n month = {Feb},\n primaryclass = {cs.LG},\n title = {Deep Learning with Limited Numerical Precision},\n url = {https://arxiv.org/abs/1502.02551v1},\n year = {2015}\n}\n\n", - "citation_id": "CKcJuj03" - }, - "arxiv:1504.04343": { - "source": "arxiv", - "identifer": "1504.04343", - "standard_citation": "arxiv:1504.04343", - "bibtex": "@article{13KjSCKB2,\n abstract = {We present Caffe con Troll (CcT), a fully compatible end-to-end version of\nthe popular framework Caffe with rebuilt internals. We built CcT to examine the\nperformance characteristics of training and deploying general-purpose\nconvolutional neural networks across different hardware architectures. We find\nthat, by employing standard batching optimizations for CPU training, we achieve\na 4.5x throughput improvement over Caffe on popular networks like CaffeNet.\nMoreover, with these improvements, the end-to-end training time for CNNs is\ndirectly proportional to the FLOPS delivered by the CPU, which enables us to\nefficiently train hybrid CPU-GPU systems for CNNs.},\n archiveprefix = {arXiv},\n author = {Stefan Hadjis and Firas Abuzaid and Ce Zhang and Christopher Ré},\n eprint = {1504.04343v2},\n file = {1504.04343v2.pdf},\n month = {Apr},\n primaryclass = {cs.LG},\n title = {Caffe con Troll: Shallow Ideas to Speed Up Deep Learning},\n url = {https://arxiv.org/abs/1504.04343v2},\n year = {2015}\n}\n\n", - "citation_id": "13KjSCKB2" - }, - "arxiv:1512.03385": { - "source": "arxiv", - "identifer": "1512.03385", - "standard_citation": "arxiv:1512.03385", - "bibtex": "@article{j7KrVyi8,\n abstract = {Deeper neural networks are more difficult to train. We present a residual\nlearning framework to ease the training of networks that are substantially\ndeeper than those used previously. We explicitly reformulate the layers as\nlearning residual functions with reference to the layer inputs, instead of\nlearning unreferenced functions. We provide comprehensive empirical evidence\nshowing that these residual networks are easier to optimize, and can gain\naccuracy from considerably increased depth. On the ImageNet dataset we evaluate\nresidual nets with a depth of up to 152 layers---8x deeper than VGG nets but\nstill having lower complexity. An ensemble of these residual nets achieves\n3.57% error on the ImageNet test set. This result won the 1st place on the\nILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100\nand 1000 layers.\nThe depth of representations is of central importance for many visual\nrecognition tasks. Solely due to our extremely deep representations, we obtain\na 28% relative improvement on the COCO object detection dataset. Deep residual\nnets are foundations of our submissions to ILSVRC & COCO 2015 competitions, where we also won the 1st places on the tasks of ImageNet detection, ImageNet\nlocalization, COCO detection, and COCO segmentation.},\n archiveprefix = {arXiv},\n author = {Kaiming He and Xiangyu Zhang and Shaoqing Ren and Jian Sun},\n eprint = {1512.03385v1},\n file = {1512.03385v1.pdf},\n month = {12},\n primaryclass = {cs.CV},\n title = {Deep Residual Learning for Image Recognition},\n url = {https://arxiv.org/abs/1512.03385v1},\n year = {2015}\n}\n\n", - "citation_id": "j7KrVyi8" - }, - "arxiv:1503.02531": { - "source": "arxiv", - "identifer": "1503.02531", - "standard_citation": "arxiv:1503.02531", - "bibtex": "@article{1CRF3gAV,\n abstract = {A very simple way to improve the performance of almost any machine learning\nalgorithm is to train many different models on the same data and then to\naverage their predictions. Unfortunately, making predictions using a whole\nensemble of models is cumbersome and may be too computationally expensive to\nallow deployment to a large number of users, especially if the individual\nmodels are large neural nets. Caruana and his collaborators have shown that it\nis possible to compress the knowledge in an ensemble into a single model which\nis much easier to deploy and we develop this approach further using a different\ncompression technique. We achieve some surprising results on MNIST and we show\nthat we can significantly improve the acoustic model of a heavily used\ncommercial system by distilling the knowledge in an ensemble of models into a\nsingle model. We also introduce a new type of ensemble composed of one or more\nfull models and many specialist models which learn to distinguish fine-grained\nclasses that the full models confuse. Unlike a mixture of experts, these\nspecialist models can be trained rapidly and in parallel.},\n archiveprefix = {arXiv},\n author = {Geoffrey Hinton and Oriol Vinyals and Jeff Dean},\n eprint = {1503.02531v1},\n file = {1503.02531v1.pdf},\n month = {Mar},\n primaryclass = {stat.ML},\n title = {Distilling the Knowledge in a Neural Network},\n url = {https://arxiv.org/abs/1503.02531v1},\n year = {2015}\n}\n\n", - "citation_id": "1CRF3gAV" - }, - "doi:10.1093/bioinformatics/btm247": { - "source": "doi", - "identifer": "10.1093/bioinformatics/btm247", - "standard_citation": "doi:10.1093/bioinformatics/btm247", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 22 - ] - ], - "date-time": "2017-08-22T14:41:09Z", - "timestamp": 1503412869103 - }, - "reference-count": 26, - "publisher": "Oxford University Press (OUP)", - "issue": "14", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2007, - 7, - 15 - ] - ] - }, - "DOI": "10.1093/bioinformatics/btm247", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2007, - 5, - 9 - ] - ], - "date-time": "2007-05-09T02:24:29Z", - "timestamp": 1178677469000 - }, - "page": "1728-1736", - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Fast model-based protein homology detection without alignment", - "prefix": "10.1093", - "volume": "23", - "author": [ - { - "given": "S.", - "family": "Hochreiter", - "affiliation": [] - }, - { - "given": "M.", - "family": "Heusel", - "affiliation": [] - }, - { - "given": "K.", - "family": "Obermayer", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2007, - 5, - 8 - ] - ] - }, - "container-title": "Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/bioinformatics/article-pdf/23/14/1728/466668/btm247.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 22 - ] - ], - "date-time": "2017-08-22T14:16:06Z", - "timestamp": 1503411366000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2007, - 5, - 8 - ] - ] - }, - "references-count": 26, - "URL": "https://doi.org/10.1093/bioinformatics/btm247", - "relation": {}, - "subject": [ - "Statistics and Probability", - "Computational Theory and Mathematics", - "Biochemistry", - "Molecular Biology", - "Computational Mathematics", - "Computer Science Applications" - ], - "container-title-short": "Bioinformatics", - "id": "G8RKF6sz" - }, - "citation_id": "G8RKF6sz" - }, - "doi:10.1093/nar/gkp327": { - "source": "doi", - "identifer": "10.1093/nar/gkp327", - "standard_citation": "doi:10.1093/nar/gkp327", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T11:22:17Z", - "timestamp": 1507116137272 - }, - "reference-count": 13, - "publisher": "Oxford University Press (OUP)", - "issue": "suppl_2", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2009, - 7, - 1 - ] - ] - }, - "DOI": "10.1093/nar/gkp327", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2009, - 5, - 9 - ] - ], - "date-time": "2009-05-09T00:49:45Z", - "timestamp": 1241830185000 - }, - "page": "W101-W105", - "source": "Crossref", - "is-referenced-by-count": 50, - "title": "Orphelia: predicting genes in metagenomic sequencing reads", - "prefix": "10.1093", - "volume": "37", - "author": [ - { - "given": "Katharina J.", - "family": "Hoff", - "affiliation": [] - }, - { - "given": "Thomas", - "family": "Lingner", - "affiliation": [] - }, - { - "given": "Peter", - "family": "Meinicke", - "affiliation": [] - }, - { - "given": "Maike", - "family": "Tech", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2009, - 5, - 8 - ] - ] - }, - "container-title": "Nucleic Acids Research", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/nar/article-pdf/37/suppl_2/W101/16758085/gkp327.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 23 - ] - ], - "date-time": "2017-08-23T20:15:17Z", - "timestamp": 1503519317000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2009, - 5, - 8 - ] - ] - }, - "references-count": 13, - "URL": "https://doi.org/10.1093/nar/gkp327", - "relation": {}, - "subject": [ - "Genetics" - ], - "id": "q1A2AEtO" - }, - "citation_id": "q1A2AEtO" - }, - "doi:10.1093/nar/20.16.4331": { - "source": "doi", - "identifer": "10.1093/nar/20.16.4331", - "standard_citation": "doi:10.1093/nar/20.16.4331", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 24 - ] - ], - "date-time": "2017-08-24T01:40:52Z", - "timestamp": 1503538852920 - }, - "reference-count": 0, - "publisher": "Oxford University Press (OUP)", - "issue": "16", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 1992 - ] - ] - }, - "DOI": "10.1093/nar/20.16.4331", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2007, - 1, - 5 - ] - ], - "date-time": "2007-01-05T02:27:37Z", - "timestamp": 1167964057000 - }, - "page": "4331-4338", - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "An assessment of neural network and statistical approaches for prediction of E.coli Promoter sites", - "prefix": "10.1093", - "volume": "20", - "author": [ - { - "given": "Paul B.", - "family": "Horton", - "affiliation": [] - }, - { - "given": "Minoru", - "family": "Kanehisa", - "affiliation": [] - } - ], - "member": "286", - "container-title": "Nucleic Acids Research", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/nar/article-pdf/20/16/4331/3858645/20-16-4331.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 23 - ] - ], - "date-time": "2017-08-23T19:49:04Z", - "timestamp": 1503517744000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 1992 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1093/nar/20.16.4331", - "relation": {}, - "subject": [ - "Genetics" - ], - "container-title-short": "Nucl Acids Res", - "id": "uZvDdFZo" - }, - "citation_id": "uZvDdFZo" - }, - "arxiv:1609.07061": { - "source": "arxiv", - "identifer": "1609.07061", - "standard_citation": "arxiv:1609.07061", - "bibtex": "@article{1GUizyE8e,\n abstract = {We introduce a method to train Quantized Neural Networks (QNNs) --- neural\nnetworks with extremely low precision (e.g., 1-bit) weights and activations, at\nrun-time. At train-time the quantized weights and activations are used for\ncomputing the parameter gradients. During the forward pass, QNNs drastically\nreduce memory size and accesses, and replace most arithmetic operations with\nbit-wise operations. As a result, power consumption is expected to be\ndrastically reduced. We trained QNNs over the MNIST, CIFAR-10, SVHN and\nImageNet datasets. The resulting QNNs achieve prediction accuracy comparable to\ntheir 32-bit counterparts. For example, our quantized version of AlexNet with\n1-bit weights and 2-bit activations achieves $51\\%$ top-1 accuracy. Moreover, we quantize the parameter gradients to 6-bits as well which enables gradients\ncomputation using only bit-wise operation. Quantized recurrent neural networks\nwere tested over the Penn Treebank dataset, and achieved comparable accuracy as\ntheir 32-bit counterparts using only 4-bits. Last but not least, we programmed\na binary matrix multiplication GPU kernel with which it is possible to run our\nMNIST QNN 7 times faster than with an unoptimized GPU kernel, without suffering\nany loss in classification accuracy. The QNN code is available online.},\n archiveprefix = {arXiv},\n author = {Itay Hubara and Matthieu Courbariaux and Daniel Soudry and Ran El-Yaniv and Yoshua Bengio},\n eprint = {1609.07061v1},\n file = {1609.07061v1.pdf},\n month = {Sep},\n primaryclass = {cs.NE},\n title = {Quantized Neural Networks: Training Neural Networks with Low Precision\nWeights and Activations},\n url = {https://arxiv.org/abs/1609.07061v1},\n year = {2016}\n}\n\n", - "citation_id": "1GUizyE8e" - }, - "doi:10.1109/access.2016.2618775": { - "source": "doi", - "identifer": "10.1109/access.2016.2618775", - "standard_citation": "doi:10.1109/access.2016.2618775", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T01:39:53Z", - "timestamp": 1502329193323 - }, - "reference-count": 67, - "publisher": "Institute of Electrical and Electronics Engineers (IEEE)", - "license": [ - { - "URL": "http://www.ieee.org/publications_standards/publications/rights/oapa.pdf", - "start": { - "date-parts": [ - [ - 2016, - 1, - 1 - ] - ], - "date-time": "2016-01-01T00:00:00Z", - "timestamp": 1451606400000 - }, - "delay-in-days": 0, - "content-version": "am" - }, - { - "URL": "http://www.ieee.org/publications_standards/publications/rights/oapa.pdf", - "start": { - "date-parts": [ - [ - 2016, - 1, - 1 - ] - ], - "date-time": "2016-01-01T00:00:00Z", - "timestamp": 1451606400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/100000001", - "name": "National Science Foundation", - "doi-asserted-by": "publisher", - "award": [ - "IIS-1527827", - "IIS-1231742" - ] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016 - ] - ] - }, - "DOI": "10.1109/access.2016.2618775", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 10, - 19 - ] - ], - "date-time": "2016-10-19T18:14:44Z", - "timestamp": 1476900884000 - }, - "page": "7988-8001", - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Predicting Complications in Critical Care Using Heterogeneous Clinical Data", - "prefix": "10.1109", - "volume": "4", - "author": [ - { - "given": "Vijay", - "family": "Huddar", - "affiliation": [] - }, - { - "given": "Bapu Koundinya", - "family": "Desiraju", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-6748-6864", - "authenticated-orcid": false, - "given": "Vaibhav", - "family": "Rajan", - "affiliation": [] - }, - { - "given": "Sakyajit", - "family": "Bhattacharya", - "affiliation": [] - }, - { - "given": "Shourya", - "family": "Roy", - "affiliation": [] - }, - { - "given": "Chandan K.", - "family": "Reddy", - "affiliation": [] - } - ], - "member": "263", - "container-title": "IEEE Access", - "original-title": [], - "link": [ - { - "URL": "http://ieeexplore.ieee.org/ielam/6287639/7419931/7593335-aam.pdf", - "content-type": "application/pdf", - "content-version": "am", - "intended-application": "syndication" - }, - { - "URL": "http://xplorestaging.ieee.org/ielx7/6287639/7419931/07593335.pdf?arnumber=7593335", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T01:35:26Z", - "timestamp": 1498354526000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016 - ] - ] - }, - "references-count": 67, - "URL": "https://doi.org/10.1109/access.2016.2618775", - "relation": {}, - "subject": [ - "General Engineering", - "General Materials Science", - "General Computer Science" - ], - "container-title-short": "IEEE Access", - "id": "fOaBA9Vc" - }, - "citation_id": "fOaBA9Vc" - }, - "doi:10.1021/acscentsci.6b00162": { - "source": "doi", - "identifer": "10.1021/acscentsci.6b00162", - "standard_citation": "doi:10.1021/acscentsci.6b00162", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 12 - ] - ], - "date-time": "2017-09-12T14:40:42Z", - "timestamp": 1505227242792 - }, - "reference-count": 68, - "publisher": "American Chemical Society (ACS)", - "issue": "8", - "license": [ - { - "URL": "http://pubs.acs.org/page/policy/authorchoice_termsofuse.html", - "start": { - "date-parts": [ - [ - 2016, - 7, - 29 - ] - ], - "date-time": "2016-07-29T00:00:00Z", - "timestamp": 1469750400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/100000092", - "name": "U.S. National Library of Medicine", - "doi-asserted-by": "publisher", - "award": [ - "R01LM012222" - ] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 8, - 24 - ] - ] - }, - "DOI": "10.1021/acscentsci.6b00162", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 7, - 29 - ] - ], - "date-time": "2016-07-29T12:52:04Z", - "timestamp": 1469796724000 - }, - "page": "529-537", - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Modeling Reactivity to Biological Macromolecules with a Deep Multitask Network", - "prefix": "10.1021", - "volume": "2", - "author": [ - { - "given": "Tyler B.", - "family": "Hughes", - "affiliation": [] - }, - { - "given": "Na Le", - "family": "Dang", - "affiliation": [] - }, - { - "given": "Grover P.", - "family": "Miller", - "affiliation": [] - }, - { - "given": "S. Joshua", - "family": "Swamidass", - "affiliation": [] - } - ], - "member": "316", - "container-title": "ACS Central Science", - "original-title": [], - "link": [ - { - "URL": "http://pubs.acs.org/doi/pdf/10.1021/acscentsci.6b00162", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "unspecified" - }, - { - "URL": "http://pubs.acs.org/doi/pdf/10.1021/acscentsci.6b00162", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 9, - 12 - ] - ], - "date-time": "2017-09-12T14:16:34Z", - "timestamp": 1505225794000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 8, - 24 - ] - ] - }, - "references-count": 68, - "alternative-id": [ - "10.1021/acscentsci.6b00162" - ], - "URL": "https://doi.org/10.1021/acscentsci.6b00162", - "relation": {}, - "container-title-short": "ACS Cent. Sci.", - "id": "1BARarxfz" - }, - "citation_id": "1BARarxfz" - }, - "doi:10.1016/j.jalz.2015.01.010": { - "source": "doi", - "identifer": "10.1016/j.jalz.2015.01.010", - "standard_citation": "doi:10.1016/j.jalz.2015.01.010", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 14 - ] - ], - "date-time": "2017-09-14T15:02:13Z", - "timestamp": 1505401333702 - }, - "reference-count": 28, - "publisher": "Elsevier BV", - "issue": "12", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2015, - 12, - 1 - ] - ], - "date-time": "2015-12-01T00:00:00Z", - "timestamp": 1448928000000 - }, - "delay-in-days": 0, - "content-version": "tdm" - } - ], - "funder": [ - { - "DOI": "10.13039/100000002", - "name": "National Institutes of Health", - "doi-asserted-by": "publisher", - "award": [ - "R01 AG040396", - "R01 AG021155" - ] - }, - { - "DOI": "10.13039/100005902", - "name": "Institute for Clinical and Translational Research, University of Wisconsin, Madison", - "doi-asserted-by": "crossref", - "award": [ - "1UL1RR025011" - ] - } - ], - "content-domain": { - "domain": [ - "clinicalkey.jp", - "clinicalkey.com", - "clinicalkey.es", - "alzheimersanddementia.org", - "clinicalkey.com.au", - "clinicalkey.fr", - "elsevier.com", - "sciencedirect.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2015, - 12 - ] - ] - }, - "DOI": "10.1016/j.jalz.2015.01.010", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 6, - 18 - ] - ], - "date-time": "2015-06-18T11:57:50Z", - "timestamp": 1434628670000 - }, - "page": "1489-1499", - "update-policy": "http://dx.doi.org/10.1016/elsevier_cm_policy", - "source": "Crossref", - "is-referenced-by-count": 8, - "title": "Imaging-based enrichment criteria using deep learning algorithms for efficient clinical trials in mild cognitive impairment", - "prefix": "10.1016", - "volume": "11", - "author": [ - { - "given": "Vamsi K.", - "family": "Ithapu", - "affiliation": [] - }, - { - "given": "Vikas", - "family": "Singh", - "affiliation": [] - }, - { - "given": "Ozioma C.", - "family": "Okonkwo", - "affiliation": [] - }, - { - "given": "Richard J.", - "family": "Chappell", - "affiliation": [] - }, - { - "given": "N. Maritza", - "family": "Dowling", - "affiliation": [] - }, - { - "given": "Sterling C.", - "family": "Johnson", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Alzheimer's & Dementia", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S1552526015001922?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S1552526015001922?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 7, - 28 - ] - ], - "date-time": "2017-07-28T01:31:05Z", - "timestamp": 1501205465000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 12 - ] - ] - }, - "references-count": 28, - "alternative-id": [ - "S1552526015001922" - ], - "URL": "https://doi.org/10.1016/j.jalz.2015.01.010", - "relation": {}, - "subject": [ - "Health Policy", - "Developmental Neuroscience", - "Epidemiology", - "Cellular and Molecular Neuroscience", - "Geriatrics and Gerontology", - "Psychiatry and Mental health", - "Clinical Neurology" - ], - "container-title-short": "Alzheimer's & Dementia", - "assertion": [ - { - "value": "Elsevier", - "name": "publisher", - "label": "This article is maintained by" - }, - { - "value": "Imaging-based enrichment criteria using deep learning algorithms for efficient clinical trials in mild cognitive impairment", - "name": "articletitle", - "label": "Article Title" - }, - { - "value": "Alzheimer's & Dementia", - "name": "journaltitle", - "label": "Journal Title" - }, - { - "value": "http://dx.doi.org/10.1016/j.jalz.2015.01.010", - "name": "articlelink", - "label": "CrossRef DOI link to publisher maintained version" - }, - { - "value": "article", - "name": "content_type", - "label": "Content Type" - }, - { - "value": "Copyright © 2015 The Alzheimer's Association. Published by Elsevier B.V. All rights reserved.", - "name": "copyright", - "label": "Copyright" - } - ], - "id": "eehGXQlY" - }, - "citation_id": "eehGXQlY" - }, - "arxiv:1609.02374": { - "source": "arxiv", - "identifer": "1609.02374", - "standard_citation": "arxiv:1609.02374", - "bibtex": "@article{1AJUcl1KV,\n abstract = {Melanoma is amongst most aggressive types of cancer. However, it is highly\ncurable if detected in its early stages. Prescreening of suspicious moles and\nlesions for malignancy is of great importance. Detection can be done by images\ncaptured by standard cameras, which are more preferable due to low cost and\navailability. One important step in computerized evaluation of skin lesions is\naccurate detection of lesion region, i.e. segmentation of an image into two\nregions as lesion and normal skin. Accurate segmentation can be challenging due\nto burdens such as illumination variation and low contrast between lesion and\nhealthy skin. In this paper, a method based on deep neural networks is proposed\nfor accurate extraction of a lesion region. The input image is preprocessed and\nthen its patches are fed to a convolutional neural network (CNN). Local texture\nand global structure of the patches are processed in order to assign pixels to\nlesion or normal classes. A method for effective selection of training patches\nis used for more accurate detection of a lesion border. The output segmentation\nmask is refined by some post processing operations. The experimental results of\nqualitative and quantitative evaluations demonstrate that our method can\noutperform other state-of-the-art algorithms exist in the literature.},\n archiveprefix = {arXiv},\n author = {Mohammad H. Jafari and Ebrahim Nasr-Esfahani and Nader Karimi and S. M. Reza Soroushmehr and Shadrokh Samavi and Kayvan Najarian},\n doi = {10.1007/s11548-017-1567-8},\n eprint = {1609.02374v1},\n file = {1609.02374v1.pdf},\n month = {Sep},\n primaryclass = {cs.CV},\n title = {Extraction of Skin Lesions from Non-Dermoscopic Images Using Deep\nLearning},\n url = {https://arxiv.org/abs/1609.02374v1},\n year = {2016}\n}\n\n", - "citation_id": "1AJUcl1KV" - }, - "doi:10.1101/104869": { - "source": "doi", - "identifer": "10.1101/104869", - "standard_citation": "doi:10.1101/104869", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T21:53:28Z", - "timestamp": 1502402008860 - }, - "posted": { - "date-parts": [ - [ - 2017, - 1, - 31 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2017, - 1, - 31 - ] - ] - }, - "abstract": "Advancements in sequencing technologies have highlighted the role of alternative splicing (AS) in increasing transcriptome complexity. This role of AS, combined with the relation of aberrant splicing to malignant states, motivated two streams of research, experimental and computational. The first involves a myriad of techniques such as RNA-Seq and CLIP-Seq to identify splicing regulators and their putative targets. The second involves probabilistic models, also known as splicing codes, which infer regulatory mechanisms and predict splicing outcome directly from genomic sequence. To date, these models have utilized only expression data. In this work we address two related challenges: Can we improve on previous models for AS outcome prediction and can we integrate additional sources of data to improve predictions for AS regulatory factors. We perform a detailed comparison of two previous modeling approaches, Bayesian and Deep Neural networks, dissecting the confounding effects of datasets and target functions. We then develop a new target function for AS prediction and show that it significantly improves model accuracy. Next, we develop a modeling framework to incorporate CLIP-Seq, knockdown and over-expression experiments, which are inherently noisy and suffer from missing values. Using several datasets involving key splice factors in mouse brain, muscle and heart we demonstrate both the prediction improvements and biological insights offered by our new models. Overall, the framework we propose offers a scalable integrative solution to improve splicing code modeling as vast amounts of relevant genomic data become available.\n\nAvailability: code and data will be available on Github following publication.", - "DOI": "10.1101/104869", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2017, - 2, - 1 - ] - ], - "date-time": "2017-02-01T06:10:13Z", - "timestamp": 1485929413000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Integrative Deep Models for Alternative Splicing", - "prefix": "10.1101", - "author": [ - { - "given": "Anupama", - "family": "Jha", - "affiliation": [] - }, - { - "given": "Matthew R", - "family": "Gazzara", - "affiliation": [] - }, - { - "given": "Yoseph", - "family": "Barash", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/104869", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 2, - 1 - ] - ], - "date-time": "2017-02-01T06:10:26Z", - "timestamp": 1485929426000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 1, - 31 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/104869", - "relation": {}, - "id": "N0HBi8MH" - }, - "citation_id": "N0HBi8MH" - }, - "arxiv:1705.00092": { - "source": "arxiv", - "identifer": "1705.00092", - "standard_citation": "arxiv:1705.00092", - "bibtex": "@article{71c6rs2z,\n abstract = {We present a conditional generative model to learn variation in cell and\nnuclear morphology and the location of subcellular structures from microscopy\nimages. Our model generalizes to a wide range of subcellular localization and\nallows for a probabilistic interpretation of cell and nuclear morphology and\nstructure localization from fluorescence images. We demonstrate the\neffectiveness of our approach by producing photo-realistic cell images using\nour generative model. The conditional nature of the model provides the ability\nto predict the localization of unobserved structures given cell and nuclear\nmorphology.},\n archiveprefix = {arXiv},\n author = {Gregory R. Johnson and Rory M. Donovan-Maiye and Mary M. Maleckar},\n eprint = {1705.00092v1},\n file = {1705.00092v1.pdf},\n month = {Apr},\n primaryclass = {stat.ML},\n title = {Generative Modeling with Conditional Autoencoders: Building an\nIntegrated Cell},\n url = {https://arxiv.org/abs/1705.00092v1},\n year = {2017}\n}\n\n", - "citation_id": "71c6rs2z" - }, - "doi:10.1530/eje-15-0916": { - "source": "doi", - "identifer": "10.1530/eje-15-0916", - "standard_citation": "doi:10.1530/eje-15-0916", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T04:30:38Z", - "timestamp": 1502253038915 - }, - "reference-count": 127, - "publisher": "BioScientifica", - "issue": "5", - "content-domain": { - "domain": [ - "eje-online.org" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2016, - 5 - ] - ] - }, - "DOI": "10.1530/eje-15-0916", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 3, - 11 - ] - ], - "date-time": "2016-03-11T03:29:06Z", - "timestamp": 1457666946000 - }, - "page": "R225-R238", - "update-policy": "http://dx.doi.org/10.1530/eje/crossmarkpolicy", - "source": "Crossref", - "is-referenced-by-count": 6, - "title": "MECHANISMS IN ENDOCRINOLOGY: Alternative splicing: the new frontier in diabetes research", - "prefix": "10.1530", - "volume": "174", - "author": [ - { - "given": "Jonàs", - "family": "Juan-Mateu", - "affiliation": [] - }, - { - "given": "Olatz", - "family": "Villate", - "affiliation": [] - }, - { - "given": "Décio L", - "family": "Eizirik", - "affiliation": [] - } - ], - "member": "416", - "published-online": { - "date-parts": [ - [ - 2015, - 12, - 1 - ] - ] - }, - "container-title": "European Journal of Endocrinology", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T08:06:53Z", - "timestamp": 1498291613000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 12, - 1 - ] - ] - }, - "references-count": 127, - "alternative-id": [ - "10.1530/EJE-15-0916" - ], - "URL": "https://doi.org/10.1530/eje-15-0916", - "relation": {}, - "subject": [ - "Endocrinology, Diabetes and Metabolism", - "Endocrinology", - "General Medicine" - ], - "container-title-short": "Eur J Endocrinol", - "id": "CNz9HwZ3" - }, - "citation_id": "CNz9HwZ3" - }, - "arxiv:1704.01942": { - "source": "arxiv", - "identifer": "1704.01942", - "standard_citation": "arxiv:1704.01942", - "bibtex": "@article{QphVo2P2,\n abstract = {While deep learning models have achieved state-of-the-art accuracies for many\nprediction tasks, understanding these models remains a challenge. Despite the\nrecent interest in developing visual tools to help users interpret deep\nlearning models, the complexity and wide variety of models deployed in\nindustry, and the large-scale datasets that they used, pose unique design\nchallenges that are inadequately addressed by existing work. Through\nparticipatory design sessions with over 15 researchers and engineers at\nFacebook, we have developed, deployed, and iteratively improved ActiVis, an\ninteractive visualization system for interpreting large-scale deep learning\nmodels and results. By tightly integrating multiple coordinated views, such as\na computation graph overview of the model architecture, and a neuron activation\nview for pattern discovery and comparison, users can explore complex deep\nneural network models at both the instance- and subset-level. ActiVis has been\ndeployed on Facebook's machine learning platform. We present case studies with\nFacebook researchers and engineers, and usage scenarios of how ActiVis may work\nwith different models.},\n archiveprefix = {arXiv},\n author = {Minsuk Kahng and Pierre Y. Andrews and Aditya Kalro and Duen Horng Chau},\n eprint = {1704.01942v2},\n file = {1704.01942v2.pdf},\n month = {Apr},\n primaryclass = {cs.HC},\n title = {ActiVis: Visual Exploration of Industry-Scale Deep Neural Network Models},\n url = {https://arxiv.org/abs/1704.01942v2},\n year = {2017}\n}\n\n", - "citation_id": "QphVo2P2" - }, - "doi:10.1128/jb.179.12.3899-3913.1997": { - "source": "doi", - "identifer": "10.1128/jb.179.12.3899-3913.1997", - "standard_citation": "doi:10.1128/jb.179.12.3899-3913.1997", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 20 - ] - ], - "date-time": "2017-09-20T15:02:12Z", - "timestamp": 1505919732632 - }, - "reference-count": 0, - "publisher": "American Society for Microbiology", - "issue": "12", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 1997, - 6 - ] - ] - }, - "abstract": "We compare and contrast genome-wide compositional biases and distributions of short oligonucleotides across 15 diverse prokaryotes that have substantial genomic sequence collections. These include seven complete genomes (Escherichia coli, Haemophilus influenzae, Mycoplasma genitalium, Mycoplasma pneumoniae, Synechocystis sp. strain PCC6803, Methanococcus jannaschii, and Pyrobaculum aerophilum). A key observation concerns the constancy of the dinucleotide relative abundance profiles over multiple 50-kb disjoint contigs within the same genome. (The profile is rhoXY* = fXY*/fX*fY* for all XY, where fX* denotes the frequency of the nucleotide X and fY* denotes the frequency of the dinucleotide XY, both computed from the sequence concatenated with its inverted complementary sequence.) On the basis of this constancy, we refer to the collection [rhoXY*] as the genome signature. We establish that the differences between [rhoXY*] vectors of 50-kb sample contigs of different genomes virtually always exceed the differences between those of the same genomes. Various di- and tetranucleotide biases are identified. In particular, we find that the dinucleotide CpG=CG is underrepresented in many thermophiles (e.g., M. jannaschii, Sulfolobus sp., and M. thermoautotrophicum) but overrepresented in halobacteria. TA is broadly underrepresented in prokaryotes and eukaryotes, but normal counts appear in Sulfolobus and P. aerophilum sequences. More than for any other bacterial genome, palindromic tetranucleotides are underrepresented in H. influenzae. The M. jannaschii sequence is unprecedented in its extreme underrepresentation of CTAG tetranucleotides and in the anomalous distribution of CTAG sites around the genome. Comparative analysis of numbers of long tetranucleotide microsatellites distinguishes H. influenzae. Dinucleotide relative abundance differences between bacterial sequences are compared. For example, in these assessments of differences, the cyanobacteria Synechocystis, Synechococcus, and Anabaena do not form a coherent group and are as far from each other as general gram-negative sequences are from general gram-positive sequences. The difference of M. jannaschii from low-G+C gram-positive proteobacteria is one-half of the difference from gram-negative proteobacteria. Interpretations and hypotheses center on the role of the genome signature in highlighting similarities and dissimilarities across different classes of prokaryotic species, possible mechanisms underlying the genome signature, the form and level of genome compositional flux, the use of the genome signature as a chronometer of molecular phylogeny, and implications with respect to the three putative eubacterial, archaeal, and eukaryote domains of life and to the origin and early evolution of eukaryotes.", - "DOI": "10.1128/jb.179.12.3899-3913.1997", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 11, - 10 - ] - ], - "date-time": "2016-11-10T14:42:18Z", - "timestamp": 1478788938000 - }, - "page": "3899-3913", - "source": "Crossref", - "is-referenced-by-count": 158, - "title": "Compositional biases of bacterial genomes and evolutionary implications.", - "prefix": "10.1128", - "volume": "179", - "author": [ - { - "given": "S", - "family": "Karlin", - "affiliation": [] - }, - { - "given": "J", - "family": "Mrázek", - "affiliation": [] - }, - { - "given": "A M", - "family": "Campbell", - "affiliation": [] - } - ], - "member": "235", - "published-online": { - "date-parts": [ - [ - 1997, - 6, - 1 - ] - ] - }, - "container-title": "Journal of Bacteriology", - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1128/jb.179.12.3899-3913.1997", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 12, - 23 - ] - ], - "date-time": "2016-12-23T13:46:11Z", - "timestamp": 1482500771000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 1997, - 6 - ] - ] - }, - "references-count": 0, - "alternative-id": [ - "10.1128/jb.179.12.3899-3913.1997" - ], - "URL": "https://doi.org/10.1128/jb.179.12.3899-3913.1997", - "relation": {}, - "subject": [ - "Molecular Biology", - "Microbiology" - ], - "container-title-short": "J. Bacteriol.", - "id": "N9NzkOjA" - }, - "citation_id": "N9NzkOjA" - }, - "arxiv:1506.02078": { - "source": "arxiv", - "identifer": "1506.02078", - "standard_citation": "arxiv:1506.02078", - "bibtex": "@article{2cpYveR4,\n abstract = {Recurrent Neural Networks (RNNs), and specifically a variant with Long\nShort-Term Memory (LSTM), are enjoying renewed interest as a result of\nsuccessful applications in a wide range of machine learning problems that\ninvolve sequential data. However, while LSTMs provide exceptional results in\npractice, the source of their performance and their limitations remain rather\npoorly understood. Using character-level language models as an interpretable\ntestbed, we aim to bridge this gap by providing an analysis of their\nrepresentations, predictions and error types. In particular, our experiments\nreveal the existence of interpretable cells that keep track of long-range\ndependencies such as line lengths, quotes and brackets. Moreover, our\ncomparative analysis with finite horizon n-gram models traces the source of the\nLSTM improvements to long-range structural dependencies. Finally, we provide\nanalysis of the remaining errors and suggests areas for further study.},\n archiveprefix = {arXiv},\n author = {Andrej Karpathy and Justin Johnson and Li Fei-Fei},\n eprint = {1506.02078v2},\n file = {1506.02078v2.pdf},\n month = {Jun},\n primaryclass = {cs.LG},\n title = {Visualizing and Understanding Recurrent Networks},\n url = {https://arxiv.org/abs/1506.02078v2},\n year = {2015}\n}\n\n", - "citation_id": "2cpYveR4" - }, - "arxiv:1606.08793": { - "source": "arxiv", - "identifer": "1606.08793", - "standard_citation": "arxiv:1606.08793", - "bibtex": "@article{uP7SgBVd,\n abstract = {Deep learning methods such as multitask neural networks have recently been\napplied to ligand-based virtual screening and other drug discovery\napplications. Using a set of industrial ADMET datasets, we compare neural\nnetworks to standard baseline models and analyze multitask learning effects\nwith both random cross-validation and a more relevant temporal validation\nscheme. We confirm that multitask learning can provide modest benefits over\nsingle-task models and show that smaller datasets tend to benefit more than\nlarger datasets from multitask learning. Additionally, we find that adding\nmassive amounts of side information is not guaranteed to improve performance\nrelative to simpler multitask learning. Our results emphasize that multitask\neffects are highly dataset-dependent, suggesting the use of dataset-specific\nmodels to maximize overall performance.},\n archiveprefix = {arXiv},\n author = {Steven Kearnes and Brian Goldman and Vijay Pande},\n eprint = {1606.08793v3},\n file = {1606.08793v3.pdf},\n month = {Jun},\n primaryclass = {stat.ML},\n title = {Modeling Industrial ADMET Data with Multitask Networks},\n url = {https://arxiv.org/abs/1606.08793v3},\n year = {2016}\n}\n\n", - "citation_id": "uP7SgBVd" - }, - "doi:10.1007/s10822-016-9938-8": { - "source": "doi", - "identifer": "10.1007/s10822-016-9938-8", - "standard_citation": "doi:10.1007/s10822-016-9938-8", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 24 - ] - ], - "date-time": "2017-08-24T13:06:21Z", - "timestamp": 1503579981504 - }, - "reference-count": 42, - "publisher": "Springer Nature", - "issue": "8", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2016, - 8, - 1 - ] - ], - "date-time": "2016-08-01T00:00:00Z", - "timestamp": 1470009600000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/100000002", - "name": "National Institutes of Health", - "doi-asserted-by": "publisher", - "award": [ - "5U19AI109662-02", - "1S10RR02664701" - ] - } - ], - "content-domain": { - "domain": [ - "link.springer.com" - ], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 8 - ] - ] - }, - "DOI": "10.1007/s10822-016-9938-8", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 8, - 24 - ] - ], - "date-time": "2016-08-24T08:46:36Z", - "timestamp": 1472028396000 - }, - "page": "595-608", - "update-policy": "http://dx.doi.org/10.1007/springer_crossmark_policy", - "source": "Crossref", - "is-referenced-by-count": 13, - "title": "Molecular graph convolutions: moving beyond fingerprints", - "prefix": "10.1007", - "volume": "30", - "author": [ - { - "ORCID": "http://orcid.org/0000-0003-4579-4388", - "authenticated-orcid": false, - "given": "Steven", - "family": "Kearnes", - "affiliation": [] - }, - { - "given": "Kevin", - "family": "McCloskey", - "affiliation": [] - }, - { - "given": "Marc", - "family": "Berndl", - "affiliation": [] - }, - { - "given": "Vijay", - "family": "Pande", - "affiliation": [] - }, - { - "given": "Patrick", - "family": "Riley", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2016, - 8, - 24 - ] - ] - }, - "container-title": "Journal of Computer-Aided Molecular Design", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1007/s10822-016-9938-8.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://link.springer.com/article/10.1007/s10822-016-9938-8/fulltext.html", - "content-type": "text/html", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T21:17:00Z", - "timestamp": 1498339020000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 8 - ] - ] - }, - "references-count": 42, - "alternative-id": [ - "9938" - ], - "URL": "https://doi.org/10.1007/s10822-016-9938-8", - "relation": { - "cites": [] - }, - "subject": [ - "Physical and Theoretical Chemistry", - "Drug Discovery", - "Computer Science Applications" - ], - "container-title-short": "J Comput Aided Mol Des", - "id": "145os4Y6t" - }, - "citation_id": "145os4Y6t" - }, - "url:https://github.com/fchollet/keras": { - "source": "url", - "identifer": "https://github.com/fchollet/keras", - "standard_citation": "url:https://github.com/fchollet/keras", - "citeproc": { - "URL": "https://github.com/fchollet/keras", - "title": "fchollet/keras", - "container-title": "GitHub", - "issued": { - "date-parts": [ - [ - 2017 - ] - ] - }, - "type": "webpage", - "id": "FwEK0msb" - }, - "citation_id": "FwEK0msb" - }, - "arxiv:1611.07270": { - "source": "arxiv", - "identifer": "1611.07270", - "standard_citation": "arxiv:1611.07270", - "bibtex": "@article{b1sc0cgP,\n abstract = {Understanding neural networks is becoming increasingly important. Over the\nlast few years different types of visualisation and explanation methods have\nbeen proposed. However, none of them explicitly considered the behaviour in the\npresence of noise and distracting elements. In this work, we will show how\nnoise and distracting dimensions can influence the result of an explanation\nmodel. This gives a new theoretical insights to aid selection of the most\nappropriate explanation model within the deep-Taylor decomposition framework.},\n archiveprefix = {arXiv},\n author = {Pieter-Jan Kindermans and Kristof Schütt and Klaus-Robert Müller and Sven Dähne},\n eprint = {1611.07270v1},\n file = {1611.07270v1.pdf},\n month = {Dec},\n primaryclass = {stat.ML},\n title = {Investigating the influence of noise and distractors on the\ninterpretation of neural networks},\n url = {https://arxiv.org/abs/1611.07270v1},\n year = {2016}\n}\n\n", - "citation_id": "b1sc0cgP" - }, - "doi:10.1111/j.1574-6976.2010.00251.x": { - "source": "doi", - "identifer": "10.1111/j.1574-6976.2010.00251.x", - "standard_citation": "doi:10.1111/j.1574-6976.2010.00251.x", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 4 - ] - ], - "date-time": "2017-09-04T13:42:08Z", - "timestamp": 1504532528657 - }, - "reference-count": 57, - "publisher": "Oxford University Press (OUP)", - "issue": "2", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2011, - 3 - ] - ] - }, - "DOI": "10.1111/j.1574-6976.2010.00251.x", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2010, - 9, - 21 - ] - ], - "date-time": "2010-09-21T13:07:26Z", - "timestamp": 1285074446000 - }, - "page": "343-359", - "source": "Crossref", - "is-referenced-by-count": 117, - "title": "Supervised classification of human microbiota", - "prefix": "10.1093", - "volume": "35", - "author": [ - { - "given": "Dan", - "family": "Knights", - "affiliation": [] - }, - { - "given": "Elizabeth K.", - "family": "Costello", - "affiliation": [] - }, - { - "given": "Rob", - "family": "Knight", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2011, - 3, - 1 - ] - ] - }, - "container-title": "FEMS Microbiology Reviews", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/femsre/article-pdf/35/2/343/19424511/35-2-343.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 23 - ] - ], - "date-time": "2017-08-23T02:58:24Z", - "timestamp": 1503457104000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2011, - 3 - ] - ] - }, - "references-count": 57, - "alternative-id": [ - "10.1111/j.1574-6976.2010.00251.x" - ], - "URL": "https://doi.org/10.1111/j.1574-6976.2010.00251.x", - "relation": {}, - "container-title-short": "FEMS Microbiol Rev", - "id": "aI9g2UOc" - }, - "citation_id": "aI9g2UOc" - }, - "doi:10.1101/052118": { - "source": "doi", - "identifer": "10.1101/052118", - "standard_citation": "doi:10.1101/052118", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T09:07:40Z", - "timestamp": 1502269660096 - }, - "posted": { - "date-parts": [ - [ - 2016, - 5, - 7 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2017, - 1, - 27 - ] - ] - }, - "abstract": "Motivation: Chromatin immunoprecipitation sequencing (ChIP-seq) experiments are commonly used to obtain genome-wide profiles of histone modifications associated with different types of functional genomic elements. However, the quality of histone ChIP-seq data is affected by a myriad of experimental parameters such as the amount of input DNA, antibody specificity, ChIP enrichment, and sequencing depth. Making accurate inferences from chromatin profiling experiments that involve diverse experimental parameters is challenging. Results: We introduce a convolutional denoising algorithm, Coda, that uses convolutional neural networks to learn a mapping from suboptimal to high-quality histone ChIP-seq data. This overcomes various sources of noise and variability, substantially enhancing and recovering signal when applied to low-quality chromatin profiling datasets across individuals, cell types, and species. Our method has the potential to improve data quality at reduced costs. More broadly, this approach -- using a high-dimensional discriminative model to encode a generative noise process -- is generally applicable to other biological domains where it is easy to generate noisy data but difficult to analytically characterize the noise or underlying data distribution. Availability: https://github.com/kundajelab/coda", - "DOI": "10.1101/052118", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2016, - 5, - 8 - ] - ], - "date-time": "2016-05-08T05:15:05Z", - "timestamp": 1462684505000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Denoising genome-wide histone ChIP-seq with convolutional neural networks", - "prefix": "10.1101", - "author": [ - { - "given": "Pang Wei", - "family": "Koh", - "affiliation": [] - }, - { - "given": "Emma", - "family": "Pierson", - "affiliation": [] - }, - { - "given": "Anshul", - "family": "Kundaje", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/052118", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 1, - 28 - ] - ], - "date-time": "2017-01-28T06:10:24Z", - "timestamp": 1485583824000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 5, - 7 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/052118", - "relation": {}, - "id": "XimuXZlz" - }, - "citation_id": "XimuXZlz" - }, - "arxiv:1703.04730": { - "source": "arxiv", - "identifer": "1703.04730", - "standard_citation": "arxiv:1703.04730", - "bibtex": "@article{69wxD9y,\n abstract = {How can we explain the predictions of a black-box model? In this paper, we\nuse influence functions -- a classic technique from robust statistics -- to\ntrace a model's prediction through the learning algorithm and back to its\ntraining data, thereby identifying training points most responsible for a given\nprediction. To scale up influence functions to modern machine learning\nsettings, we develop a simple, efficient implementation that requires only\noracle access to gradients and Hessian-vector products. We show that even on\nnon-convex and non-differentiable models where the theory breaks down, approximations to influence functions can still provide valuable information.\nOn linear models and convolutional neural networks, we demonstrate that\ninfluence functions are useful for multiple purposes: understanding model\nbehavior, debugging models, detecting dataset errors, and even creating\nvisually-indistinguishable training-set attacks.},\n archiveprefix = {arXiv},\n author = {Pang Wei Koh and Percy Liang},\n eprint = {1703.04730v2},\n file = {1703.04730v2.pdf},\n month = {Mar},\n primaryclass = {stat.ML},\n title = {Understanding Black-box Predictions via Influence Functions},\n url = {https://arxiv.org/abs/1703.04730v2},\n year = {2017}\n}\n\n", - "citation_id": "69wxD9y" - }, - "doi:10.1016/j.media.2016.07.007": { - "source": "doi", - "identifer": "10.1016/j.media.2016.07.007", - "standard_citation": "doi:10.1016/j.media.2016.07.007", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 20 - ] - ], - "date-time": "2017-09-20T16:22:11Z", - "timestamp": 1505924531582 - }, - "reference-count": 65, - "publisher": "Elsevier BV", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2017, - 1, - 1 - ] - ], - "date-time": "2017-01-01T00:00:00Z", - "timestamp": 1483228800000 - }, - "delay-in-days": 0, - "content-version": "tdm" - } - ], - "content-domain": { - "domain": [ - "medicalimageanalysisjournal.com", - "elsevier.com", - "sciencedirect.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2017, - 1 - ] - ] - }, - "DOI": "10.1016/j.media.2016.07.007", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 8, - 3 - ] - ], - "date-time": "2016-08-03T16:52:24Z", - "timestamp": 1470243144000 - }, - "page": "303-312", - "update-policy": "http://dx.doi.org/10.1016/elsevier_cm_policy", - "source": "Crossref", - "is-referenced-by-count": 13, - "title": "Large scale deep learning for computer aided detection of mammographic lesions", - "prefix": "10.1016", - "volume": "35", - "author": [ - { - "ORCID": "http://orcid.org/0000-0001-5565-5029", - "authenticated-orcid": false, - "given": "Thijs", - "family": "Kooi", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0003-1554-1291", - "authenticated-orcid": false, - "given": "Geert", - "family": "Litjens", - "affiliation": [] - }, - { - "given": "Bram", - "family": "van Ginneken", - "affiliation": [] - }, - { - "given": "Albert", - "family": "Gubern-Mérida", - "affiliation": [] - }, - { - "given": "Clara I.", - "family": "Sánchez", - "affiliation": [] - }, - { - "given": "Ritse", - "family": "Mann", - "affiliation": [] - }, - { - "given": "Ard", - "family": "den Heeten", - "affiliation": [] - }, - { - "given": "Nico", - "family": "Karssemeijer", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Medical Image Analysis", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S1361841516301244?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S1361841516301244?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 7, - 24 - ] - ], - "date-time": "2017-07-24T18:15:22Z", - "timestamp": 1500920122000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 1 - ] - ] - }, - "references-count": 65, - "alternative-id": [ - "S1361841516301244" - ], - "URL": "https://doi.org/10.1016/j.media.2016.07.007", - "relation": {}, - "subject": [ - "Radiological and Ultrasound Technology", - "Health Informatics", - "Radiology Nuclear Medicine and imaging", - "Computer Vision and Pattern Recognition", - "Computer Graphics and Computer-Aided Design" - ], - "container-title-short": "Medical Image Analysis", - "assertion": [ - { - "value": "Elsevier", - "name": "publisher", - "label": "This article is maintained by" - }, - { - "value": "Large scale deep learning for computer aided detection of mammographic lesions", - "name": "articletitle", - "label": "Article Title" - }, - { - "value": "Medical Image Analysis", - "name": "journaltitle", - "label": "Journal Title" - }, - { - "value": "http://dx.doi.org/10.1016/j.media.2016.07.007", - "name": "articlelink", - "label": "CrossRef DOI link to publisher maintained version" - }, - { - "value": "article", - "name": "content_type", - "label": "Content Type" - }, - { - "value": "© 2016 Elsevier B.V. All rights reserved.", - "name": "copyright", - "label": "Copyright" - } - ], - "id": "18cZbigDD" - }, - "citation_id": "18cZbigDD" - }, - "doi:10.15252/msb.20177551": { - "source": "doi", - "identifer": "10.15252/msb.20177551", - "standard_citation": "doi:10.15252/msb.20177551", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 31 - ] - ], - "date-time": "2017-08-31T16:02:17Z", - "timestamp": 1504195337883 - }, - "reference-count": 0, - "publisher": "EMBO", - "issue": "4", - "funder": [ - { - "DOI": "10.13039/501100001805", - "name": "Canada Foundation for Innovation", - "doi-asserted-by": "publisher", - "award": [ - "21475" - ] - }, - { - "DOI": "10.13039/501100003400", - "name": "Ministry of Research and Innovation", - "doi-asserted-by": "publisher", - "award": [ - "21475" - ] - }, - { - "DOI": "10.13039/501100000024", - "name": "Canadian Institutes of Health Research", - "doi-asserted-by": "publisher", - "award": [ - "FDN‐143264", - "FDN‐143265" - ] - }, - { - "DOI": "10.13039/100000002", - "name": "National Institutes of Health", - "doi-asserted-by": "publisher", - "award": [ - "R01HG005853" - ] - }, - { - "DOI": "10.13039/501100007224", - "name": "Connaught Fund", - "doi-asserted-by": "publisher", - "award": [ - "GCDF:2013‐14" - ] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2017, - 4 - ] - ] - }, - "DOI": "10.15252/msb.20177551", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2017, - 4, - 19 - ] - ], - "date-time": "2017-04-19T00:10:54Z", - "timestamp": 1492560654000 - }, - "page": "924", - "source": "Crossref", - "is-referenced-by-count": 4, - "title": "Automated analysis of high‐content microscopy data with deep learning", - "prefix": "10.15252", - "volume": "13", - "author": [ - { - "ORCID": "http://orcid.org/0000-0002-6328-9492", - "authenticated-orcid": false, - "given": "Oren Z", - "family": "Kraus", - "affiliation": [] - }, - { - "given": "Ben T", - "family": "Grys", - "affiliation": [] - }, - { - "given": "Jimmy", - "family": "Ba", - "affiliation": [] - }, - { - "given": "Yolanda", - "family": "Chong", - "affiliation": [] - }, - { - "given": "Brendan J", - "family": "Frey", - "affiliation": [] - }, - { - "given": "Charles", - "family": "Boone", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0001-6427-6493", - "authenticated-orcid": false, - "given": "Brenda J", - "family": "Andrews", - "affiliation": [] - } - ], - "member": "79", - "published-online": { - "date-parts": [ - [ - 2017, - 4, - 18 - ] - ] - }, - "container-title": "Molecular Systems Biology", - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.15252/msb.20177551", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 4, - 19 - ] - ], - "date-time": "2017-04-19T00:11:02Z", - "timestamp": 1492560662000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 4 - ] - ] - }, - "references-count": 0, - "alternative-id": [ - "10.15252/msb.20177551" - ], - "URL": "https://doi.org/10.15252/msb.20177551", - "relation": {}, - "subject": [ - "General Biochemistry, Genetics and Molecular Biology", - "Computational Theory and Mathematics", - "General Immunology and Microbiology", - "Applied Mathematics", - "General Agricultural and Biological Sciences", - "Information Systems" - ], - "container-title-short": "Mol Syst Biol", - "id": "DcnNfASG" - }, - "citation_id": "DcnNfASG" - }, - "url:https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf": { - "source": "url", - "identifer": "https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf", - "standard_citation": "url:https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf", - "citeproc": { - "type": "paper-conference", - "title": "ImageNet Classification with Deep Convolutional Neural Networks", - "container-title": "Proceedings of the 25th International Conference on Neural Information Processing Systems", - "collection-title": "NIPS'12", - "publisher": "Curran Associates Inc.", - "publisher-place": "USA", - "page": "1097–1105", - "source": "ACM Digital Library", - "event-place": "USA", - "abstract": "We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 different classes. On the test data, we achieved top-1 and top-5 error rates of 37.5% and 17.0% which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implementation of the convolution operation. To reduce overriding in the fully-connected layers we employed a recently-developed regularization method called \"dropout\" that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3%, compared to 26.2% achieved by the second-best entry.", - "URL": "http://dl.acm.org/citation.cfm?id=2999134.2999257", - "author": [ - { - "family": "Krizhevsky", - "given": "Alex" - }, - { - "family": "Sutskever", - "given": "Ilya" - }, - { - "family": "Hinton", - "given": "Geoffrey E." - } - ], - "issued": { - "date-parts": [ - [ - "2012" - ] - ] - }, - "accessed": { - "date-parts": [ - [ - "2017", - 5, - 23 - ] - ] - }, - "id": "CCS5KSIM" - }, - "citation_id": "CCS5KSIM" - }, - "arxiv:1404.5997": { - "source": "arxiv", - "identifer": "1404.5997", - "standard_citation": "arxiv:1404.5997", - "bibtex": "@article{ZSVsnPVO,\n abstract = {I present a new way to parallelize the training of convolutional neural\nnetworks across multiple GPUs. The method scales significantly better than all\nalternatives when applied to modern convolutional neural networks.},\n archiveprefix = {arXiv},\n author = {Alex Krizhevsky},\n eprint = {1404.5997v2},\n file = {1404.5997v2.pdf},\n month = {Apr},\n primaryclass = {cs.NE},\n title = {One weird trick for parallelizing convolutional neural networks},\n url = {https://arxiv.org/abs/1404.5997v2},\n year = {2014}\n}\n\n", - "citation_id": "ZSVsnPVO" - }, - "arxiv:1602.04283": { - "source": "arxiv", - "identifer": "1602.04283", - "standard_citation": "arxiv:1602.04283", - "bibtex": "@article{9NKsJjSw,\n abstract = {The rapid growth of data size and accessibility in recent years has\ninstigated a shift of philosophy in algorithm design for artificial\nintelligence. Instead of engineering algorithms by hand, the ability to learn\ncomposable systems automatically from massive amounts of data has led to\nground-breaking performance in important domains such as computer vision, speech recognition, and natural language processing. The most popular class of\ntechniques used in these domains is called deep learning, and is seeing\nsignificant attention from industry. However, these models require incredible\namounts of data and compute power to train, and are limited by the need for\nbetter hardware acceleration to accommodate scaling beyond current data and\nmodel sizes. While the current solution has been to use clusters of graphics\nprocessing units (GPU) as general purpose processors (GPGPU), the use of field\nprogrammable gate arrays (FPGA) provide an interesting alternative. Current\ntrends in design tools for FPGAs have made them more compatible with the\nhigh-level software practices typically practiced in the deep learning\ncommunity, making FPGAs more accessible to those who build and deploy models.\nSince FPGA architectures are flexible, this could also allow researchers the\nability to explore model-level optimizations beyond what is possible on fixed\narchitectures such as GPUs. As well, FPGAs tend to provide high performance per\nwatt of power consumption, which is of particular importance for application\nscientists interested in large scale server-based deployment or\nresource-limited embedded applications. This review takes a look at deep\nlearning and FPGAs from a hardware acceleration perspective, identifying trends\nand innovations that make these technologies a natural fit, and motivates a\ndiscussion on how FPGAs may best serve the needs of the deep learning community\nmoving forward.},\n archiveprefix = {arXiv},\n author = {Griffin Lacey and Graham W. Taylor and Shawki Areibi},\n eprint = {1602.04283v1},\n file = {1602.04283v1.pdf},\n month = {Feb},\n primaryclass = {cs.DC},\n title = {Deep Learning on FPGAs: Past, Present, and Future},\n url = {https://arxiv.org/abs/1602.04283v1},\n year = {2016}\n}\n\n", - "citation_id": "9NKsJjSw" - }, - "arxiv:1608.03644": { - "source": "arxiv", - "identifer": "1608.03644", - "standard_citation": "arxiv:1608.03644", - "bibtex": "@article{Dwi2eAvT,\n abstract = {Deep neural network (DNN) models have recently obtained state-of-the-art\nprediction accuracy for the transcription factor binding (TFBS) site\nclassification task. However, it remains unclear how these approaches identify\nmeaningful DNA sequence signals and give insights as to why TFs bind to certain\nlocations. In this paper, we propose a toolkit called the Deep Motif Dashboard\n(DeMo Dashboard) which provides a suite of visualization strategies to extract\nmotifs, or sequence patterns from deep neural network models for TFBS\nclassification. We demonstrate how to visualize and understand three important\nDNN models: convolutional, recurrent, and convolutional-recurrent networks. Our\nfirst visualization method is finding a test sequence's saliency map which uses\nfirst-order derivatives to describe the importance of each nucleotide in making\nthe final prediction. Second, considering recurrent models make predictions in\na temporal manner (from one end of a TFBS sequence to the other), we introduce\ntemporal output scores, indicating the prediction score of a model over time\nfor a sequential input. Lastly, a class-specific visualization strategy finds\nthe optimal input sequence for a given TFBS positive class via stochastic\ngradient optimization. Our experimental results indicate that a\nconvolutional-recurrent architecture performs the best among the three\narchitectures. The visualization techniques indicate that CNN-RNN makes\npredictions by modeling both motifs as well as dependencies among them.},\n archiveprefix = {arXiv},\n author = {Jack Lanchantin and Ritambhara Singh and Beilun Wang and Yanjun Qi},\n eprint = {1608.03644v4},\n file = {1608.03644v4.pdf},\n month = {Aug},\n primaryclass = {cs.LG},\n title = {Deep Motif Dashboard: Visualizing and Understanding Genomic Sequences\nUsing Deep Neural Networks},\n url = {https://arxiv.org/abs/1608.03644v4},\n year = {2016}\n}\n\n", - "citation_id": "Dwi2eAvT" - }, - "arxiv:1603.09123v2": { - "source": "arxiv", - "identifer": "1603.09123v2", - "standard_citation": "arxiv:1603.09123v2", - "bibtex": "@article{1GwC1ll6h,\n abstract = {MicroRNAs (miRNAs) are short sequences of ribonucleic acids that control the\nexpression of target messenger RNAs (mRNAs) by binding them. Robust prediction\nof miRNA-mRNA pairs is of utmost importance in deciphering gene regulations but\nhas been challenging because of high false positive rates, despite a deluge of\ncomputational tools that normally require laborious manual feature extraction.\nThis paper presents an end-to-end machine learning framework for miRNA target\nprediction. Leveraged by deep recurrent neural networks-based auto-encoding and\nsequence-sequence interaction learning, our approach not only delivers an\nunprecedented level of accuracy but also eliminates the need for manual feature\nextraction. The performance gap between the proposed method and existing\nalternatives is substantial (over 25% increase in F-measure), and deepTarget\ndelivers a quantum leap in the long-standing challenge of robust miRNA target\nprediction.},\n archiveprefix = {arXiv},\n author = {Byunghan Lee and Junghwan Baek and Seunghyun Park and Sungroh Yoon},\n eprint = {1603.09123v2},\n file = {1603.09123v2.pdf},\n month = {Mar},\n primaryclass = {cs.LG},\n title = {deepTarget: End-to-end Learning Framework for microRNA Target Prediction\nusing Deep Recurrent Neural Networks},\n url = {https://arxiv.org/abs/1603.09123v2},\n year = {2016}\n}\n\n", - "citation_id": "1GwC1ll6h" - }, - "doi:10.1101/094276": { - "source": "doi", - "identifer": "10.1101/094276", - "standard_citation": "doi:10.1101/094276", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T05:20:36Z", - "timestamp": 1502342436479 - }, - "posted": { - "date-parts": [ - [ - 2016, - 12, - 14 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2016, - 12, - 14 - ] - ] - }, - "abstract": "Objective: The advent of Electronic Medical Records (EMR) with large electronic imaging databases along with advances in deep neural networks with machine learning has provided a unique opportunity to achieve milestones in automated image analysis. Optical coherence tomography (OCT) is the most commonly obtained imaging modality in ophthalmology and represents a dense and rich dataset when combined with labels derived from the EMR. We sought to determine if deep learning could be utilized to distinguish normal OCT images from images from patients with Age-related Macular Degeneration (AMD). Design: EMR and OCT database study Subjects: Normal and AMD patients who had a macular OCT. Methods: Automated extraction of an OCT imaging database was performed and linked to clinical endpoints from the EMR. OCT macula scans were obtained by Heidelberg Spectralis, and each OCT scan was linked to EMR clinical endpoints extracted from EPIC. The central 11 images were selected from each OCT scan of two cohorts of patients: normal and AMD. Cross-validation was performed using a random subset of patients. Receiver operator curves (ROC) were constructed at an independent image level, macular OCT level, and patient level. Main outcome measure: Area under the ROC. Results: Of a recent extraction of 2.6 million OCT images linked to clinical datapoints from the EMR, 52,690 normal macular OCT images and 48,312 AMD macular OCT images were selected. A deep neural network was trained to categorize images as either normal or AMD. At the image level, we achieved an area under the ROC of 92.78% with an accuracy of 87.63%. At the macula level, we achieved an area under the ROC of 93.83% with an accuracy of 88.98%. At a patient level, we achieved an area under the ROC of 97.45% with an accuracy of 93.45%. Peak sensitivity and specificity with optimal cutoffs were 92.64% and 93.69% respectively. Conclusions: Deep learning techniques achieve high accuracy and is effective as a new image classification technique. These findings have important implications in utilizing OCT in automated screening and the development of computer aided diagnosis tools in the future.", - "DOI": "10.1101/094276", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2016, - 12, - 15 - ] - ], - "date-time": "2016-12-15T06:10:40Z", - "timestamp": 1481782240000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Deep learning is effective for the classification of OCT images of normal versus Age-related Macular Degeneration", - "prefix": "10.1101", - "author": [ - { - "ORCID": "http://orcid.org/0000-0003-1994-7213", - "authenticated-orcid": false, - "given": "Cecilia S", - "family": "Lee", - "affiliation": [] - }, - { - "given": "Doug M", - "family": "Baughman", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-7452-1648", - "authenticated-orcid": false, - "given": "Aaron Y", - "family": "Lee", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/094276", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 1, - 4 - ] - ], - "date-time": "2017-01-04T06:17:57Z", - "timestamp": 1483510677000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 12, - 14 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/094276", - "relation": {}, - "id": "SxsZyrVM" - }, - "citation_id": "SxsZyrVM" - }, - "arxiv:1606.04155": { - "source": "arxiv", - "identifer": "1606.04155", - "standard_citation": "arxiv:1606.04155", - "bibtex": "@article{ZUCVI5eU,\n abstract = {Prediction without justification has limited applicability. As a remedy, we\nlearn to extract pieces of input text as justifications -- rationales -- that\nare tailored to be short and coherent, yet sufficient for making the same\nprediction. Our approach combines two modular components, generator and\nencoder, which are trained to operate well together. The generator specifies a\ndistribution over text fragments as candidate rationales and these are passed\nthrough the encoder for prediction. Rationales are never given during training.\nInstead, the model is regularized by desiderata for rationales. We evaluate the\napproach on multi-aspect sentiment analysis against manually annotated test\ncases. Our approach outperforms attention-based baseline by a significant\nmargin. We also successfully illustrate the method on the question retrieval\ntask.},\n archiveprefix = {arXiv},\n author = {Tao Lei and Regina Barzilay and Tommi Jaakkola},\n eprint = {1606.04155v2},\n file = {1606.04155v2.pdf},\n month = {Jun},\n primaryclass = {cs.CL},\n title = {Rationalizing Neural Predictions},\n url = {https://arxiv.org/abs/1606.04155v2},\n year = {2016}\n}\n\n", - "citation_id": "ZUCVI5eU" - }, - "doi:10.1101/084210": { - "source": "doi", - "identifer": "10.1101/084210", - "standard_citation": "doi:10.1101/084210", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T02:17:43Z", - "timestamp": 1502331463361 - }, - "posted": { - "date-parts": [ - [ - 2016, - 10, - 28 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2017, - 8, - 2 - ] - ] - }, - "abstract": "Deep learning (DL) has revolutionized the field of computer vision and image processing. In medical imaging, algorithmic solutions based on DL have been shown to achieve high performance on tasks that previously required medical experts. However, DL-based solutions for disease detection have been proposed without methods to quantify and control their uncertainty in a decision. In contrast, a physician knows whether she is uncertain about a case and will consult more experienced colleagues if needed. Here we evaluate the uncertainty of DL in medical diagnostics based on a recent theoretical insight on the link between dropout networks and approximate Bayesian inference. Using the example of detecting diabetic retinopathy (DR) from fundus photographs, we show that uncertainty informed decision referral improves diagnostic performance. Experiments across different networks, tasks and datasets showed robust generalization. Depending on network capacity and task/dataset difficulty, we surpass 85% sensitivity and 80% specificity as recommended by the NHS when referring 0%-20% of the most uncertain decisions for further inspection. We analyse causes of uncertainty by relating intuitions from 2D visualizations to the high-dimensional image space. While uncertainty is sensitive to clinically relevant cases, sensitivity to unfamiliar data samples is task dependent, but can be rendered more robust.", - "DOI": "10.1101/084210", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2016, - 10, - 29 - ] - ], - "date-time": "2016-10-29T05:12:12Z", - "timestamp": 1477717932000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Leveraging uncertainty information from deep neural networks for disease detection", - "prefix": "10.1101", - "author": [ - { - "given": "Christian", - "family": "Leibig", - "affiliation": [] - }, - { - "given": "Vaneeda", - "family": "Allken", - "affiliation": [] - }, - { - "given": "Philipp", - "family": "Berens", - "affiliation": [] - }, - { - "given": "Siegfried", - "family": "Wahl", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/084210", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 3 - ] - ], - "date-time": "2017-08-03T05:10:43Z", - "timestamp": 1501737043000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 10, - 28 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/084210", - "relation": {}, - "id": "14Ovc5nPg" - }, - "citation_id": "14Ovc5nPg" - }, - "doi:10.1145/2623330.2623612": { - "source": "doi", - "identifer": "10.1145/2623330.2623612", - "standard_citation": "doi:10.1145/2623330.2623612", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 15 - ] - ], - "date-time": "2017-09-15T19:22:11Z", - "timestamp": 1505503331318 - }, - "publisher-location": "New York, New York, USA", - "reference-count": 0, - "publisher": "ACM Press", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2014 - ] - ] - }, - "DOI": "10.1145/2623330.2623612", - "type": "paper-conference", - "created": { - "date-parts": [ - [ - 2014, - 8, - 22 - ] - ], - "date-time": "2014-08-22T19:38:46Z", - "timestamp": 1408736326000 - }, - "source": "Crossref", - "is-referenced-by-count": 21, - "title": "Efficient mini-batch training for stochastic optimization", - "prefix": "10.1145", - "author": [ - { - "given": "Mu", - "family": "Li", - "affiliation": [] - }, - { - "given": "Tong", - "family": "Zhang", - "affiliation": [] - }, - { - "given": "Yuqiang", - "family": "Chen", - "affiliation": [] - }, - { - "given": "Alexander J.", - "family": "Smola", - "affiliation": [] - } - ], - "member": "320", - "container-title": "Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining - KDD '14", - "original-title": [], - "link": [ - { - "URL": "http://dl.acm.org/ft_gateway.cfm?id=2623612&ftid=1494890&dwn=1", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 12, - 7 - ] - ], - "date-time": "2016-12-07T02:06:13Z", - "timestamp": 1481076373000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1145/2623330.2623612", - "relation": {}, - "id": "fNkl8HFz" - }, - "citation_id": "fNkl8HFz" - }, - "doi:10.1126/science.aad9417": { - "source": "doi", - "identifer": "10.1126/science.aad9417", - "standard_citation": "doi:10.1126/science.aad9417", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 2 - ] - ], - "date-time": "2017-10-02T17:42:12Z", - "timestamp": 1506966132073 - }, - "reference-count": 50, - "publisher": "American Association for the Advancement of Science (AAAS)", - "issue": "6285", - "license": [ - { - "URL": "http://www.sciencemag.org/about/science-licenses-journal-article-reuse", - "start": { - "date-parts": [ - [ - 2017, - 4, - 29 - ] - ], - "date-time": "2017-04-29T00:00:00Z", - "timestamp": 1493424000000 - }, - "delay-in-days": 366, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/100000002", - "name": "National Institutes of Health", - "doi-asserted-by": "publisher", - "award": [ - "R01MH084703", - "U54CA149145", - "RO1MH101825", - "U01HG007036" - ] - }, - { - "DOI": "10.13039/100000011", - "name": "Howard Hughes Medical Institute", - "doi-asserted-by": "publisher", - "award": [] - }, - { - "name": "Center for Computational, Evolutionary and Human Genomics Fellowship", - "award": [] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 4, - 29 - ] - ] - }, - "DOI": "10.1126/science.aad9417", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 4, - 28 - ] - ], - "date-time": "2016-04-28T19:09:18Z", - "timestamp": 1461870558000 - }, - "page": "600-604", - "source": "Crossref", - "is-referenced-by-count": 41, - "title": "RNA splicing is a primary link between genetic variation and disease", - "prefix": "10.1126", - "volume": "352", - "author": [ - { - "given": "Y. I.", - "family": "Li", - "affiliation": [] - }, - { - "given": "B.", - "family": "van de Geijn", - "affiliation": [] - }, - { - "given": "A.", - "family": "Raj", - "affiliation": [] - }, - { - "given": "D. A.", - "family": "Knowles", - "affiliation": [] - }, - { - "given": "A. A.", - "family": "Petti", - "affiliation": [] - }, - { - "given": "D.", - "family": "Golan", - "affiliation": [] - }, - { - "given": "Y.", - "family": "Gilad", - "affiliation": [] - }, - { - "given": "J. K.", - "family": "Pritchard", - "affiliation": [] - } - ], - "member": "221", - "published-online": { - "date-parts": [ - [ - 2016, - 4, - 28 - ] - ] - }, - "container-title": "Science", - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1126/science.aad9417", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T12:02:14Z", - "timestamp": 1498305734000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 4, - 28 - ] - ] - }, - "references-count": 50, - "URL": "https://doi.org/10.1126/science.aad9417", - "relation": { - "cites": [] - }, - "subject": [ - "Multidisciplinary" - ], - "container-title-short": "Science", - "id": "b6p6wxpC" - }, - "citation_id": "b6p6wxpC" - }, - "doi:10.1109/tcbb.2014.2377729": { - "source": "doi", - "identifer": "10.1109/tcbb.2014.2377729", - "standard_citation": "doi:10.1109/tcbb.2014.2377729", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T07:22:12Z", - "timestamp": 1507015332885 - }, - "reference-count": 30, - "publisher": "Institute of Electrical and Electronics Engineers (IEEE)", - "issue": "4", - "funder": [ - { - "DOI": "10.13039/501100001809", - "name": "National Natural Science Foundation of China", - "doi-asserted-by": "publisher", - "award": [ - "61361136003", - "61472205", - "61033001" - ] - }, - { - "name": "National Basic Research Program of China", - "award": [ - "2011CBA00301", - "2011CBA00300" - ] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015, - 7, - 1 - ] - ] - }, - "DOI": "10.1109/tcbb.2014.2377729", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 12, - 6 - ] - ], - "date-time": "2014-12-06T00:20:03Z", - "timestamp": 1417825203000 - }, - "page": "928-937", - "source": "Crossref", - "is-referenced-by-count": 13, - "title": "Integrative Data Analysis of Multi-Platform Cancer Data with a Multimodal Deep Learning Approach", - "prefix": "10.1109", - "volume": "12", - "author": [ - { - "given": "Muxuan", - "family": "Liang", - "affiliation": [] - }, - { - "given": "Zhizhong", - "family": "Li", - "affiliation": [] - }, - { - "given": "Ting", - "family": "Chen", - "affiliation": [] - }, - { - "given": "Jianyang", - "family": "Zeng", - "affiliation": [] - } - ], - "member": "263", - "container-title": "IEEE/ACM Transactions on Computational Biology and Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://xplorestaging.ieee.org/ielx7/8857/7180481/06977954.pdf?arnumber=6977954", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 23 - ] - ], - "date-time": "2017-06-23T01:36:17Z", - "timestamp": 1498181777000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 7, - 1 - ] - ] - }, - "references-count": 30, - "URL": "https://doi.org/10.1109/tcbb.2014.2377729", - "relation": {}, - "subject": [ - "Biotechnology", - "Genetics", - "Applied Mathematics" - ], - "container-title-short": "IEEE/ACM Trans. Comput. Biol. and Bioinf.", - "id": "1EtavGKI4" - }, - "citation_id": "1EtavGKI4" - }, - "arxiv:1510.07641": { - "source": "arxiv", - "identifer": "1510.07641", - "standard_citation": "arxiv:1510.07641", - "bibtex": "@article{glyI7H6F,\n abstract = {We present a novel application of LSTM recurrent neural networks to\nmultilabel classification of diagnoses given variable-length time series of\nclinical measurements. Our method outperforms a strong baseline on a variety of\nmetrics.},\n archiveprefix = {arXiv},\n author = {Zachary C. Lipton and David C. Kale and Randall C. Wetzel},\n eprint = {1510.07641v2},\n file = {1510.07641v2.pdf},\n month = {Nov},\n primaryclass = {cs.LG},\n title = {Phenotyping of Clinical Time Series with LSTM Recurrent Neural Networks},\n url = {https://arxiv.org/abs/1510.07641v2},\n year = {2015}\n}\n\n", - "citation_id": "glyI7H6F" - }, - "arxiv:1606.04130": { - "source": "arxiv", - "identifer": "1606.04130", - "standard_citation": "arxiv:1606.04130", - "bibtex": "@article{4zpZxjHR,\n abstract = {We demonstrate a simple strategy to cope with missing data in sequential\ninputs, addressing the task of multilabel classification of diagnoses given\nclinical time series. Collected from the pediatric intensive care unit (PICU)\nat Children's Hospital Los Angeles, our data consists of multivariate time\nseries of observations. The measurements are irregularly spaced, leading to\nmissingness patterns in temporally discretized sequences. While these artifacts\nare typically handled by imputation, we achieve superior predictive performance\nby treating the artifacts as features. Unlike linear models, recurrent neural\nnetworks can realize this improvement using only simple binary indicators of\nmissingness. For linear models, we show an alternative strategy to capture this\nsignal. Training models on missingness patterns only, we show that for some\ndiseases, what tests are run can be as predictive as the results themselves.},\n archiveprefix = {arXiv},\n author = {Zachary C. Lipton and David C. Kale and Randall Wetzel},\n eprint = {1606.04130v5},\n file = {1606.04130v5.pdf},\n month = {Jun},\n primaryclass = {cs.LG},\n title = {Modeling Missing Data in Clinical Time Series with RNNs},\n url = {https://arxiv.org/abs/1606.04130v5},\n year = {2016}\n}\n\n", - "citation_id": "4zpZxjHR" - }, - "doi:10.1016/j.neunet.2005.10.007": { - "source": "doi", - "identifer": "10.1016/j.neunet.2005.10.007", - "standard_citation": "doi:10.1016/j.neunet.2005.10.007", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T09:02:22Z", - "timestamp": 1507107742823 - }, - "reference-count": 57, - "publisher": "Elsevier BV", - "issue": "4", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2006, - 5, - 1 - ] - ], - "date-time": "2006-05-01T00:00:00Z", - "timestamp": 1146441600000 - }, - "delay-in-days": 0, - "content-version": "tdm" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2006, - 5 - ] - ] - }, - "DOI": "10.1016/j.neunet.2005.10.007", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2006, - 2, - 15 - ] - ], - "date-time": "2006-02-15T12:21:06Z", - "timestamp": 1140006066000 - }, - "page": "408-415", - "source": "Crossref", - "is-referenced-by-count": 164, - "title": "The use of artificial neural networks in decision support in cancer: A systematic review", - "prefix": "10.1016", - "volume": "19", - "author": [ - { - "given": "Paulo J.", - "family": "Lisboa", - "affiliation": [] - }, - { - "given": "Azzam F.G.", - "family": "Taktak", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Neural Networks", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S0893608005002844?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S0893608005002844?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 17 - ] - ], - "date-time": "2017-06-17T03:11:16Z", - "timestamp": 1497669076000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2006, - 5 - ] - ] - }, - "references-count": 57, - "alternative-id": [ - "S0893608005002844" - ], - "URL": "https://doi.org/10.1016/j.neunet.2005.10.007", - "relation": {}, - "subject": [ - "Cognitive Neuroscience", - "Artificial Intelligence" - ], - "container-title-short": "Neural Networks", - "id": "qxxwkSAT" - }, - "citation_id": "qxxwkSAT" - }, - "doi:10.1038/srep26286": { - "source": "doi", - "identifer": "10.1038/srep26286", - "standard_citation": "doi:10.1038/srep26286", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T02:02:17Z", - "timestamp": 1506996137368 - }, - "reference-count": 27, - "publisher": "Springer Nature", - "issue": "1", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2016, - 5, - 23 - ] - ], - "date-time": "2016-05-23T00:00:00Z", - "timestamp": 1463961600000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "link.springer.com" - ], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 9 - ] - ] - }, - "DOI": "10.1038/srep26286", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 5, - 24 - ] - ], - "date-time": "2016-05-24T06:23:18Z", - "timestamp": 1464070998000 - }, - "update-policy": "http://dx.doi.org/10.1007/springer_crossmark_policy", - "source": "Crossref", - "is-referenced-by-count": 28, - "title": "Deep learning as a tool for increased accuracy and efficiency of histopathological diagnosis", - "prefix": "10.1038", - "volume": "6", - "author": [ - { - "given": "Geert", - "family": "Litjens", - "affiliation": [] - }, - { - "given": "Clara I.", - "family": "Sánchez", - "affiliation": [] - }, - { - "given": "Nadya", - "family": "Timofeeva", - "affiliation": [] - }, - { - "given": "Meyke", - "family": "Hermsen", - "affiliation": [] - }, - { - "given": "Iris", - "family": "Nagtegaal", - "affiliation": [] - }, - { - "given": "Iringo", - "family": "Kovacs", - "affiliation": [] - }, - { - "given": "Christina", - "family": "Hulsbergen - van de Kaa", - "affiliation": [] - }, - { - "given": "Peter", - "family": "Bult", - "affiliation": [] - }, - { - "given": "Bram", - "family": "van Ginneken", - "affiliation": [] - }, - { - "given": "Jeroen", - "family": "van der Laak", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2016, - 5, - 23 - ] - ] - }, - "container-title": "Scientific Reports", - "original-title": [], - "link": [ - { - "URL": "http://www.nature.com/articles/srep26286", - "content-type": "text/html", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://www.nature.com/articles/srep26286.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://www.nature.com/articles/srep26286.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 5 - ] - ], - "date-time": "2017-08-05T02:24:46Z", - "timestamp": 1501899886000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 5, - 23 - ] - ] - }, - "references-count": 27, - "alternative-id": [ - "BFsrep26286" - ], - "URL": "https://doi.org/10.1038/srep26286", - "relation": { - "cites": [] - }, - "subject": [ - "Multidisciplinary" - ], - "container-title-short": "Sci Rep", - "article-number": "26286", - "id": "dQCjqq0Q" - }, - "citation_id": "dQCjqq0Q" - }, - "arxiv:1702.05747": { - "source": "arxiv", - "identifer": "1702.05747", - "standard_citation": "arxiv:1702.05747", - "bibtex": "@article{LL5huVs3,\n abstract = {Deep learning algorithms, in particular convolutional networks, have rapidly\nbecome a methodology of choice for analyzing medical images. This paper reviews\nthe major deep learning concepts pertinent to medical image analysis and\nsummarizes over 300 contributions to the field, most of which appeared in the\nlast year. We survey the use of deep learning for image classification, object\ndetection, segmentation, registration, and other tasks and provide concise\noverviews of studies per application area. Open challenges and directions for\nfuture research are discussed.},\n archiveprefix = {arXiv},\n author = {Geert Litjens and Thijs Kooi and Babak Ehteshami Bejnordi and Arnaud Arindra Adiyoso Setio and Francesco Ciompi and Mohsen Ghafoorian and Jeroen A. W. M. van der Laak and Bram van Ginneken and Clara I. Sánchez},\n doi = {10.1016/j.media.2017.07.005},\n eprint = {1702.05747v2},\n file = {1702.05747v2.pdf},\n month = {Feb},\n primaryclass = {cs.CV},\n title = {A Survey on Deep Learning in Medical Image Analysis},\n url = {https://arxiv.org/abs/1702.05747v2},\n year = {2017}\n}\n\n", - "citation_id": "LL5huVs3" - }, - "doi:10.1371/journal.pone.0097958": { - "source": "doi", - "identifer": "10.1371/journal.pone.0097958", - "standard_citation": "doi:10.1371/journal.pone.0097958", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 11 - ] - ], - "date-time": "2017-08-11T01:40:46Z", - "timestamp": 1502415646103 - }, - "update-to": [ - { - "updated": { - "date-parts": [ - [ - 2014, - 5, - 12 - ] - ], - "date-time": "2014-05-12T00:00:00Z", - "timestamp": 1399852800000 - }, - "DOI": "10.1371/journal.pone.0053253", - "type": "correction", - "label": "Correction" - } - ], - "reference-count": 0, - "publisher": "Public Library of Science (PLoS)", - "issue": "5", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2014, - 5, - 12 - ] - ], - "date-time": "2014-05-12T00:00:00Z", - "timestamp": 1399852800000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "www.plosone.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pone.0097958", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 5, - 12 - ] - ], - "date-time": "2014-05-12T21:24:23Z", - "timestamp": 1399929863000 - }, - "page": "e97958", - "update-policy": "http://dx.doi.org/10.1371/journal.pone.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Correction: Class Prediction and Feature Selection with Linear Optimization for Metagenomic Count Data", - "prefix": "10.1371", - "volume": "9", - "author": [ - { - "given": "Zhenqiu", - "family": "Liu", - "affiliation": [] - }, - { - "given": "Dechang", - "family": "Chen", - "affiliation": [] - }, - { - "given": "Li", - "family": "Sheng", - "affiliation": [] - }, - { - "given": "Amy Y.", - "family": "Liu", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2014, - 5, - 12 - ] - ] - }, - "container-title": "PLoS ONE", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2016, - 4, - 20 - ] - ], - "date-time": "2016-04-20T18:04:46Z", - "timestamp": 1461175486000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 5, - 12 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1371/journal.pone.0097958", - "relation": {}, - "subject": [ - "General Biochemistry, Genetics and Molecular Biology", - "General Agricultural and Biological Sciences", - "General Medicine" - ], - "container-title-short": "PLoS ONE", - "id": "Kt9NojjR" - }, - "citation_id": "Kt9NojjR" - }, - "doi:10.12688/f1000research.7223.1": { - "source": "doi", - "identifer": "10.12688/f1000research.7223.1", - "standard_citation": "doi:10.12688/f1000research.7223.1", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 12 - ] - ], - "date-time": "2017-09-12T07:22:15Z", - "timestamp": 1505200935894 - }, - "reference-count": 0, - "publisher": "F1000 Research, Ltd.", - "content-domain": { - "domain": [ - "f1000research.com" - ], - "crossmark-restriction": true - }, - "DOI": "10.12688/f1000research.7223.1", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 2, - 17 - ] - ], - "date-time": "2016-02-17T11:35:11Z", - "timestamp": 1455708911000 - }, - "update-policy": "http://dx.doi.org/10.12688/f1000research.crossmark-policy", - "source": "Crossref", - "is-referenced-by-count": 5, - "title": "Single-cell transcriptome sequencing: recent advances and remaining challenges", - "prefix": "10.12688", - "author": [ - { - "given": "Serena", - "family": "Liu", - "affiliation": [] - }, - { - "given": "Cole", - "family": "Trapnell", - "affiliation": [] - } - ], - "member": "4950", - "published-online": { - "date-parts": [ - [ - 2016, - 2, - 17 - ] - ] - }, - "container-title": "F1000Research", - "original-title": [], - "link": [ - { - "URL": "http://f1000research.com/articles/5-182/v1/iparadigms", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 2, - 17 - ] - ], - "date-time": "2016-02-17T16:05:07Z", - "timestamp": 1455725107000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 2, - 17 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.12688/f1000research.7223.1", - "relation": {}, - "container-title-short": "F1000Res", - "assertion": [ - { - "value": "Indexed", - "URL": "http://f1000research.com/articles/5-182/v1#article-reports", - "order": 0, - "name": "referee-status", - "label": "Referee status", - "group": { - "name": "current-referee-status", - "label": "Current Referee Status" - } - }, - { - "value": "10.5256/f1000research.7781.r12513, Sten Linnarsson, Unit of Molecular Neurobiology, Department of Medical Biochemistry and Biophysics, Karolinska Institutet, Stockholm, Sweden, 17 Feb 2016, version 1, indexed", - "URL": "http://f1000research.com/articles/5-182/v1#referee-response-12513", - "order": 0, - "name": "referee-response-12513", - "label": "Referee Report", - "group": { - "name": "article-reports", - "label": "Article Reports" - } - }, - { - "value": "10.5256/f1000research.7781.r12514, Roger S Lasken, Sara B Linker, J. Craig Venter Institute, La Jolla, CA, 92121, USA, 17 Feb 2016, version 1, indexed", - "URL": "http://f1000research.com/articles/5-182/v1#referee-response-12514", - "order": 1, - "name": "referee-response-12514", - "label": "Referee Report", - "group": { - "name": "article-reports", - "label": "Article Reports" - } - }, - { - "value": "CT is supported by an NIH DP2 HD088158 grant, an Alfred P. Sloan Fellowship, and a Dale F. Frey Award for Breakthrough Scientists from the Damon Runyon Cancer Research Foundation. SL is supported by an NSF IGERT grant DGE-1258485.The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.", - "order": 2, - "name": "grant-information", - "label": "Grant Information" - }, - { - "value": "This is an open access article distributed under the terms of the Creative Commons Attribution Licence, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.", - "order": 0, - "name": "copyright-info", - "label": "Copyright" - } - ], - "id": "QafUwNKn" - }, - "citation_id": "QafUwNKn" - }, - "arxiv:1604.07043": { - "source": "arxiv", - "identifer": "1604.07043", - "standard_citation": "arxiv:1604.07043", - "bibtex": "@article{AEc66xxR,\n abstract = {Deep convolutional neural networks (CNNs) have achieved breakthrough\nperformance in many pattern recognition tasks such as image classification.\nHowever, the development of high-quality deep models typically relies on a\nsubstantial amount of trial-and-error, as there is still no clear understanding\nof when and why a deep model works. In this paper, we present a visual\nanalytics approach for better understanding, diagnosing, and refining deep\nCNNs. We formulate a deep CNN as a directed acyclic graph. Based on this\nformulation, a hybrid visualization is developed to disclose the multiple\nfacets of each neuron and the interactions between them. In particular, we\nintroduce a hierarchical rectangle packing algorithm and a matrix reordering\nalgorithm to show the derived features of a neuron cluster. We also propose a\nbiclustering-based edge bundling method to reduce visual clutter caused by a\nlarge number of connections between neurons. We evaluated our method on a set\nof CNNs and the results are generally favorable.},\n archiveprefix = {arXiv},\n author = {Mengchen Liu and Jiaxin Shi and Zhen Li and Chongxuan Li and Jun Zhu and Shixia Liu},\n eprint = {1604.07043v3},\n file = {1604.07043v3.pdf},\n month = {Apr},\n primaryclass = {cs.CV},\n title = {Towards Better Analysis of Deep Convolutional Neural Networks},\n url = {https://arxiv.org/abs/1604.07043v3},\n year = {2016}\n}\n\n", - "citation_id": "AEc66xxR" - }, - "doi:10.1126/science.aab1785": { - "source": "doi", - "identifer": "10.1126/science.aab1785", - "standard_citation": "doi:10.1126/science.aab1785", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 30 - ] - ], - "date-time": "2017-09-30T00:22:12Z", - "timestamp": 1506730932761 - }, - "reference-count": 54, - "publisher": "American Association for the Advancement of Science (AAAS)", - "issue": "6256", - "license": [ - { - "URL": "http://www.sciencemag.org/about/science-licenses-journal-article-reuse", - "start": { - "date-parts": [ - [ - 2016, - 10, - 2 - ] - ], - "date-time": "2016-10-02T00:00:00Z", - "timestamp": 1475366400000 - }, - "delay-in-days": 367, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/100000011", - "name": "Howard Hughes Medical Institute", - "doi-asserted-by": "publisher", - "award": [] - }, - { - "DOI": "10.13039/100000025", - "name": "National Institute of Mental Health", - "doi-asserted-by": "publisher", - "award": [ - "P50 MH106933" - ] - }, - { - "DOI": "10.13039/100000049", - "name": "National Institute on Aging", - "doi-asserted-by": "publisher", - "award": [ - "T32 AG000222" - ] - }, - { - "DOI": "10.13039/100000057", - "name": "National Institute of General Medical Sciences", - "doi-asserted-by": "publisher", - "award": [ - "T32 GM007226", - "T32 GM007753" - ] - }, - { - "DOI": "10.13039/100000065", - "name": "National Institute of Neurological Disorders and Stroke", - "doi-asserted-by": "publisher", - "award": [ - "U01 MH106883", - "R01 NS079277", - "R01 NS032457" - ] - }, - { - "DOI": "10.13039/100000097", - "name": "National Center for Research Resources", - "doi-asserted-by": "publisher", - "award": [ - "1S10RR028832-01" - ] - }, - { - "DOI": "10.13039/100000952", - "name": "Paul G. Allen Family Foundation", - "doi-asserted-by": "publisher", - "award": [] - }, - { - "name": "Leonard and Isabelle Goldenson Research Fellowship", - "award": [] - }, - { - "name": "Louis Lange III Scholarship in Translational Research", - "award": [] - }, - { - "name": "Eleanor and Miles Shore Fellowship", - "award": [] - }, - { - "name": "Manton Center for Orphan Disease Research", - "award": [] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015, - 10, - 2 - ] - ] - }, - "DOI": "10.1126/science.aab1785", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 10, - 1 - ] - ], - "date-time": "2015-10-01T18:14:33Z", - "timestamp": 1443723273000 - }, - "page": "94-98", - "source": "Crossref", - "is-referenced-by-count": 73, - "title": "Somatic mutation in single human neurons tracks developmental and transcriptional history", - "prefix": "10.1126", - "volume": "350", - "author": [ - { - "given": "M. A.", - "family": "Lodato", - "affiliation": [] - }, - { - "given": "M. B.", - "family": "Woodworth", - "affiliation": [] - }, - { - "given": "S.", - "family": "Lee", - "affiliation": [] - }, - { - "given": "G. D.", - "family": "Evrony", - "affiliation": [] - }, - { - "given": "B. K.", - "family": "Mehta", - "affiliation": [] - }, - { - "given": "A.", - "family": "Karger", - "affiliation": [] - }, - { - "given": "S.", - "family": "Lee", - "affiliation": [] - }, - { - "given": "T. W.", - "family": "Chittenden", - "affiliation": [] - }, - { - "given": "A. M.", - "family": "D'Gama", - "affiliation": [] - }, - { - "given": "X.", - "family": "Cai", - "affiliation": [] - }, - { - "given": "L. J.", - "family": "Luquette", - "affiliation": [] - }, - { - "given": "E.", - "family": "Lee", - "affiliation": [] - }, - { - "given": "P. J.", - "family": "Park", - "affiliation": [] - }, - { - "given": "C. A.", - "family": "Walsh", - "affiliation": [] - } - ], - "member": "221", - "published-online": { - "date-parts": [ - [ - 2015, - 10, - 1 - ] - ] - }, - "container-title": "Science", - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1126/science.aab1785", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 23 - ] - ], - "date-time": "2017-06-23T20:35:55Z", - "timestamp": 1498250155000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 10, - 1 - ] - ] - }, - "references-count": 54, - "URL": "https://doi.org/10.1126/science.aab1785", - "relation": { - "cites": [] - }, - "subject": [ - "Multidisciplinary" - ], - "container-title-short": "Science", - "id": "1GvfSy48x" - }, - "citation_id": "1GvfSy48x" - }, - "arxiv:1611.07478": { - "source": "arxiv", - "identifer": "1611.07478", - "standard_citation": "arxiv:1611.07478", - "bibtex": "@article{DeOI1oGf,\n abstract = {Understanding why a model made a certain prediction is crucial in many data\nscience fields. Interpretable predictions engender appropriate trust and\nprovide insight into how the model may be improved. However, with large modern\ndatasets the best accuracy is often achieved by complex models even experts\nstruggle to interpret, which creates a tension between accuracy and\ninterpretability. Recently, several methods have been proposed for interpreting\npredictions from complex models by estimating the importance of input features.\nHere, we present how a model-agnostic additive representation of the importance\nof input features unifies current methods. This representation is optimal, in\nthe sense that it is the only set of additive values that satisfies important\nproperties. We show how we can leverage these properties to create novel visual\nexplanations of model predictions. The thread of unity that this representation\nweaves through the literature indicates that there are common principles to be\nlearned about the interpretation of model predictions that apply in many\nscenarios.},\n archiveprefix = {arXiv},\n author = {Scott Lundberg and Su-In Lee},\n eprint = {1611.07478v3},\n file = {1611.07478v3.pdf},\n month = {Dec},\n primaryclass = {cs.AI},\n title = {An unexpected unity among methods for interpreting model predictions},\n url = {https://arxiv.org/abs/1611.07478v3},\n year = {2016}\n}\n\n", - "citation_id": "DeOI1oGf" - }, - "doi:10.1021/ci400187y": { - "source": "doi", - "identifer": "10.1021/ci400187y", - "standard_citation": "doi:10.1021/ci400187y", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 27 - ] - ], - "date-time": "2017-09-27T10:42:17Z", - "timestamp": 1506508937735 - }, - "reference-count": 73, - "publisher": "American Chemical Society (ACS)", - "issue": "7", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2013, - 7, - 22 - ] - ] - }, - "DOI": "10.1021/ci400187y", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2013, - 6, - 24 - ] - ], - "date-time": "2013-06-24T14:26:33Z", - "timestamp": 1372083993000 - }, - "page": "1563-1575", - "source": "Crossref", - "is-referenced-by-count": 61, - "title": "Deep Architectures and Deep Learning in Chemoinformatics: The Prediction of Aqueous Solubility for Drug-Like Molecules", - "prefix": "10.1021", - "volume": "53", - "author": [ - { - "given": "Alessandro", - "family": "Lusci", - "affiliation": [] - }, - { - "given": "Gianluca", - "family": "Pollastri", - "affiliation": [] - }, - { - "given": "Pierre", - "family": "Baldi", - "affiliation": [] - } - ], - "member": "316", - "container-title": "Journal of Chemical Information and Modeling", - "original-title": [], - "link": [ - { - "URL": "http://pubs.acs.org/doi/pdf/10.1021/ci400187y", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 21 - ] - ], - "date-time": "2017-06-21T15:08:23Z", - "timestamp": 1498057703000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2013, - 7, - 22 - ] - ] - }, - "references-count": 73, - "alternative-id": [ - "10.1021/ci400187y" - ], - "URL": "https://doi.org/10.1021/ci400187y", - "relation": {}, - "subject": [ - "General Chemistry", - "General Chemical Engineering", - "Library and Information Sciences", - "Computer Science Applications" - ], - "container-title-short": "J. Chem. Inf. Model.", - "id": "17Wih4Hd5" - }, - "citation_id": "17Wih4Hd5" - }, - "doi:10.1186/s13321-015-0110-6": { - "source": "doi", - "identifer": "10.1186/s13321-015-0110-6", - "standard_citation": "doi:10.1186/s13321-015-0110-6", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 14 - ] - ], - "date-time": "2017-08-14T12:22:21Z", - "timestamp": 1502713341228 - }, - "reference-count": 57, - "publisher": "Springer Nature", - "issue": "1", - "funder": [ - { - "DOI": "10.13039/100000001", - "name": "National Science Foundation", - "doi-asserted-by": "publisher", - "award": [ - "IIS-0513376" - ] - }, - { - "DOI": "10.13039/100000002", - "name": "National Institutes of Health", - "doi-asserted-by": "publisher", - "award": [ - "T15 LM07443", - "LM010235" - ] - }, - { - "DOI": "10.13039/100006785", - "name": "Google", - "doi-asserted-by": "publisher", - "award": [ - "Faculty Research Award" - ] - } - ], - "content-domain": { - "domain": [ - "link.springer.com" - ], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015, - 12 - ] - ] - }, - "DOI": "10.1186/s13321-015-0110-6", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 12, - 29 - ] - ], - "date-time": "2015-12-29T13:33:20Z", - "timestamp": 1451396000000 - }, - "update-policy": "http://dx.doi.org/10.1007/springer_crossmark_policy", - "source": "Crossref", - "is-referenced-by-count": 7, - "title": "Accurate and efficient target prediction using a potency-sensitive influence-relevance voter", - "prefix": "10.1186", - "volume": "7", - "author": [ - { - "given": "Alessandro", - "family": "Lusci", - "affiliation": [] - }, - { - "given": "David", - "family": "Fooshee", - "affiliation": [] - }, - { - "given": "Michael", - "family": "Browning", - "affiliation": [] - }, - { - "given": "Joshua", - "family": "Swamidass", - "affiliation": [] - }, - { - "given": "Pierre", - "family": "Baldi", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2015, - 12, - 29 - ] - ] - }, - "container-title": "Journal of Cheminformatics", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1186/s13321-015-0110-6", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T02:40:39Z", - "timestamp": 1498272039000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 12 - ] - ] - }, - "references-count": 57, - "alternative-id": [ - "110" - ], - "URL": "https://doi.org/10.1186/s13321-015-0110-6", - "relation": { - "cites": [] - }, - "subject": [ - "Physical and Theoretical Chemistry", - "Library and Information Sciences", - "Computer Graphics and Computer-Aided Design", - "Computer Science Applications" - ], - "container-title-short": "J Cheminform", - "article-number": "63", - "id": "1E0x7QgLP" - }, - "citation_id": "1E0x7QgLP" - }, - "doi:10.1021/ci500747n": { - "source": "doi", - "identifer": "10.1021/ci500747n", - "standard_citation": "doi:10.1021/ci500747n", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T01:42:17Z", - "timestamp": 1506994937309 - }, - "reference-count": 20, - "publisher": "American Chemical Society (ACS)", - "issue": "2", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015, - 2, - 23 - ] - ] - }, - "DOI": "10.1021/ci500747n", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 1, - 30 - ] - ], - "date-time": "2015-01-30T19:25:01Z", - "timestamp": 1422645901000 - }, - "page": "263-274", - "source": "Crossref", - "is-referenced-by-count": 66, - "title": "Deep Neural Nets as a Method for Quantitative Structure–Activity Relationships", - "prefix": "10.1021", - "volume": "55", - "author": [ - { - "given": "Junshui", - "family": "Ma", - "affiliation": [] - }, - { - "given": "Robert P.", - "family": "Sheridan", - "affiliation": [] - }, - { - "given": "Andy", - "family": "Liaw", - "affiliation": [] - }, - { - "given": "George E.", - "family": "Dahl", - "affiliation": [] - }, - { - "given": "Vladimir", - "family": "Svetnik", - "affiliation": [] - } - ], - "member": "316", - "container-title": "Journal of Chemical Information and Modeling", - "original-title": [], - "link": [ - { - "URL": "http://pubs.acs.org/doi/pdf/10.1021/ci500747n", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 23 - ] - ], - "date-time": "2017-06-23T04:40:33Z", - "timestamp": 1498192833000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 2, - 23 - ] - ] - }, - "references-count": 20, - "alternative-id": [ - "10.1021/ci500747n" - ], - "URL": "https://doi.org/10.1021/ci500747n", - "relation": {}, - "subject": [ - "General Chemistry", - "General Chemical Engineering", - "Library and Information Sciences", - "Computer Science Applications" - ], - "container-title-short": "J. Chem. Inf. Model.", - "id": "xOaTIeBY" - }, - "citation_id": "xOaTIeBY" - }, - "arxiv:1412.0035": { - "source": "arxiv", - "identifer": "1412.0035", - "standard_citation": "arxiv:1412.0035", - "bibtex": "@article{19mGl6pfy,\n abstract = {Image representations, from SIFT and Bag of Visual Words to Convolutional\nNeural Networks (CNNs), are a crucial component of almost any image\nunderstanding system. Nevertheless, our understanding of them remains limited.\nIn this paper we conduct a direct analysis of the visual information contained\nin representations by asking the following question: given an encoding of an\nimage, to which extent is it possible to reconstruct the image itself? To\nanswer this question we contribute a general framework to invert\nrepresentations. We show that this method can invert representations such as\nHOG and SIFT more accurately than recent alternatives while being applicable to\nCNNs too. We then use this technique to study the inverse of recent\nstate-of-the-art CNN image representations for the first time. Among our\nfindings, we show that several layers in CNNs retain photographically accurate\ninformation about the image, with different degrees of geometric and\nphotometric invariance.},\n archiveprefix = {arXiv},\n author = {Aravindh Mahendran and Andrea Vedaldi},\n eprint = {1412.0035v1},\n file = {1412.0035v1.pdf},\n month = {Dec},\n primaryclass = {cs.CV},\n title = {Understanding Deep Image Representations by Inverting Them},\n url = {https://arxiv.org/abs/1412.0035v1},\n year = {2014}\n}\n\n", - "citation_id": "19mGl6pfy" - }, - "doi:10.1007/978-3-319-46466-4_8": { - "source": "doi", - "identifer": "10.1007/978-3-319-46466-4_8", - "standard_citation": "doi:10.1007/978-3-319-46466-4_8", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T23:12:57Z", - "timestamp": 1502320377534 - }, - "publisher-location": "Cham", - "reference-count": 27, - "publisher": "Springer International Publishing", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2016, - 1, - 1 - ] - ], - "date-time": "2016-01-01T00:00:00Z", - "timestamp": 1451606400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016 - ] - ] - }, - "DOI": "10.1007/978-3-319-46466-4_8", - "type": "chapter", - "created": { - "date-parts": [ - [ - 2016, - 9, - 16 - ] - ], - "date-time": "2016-09-16T09:31:58Z", - "timestamp": 1474018318000 - }, - "page": "120-135", - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Salient Deconvolutional Networks", - "prefix": "10.1007", - "author": [ - { - "given": "Aravindh", - "family": "Mahendran", - "affiliation": [] - }, - { - "given": "Andrea", - "family": "Vedaldi", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2016, - 9, - 17 - ] - ] - }, - "container-title": "Computer Vision – ECCV 2016", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1007/978-3-319-46466-4_8", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T22:57:42Z", - "timestamp": 1498345062000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016 - ] - ] - }, - "references-count": 27, - "URL": "https://doi.org/10.1007/978-3-319-46466-4_8", - "relation": { - "cites": [] - }, - "id": "vjXoJqO3" - }, - "citation_id": "vjXoJqO3" - }, - "doi:10.1007/s11263-016-0911-8": { - "source": "doi", - "identifer": "10.1007/s11263-016-0911-8", - "standard_citation": "doi:10.1007/s11263-016-0911-8", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 29 - ] - ], - "date-time": "2017-09-29T16:42:13Z", - "timestamp": 1506703333983 - }, - "reference-count": 55, - "publisher": "Springer Nature", - "issue": "3", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2016, - 5, - 18 - ] - ], - "date-time": "2016-05-18T00:00:00Z", - "timestamp": 1463529600000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/100004364", - "name": "BP", - "doi-asserted-by": "publisher", - "award": [] - }, - { - "name": "ERC-STG", - "award": [ - "IDIU-638009" - ] - } - ], - "content-domain": { - "domain": [ - "link.springer.com" - ], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 12 - ] - ] - }, - "DOI": "10.1007/s11263-016-0911-8", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 5, - 18 - ] - ], - "date-time": "2016-05-18T05:50:31Z", - "timestamp": 1463550631000 - }, - "page": "233-255", - "update-policy": "http://dx.doi.org/10.1007/springer_crossmark_policy", - "source": "Crossref", - "is-referenced-by-count": 10, - "title": "Visualizing Deep Convolutional Neural Networks Using Natural Pre-images", - "prefix": "10.1007", - "volume": "120", - "author": [ - { - "ORCID": "http://orcid.org/0000-0002-2650-9871", - "authenticated-orcid": false, - "given": "Aravindh", - "family": "Mahendran", - "affiliation": [] - }, - { - "given": "Andrea", - "family": "Vedaldi", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2016, - 5, - 18 - ] - ] - }, - "container-title": "International Journal of Computer Vision", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1007/s11263-016-0911-8.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://link.springer.com/article/10.1007/s11263-016-0911-8/fulltext.html", - "content-type": "text/html", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://link.springer.com/content/pdf/10.1007/s11263-016-0911-8", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T13:52:09Z", - "timestamp": 1498312329000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 5, - 18 - ] - ] - }, - "references-count": 55, - "alternative-id": [ - "911" - ], - "URL": "https://doi.org/10.1007/s11263-016-0911-8", - "relation": { - "cites": [] - }, - "subject": [ - "Software", - "Artificial Intelligence", - "Computer Vision and Pattern Recognition" - ], - "container-title-short": "Int J Comput Vis", - "id": "1FkT6C6oa" - }, - "citation_id": "1FkT6C6oa" - }, - "doi:10.1145/1327452.1327492": { - "source": "doi", - "identifer": "10.1145/1327452.1327492", - "standard_citation": "doi:10.1145/1327452.1327492", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T08:22:17Z", - "timestamp": 1507105337520 - }, - "reference-count": 0, - "publisher": "Association for Computing Machinery (ACM)", - "issue": "1", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2008, - 1, - 1 - ] - ] - }, - "DOI": "10.1145/1327452.1327492", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2008, - 1, - 3 - ] - ], - "date-time": "2008-01-03T18:20:10Z", - "timestamp": 1199384410000 - }, - "page": "107", - "source": "Crossref", - "is-referenced-by-count": 4905, - "title": "MapReduce", - "prefix": "10.1145", - "volume": "51", - "author": [ - { - "given": "Jeffrey", - "family": "Dean", - "affiliation": [] - }, - { - "given": "Sanjay", - "family": "Ghemawat", - "affiliation": [] - } - ], - "member": "320", - "container-title": "Communications of the ACM", - "original-title": [], - "link": [ - { - "URL": "http://dl.acm.org/ft_gateway.cfm?id=1327492&ftid=476270&dwn=1", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 12, - 16 - ] - ], - "date-time": "2016-12-16T14:57:01Z", - "timestamp": 1481900221000 - }, - "score": 1.0, - "subtitle": [ - "simplified data processing on large clusters" - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2008, - 1, - 1 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1145/1327452.1327492", - "relation": {}, - "subject": [ - "General Computer Science" - ], - "container-title-short": "Commun. ACM", - "id": "xE3EYmck" - }, - "citation_id": "xE3EYmck" - }, - "doi:10.3389/fenvs.2015.00080": { - "source": "doi", - "identifer": "10.3389/fenvs.2015.00080", - "standard_citation": "doi:10.3389/fenvs.2015.00080", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 4 - ] - ], - "date-time": "2017-09-04T12:22:10Z", - "timestamp": 1504527730821 - }, - "reference-count": 72, - "publisher": "Frontiers Media SA", - "funder": [ - { - "DOI": "10.13039/501100000780", - "name": "European Commission", - "doi-asserted-by": "publisher", - "award": [] - }, - { - "name": "ChemBioBridge", - "award": [] - } - ], - "content-domain": { - "domain": [ - "frontiersin.org" - ], - "crossmark-restriction": true - }, - "DOI": "10.3389/fenvs.2015.00080", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 2, - 2 - ] - ], - "date-time": "2016-02-02T06:27:12Z", - "timestamp": 1454394432000 - }, - "update-policy": "http://dx.doi.org/10.3389/crossmark-policy", - "source": "Crossref", - "is-referenced-by-count": 12, - "title": "DeepTox: Toxicity Prediction using Deep Learning", - "prefix": "10.3389", - "volume": "3", - "author": [ - { - "given": "Andreas", - "family": "Mayr", - "affiliation": [] - }, - { - "given": "Günter", - "family": "Klambauer", - "affiliation": [] - }, - { - "given": "Thomas", - "family": "Unterthiner", - "affiliation": [] - }, - { - "given": "Sepp", - "family": "Hochreiter", - "affiliation": [] - } - ], - "member": "1965", - "published-online": { - "date-parts": [ - [ - 2016, - 2, - 2 - ] - ] - }, - "container-title": "Frontiers in Environmental Science", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T05:00:50Z", - "timestamp": 1498280450000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 2, - 2 - ] - ] - }, - "references-count": 72, - "URL": "https://doi.org/10.3389/fenvs.2015.00080", - "relation": { - "cites": [] - }, - "container-title-short": "Front. Environ. Sci.", - "id": "Y1D0SZrO" - }, - "citation_id": "Y1D0SZrO" - }, - "doi:10.1038/nmeth976": { - "source": "doi", - "identifer": "10.1038/nmeth976", - "standard_citation": "doi:10.1038/nmeth976", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T11:22:17Z", - "timestamp": 1507116137275 - }, - "reference-count": 35, - "publisher": "Springer Nature", - "issue": "1", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2007, - 1 - ] - ] - }, - "DOI": "10.1038/nmeth976", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2006, - 12, - 18 - ] - ], - "date-time": "2006-12-18T15:32:55Z", - "timestamp": 1166455975000 - }, - "page": "63-72", - "source": "Crossref", - "is-referenced-by-count": 229, - "title": "Accurate phylogenetic classification of variable-length DNA fragments", - "prefix": "10.1038", - "volume": "4", - "author": [ - { - "given": "Alice Carolyn", - "family": "McHardy", - "affiliation": [] - }, - { - "given": "Héctor García", - "family": "Martín", - "affiliation": [] - }, - { - "given": "Aristotelis", - "family": "Tsirigos", - "affiliation": [] - }, - { - "given": "Philip", - "family": "Hugenholtz", - "affiliation": [] - }, - { - "given": "Isidore", - "family": "Rigoutsos", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2006, - 12, - 10 - ] - ] - }, - "container-title": "Nature Methods", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 17 - ] - ], - "date-time": "2017-06-17T12:21:37Z", - "timestamp": 1497702097000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2006, - 12, - 10 - ] - ] - }, - "references-count": 35, - "alternative-id": [ - "nmeth976" - ], - "URL": "https://doi.org/10.1038/nmeth976", - "relation": { - "cites": [] - }, - "subject": [ - "Biotechnology", - "Cell Biology", - "Biochemistry", - "Molecular Biology" - ], - "container-title-short": "Nat Meth", - "id": "QV551Nlx" - }, - "citation_id": "QV551Nlx" - }, - "arxiv:1505.06807": { - "source": "arxiv", - "identifer": "1505.06807", - "standard_citation": "arxiv:1505.06807", - "bibtex": "@article{rZnxDitd,\n abstract = {Apache Spark is a popular open-source platform for large-scale data\nprocessing that is well-suited for iterative machine learning tasks. In this\npaper we present MLlib, Spark's open-source distributed machine learning\nlibrary. MLlib provides efficient functionality for a wide range of learning\nsettings and includes several underlying statistical, optimization, and linear\nalgebra primitives. Shipped with Spark, MLlib supports several languages and\nprovides a high-level API that leverages Spark's rich ecosystem to simplify the\ndevelopment of end-to-end machine learning pipelines. MLlib has experienced a\nrapid growth due to its vibrant open-source community of over 140 contributors, and includes extensive documentation to support further growth and to let users\nquickly get up to speed.},\n archiveprefix = {arXiv},\n author = {Xiangrui Meng and Joseph Bradley and Burak Yavuz and Evan Sparks and Shivaram Venkataraman and Davies Liu and Jeremy Freeman and DB Tsai and Manish Amde and Sean Owen and Doris Xin and Reynold Xin and Michael J. Franklin and Reza Zadeh and Matei Zaharia and Ameet Talwalkar},\n eprint = {1505.06807v1},\n file = {1505.06807v1.pdf},\n month = {May},\n primaryclass = {cs.LG},\n title = {MLlib: Machine Learning in Apache Spark},\n url = {https://arxiv.org/abs/1505.06807v1},\n year = {2015}\n}\n\n", - "citation_id": "rZnxDitd" - }, - "doi:10.1038/nmeth.2066": { - "source": "doi", - "identifer": "10.1038/nmeth.2066", - "standard_citation": "doi:10.1038/nmeth.2066", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T11:22:17Z", - "timestamp": 1507116137281 - }, - "reference-count": 37, - "publisher": "Springer Nature", - "issue": "8", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/nmeth.2066", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2012, - 6, - 11 - ] - ], - "date-time": "2012-06-11T13:13:51Z", - "timestamp": 1339420431000 - }, - "page": "811-814", - "source": "Crossref", - "is-referenced-by-count": 357, - "title": "Metagenomic microbial community profiling using unique clade-specific marker genes", - "prefix": "10.1038", - "volume": "9", - "author": [ - { - "given": "Nicola", - "family": "Segata", - "affiliation": [] - }, - { - "given": "Levi", - "family": "Waldron", - "affiliation": [] - }, - { - "given": "Annalisa", - "family": "Ballarini", - "affiliation": [] - }, - { - "given": "Vagheesh", - "family": "Narasimhan", - "affiliation": [] - }, - { - "given": "Olivier", - "family": "Jousson", - "affiliation": [] - }, - { - "given": "Curtis", - "family": "Huttenhower", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2012, - 6, - 10 - ] - ] - }, - "container-title": "Nature Methods", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 20 - ] - ], - "date-time": "2017-06-20T19:54:29Z", - "timestamp": 1497988469000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2012, - 6, - 10 - ] - ] - }, - "references-count": 37, - "alternative-id": [ - "nmeth.2066" - ], - "URL": "https://doi.org/10.1038/nmeth.2066", - "relation": { - "cites": [] - }, - "subject": [ - "Biotechnology", - "Cell Biology", - "Biochemistry", - "Molecular Biology" - ], - "container-title-short": "Nat Meth", - "id": "56wEWVIl" - }, - "citation_id": "56wEWVIl" - }, - "doi:10.1109/bibm.2016.7822593": { - "source": "doi", - "identifer": "10.1109/bibm.2016.7822593", - "standard_citation": "doi:10.1109/bibm.2016.7822593", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T20:35:36Z", - "timestamp": 1502397336252 - }, - "reference-count": 34, - "publisher": "IEEE", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 12 - ] - ] - }, - "DOI": "10.1109/bibm.2016.7822593", - "type": "paper-conference", - "created": { - "date-parts": [ - [ - 2017, - 1, - 20 - ] - ], - "date-time": "2017-01-20T03:36:10Z", - "timestamp": 1484883370000 - }, - "source": "Crossref", - "is-referenced-by-count": 1, - "title": "DeepEnhancer: Predicting enhancers by convolutional neural networks", - "prefix": "10.1109", - "author": [ - { - "family": "Xu Min", - "affiliation": [] - }, - { - "family": "Ning Chen", - "affiliation": [] - }, - { - "family": "Ting Chen", - "affiliation": [] - }, - { - "family": "Rui Jiang", - "affiliation": [] - } - ], - "member": "263", - "container-title": "2016 IEEE International Conference on Bioinformatics and Biomedicine (BIBM)", - "original-title": [], - "link": [ - { - "URL": "http://xplorestaging.ieee.org/ielx7/7811899/7822470/07822593.pdf?arnumber=7822593", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T08:41:08Z", - "timestamp": 1498380068000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 12 - ] - ] - }, - "references-count": 34, - "URL": "https://doi.org/10.1109/bibm.2016.7822593", - "relation": {}, - "id": "jV2YerUS" - }, - "citation_id": "jV2YerUS" - }, - "url:http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html": { - "source": "url", - "identifer": "http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html", - "standard_citation": "url:http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html", - "citeproc": { - "URL": "http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html", - "greycite-canonical-uri": "https://research.googleblog.com/2015/06/inceptionism-going-deeper-into-neural.html", - "title": "Inceptionism: Going Deeper into Neural Networks", - "container-title": "Google Research Blog", - "issued": { - "date-parts": [ - [ - 2015, - 6 - ] - ] - }, - "author": [ - { - "given": "Alexander", - "family": "Mordvintsev" - }, - { - "given": "Christopher", - "family": "Olah" - }, - { - "given": "Mike", - "family": "Tyka" - } - ], - "archives": [ - "http://wayback.archive.org/web/http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html" - ], - "type": "webpage", - "id": "XLHInhc1" - }, - "citation_id": "XLHInhc1" - }, - "arxiv:1511.06051": { - "source": "arxiv", - "identifer": "1511.06051", - "standard_citation": "arxiv:1511.06051", - "bibtex": "@article{rmJZ2Aui,\n abstract = {Training deep networks is a time-consuming process, with networks for object\nrecognition often requiring multiple days to train. For this reason, leveraging\nthe resources of a cluster to speed up training is an important area of work.\nHowever, widely-popular batch-processing computational frameworks like\nMapReduce and Spark were not designed to support the asynchronous and\ncommunication-intensive workloads of existing distributed deep learning\nsystems. We introduce SparkNet, a framework for training deep networks in\nSpark. Our implementation includes a convenient interface for reading data from\nSpark RDDs, a Scala interface to the Caffe deep learning framework, and a\nlightweight multi-dimensional tensor library. Using a simple parallelization\nscheme for stochastic gradient descent, SparkNet scales well with the cluster\nsize and tolerates very high-latency communication. Furthermore, it is easy to\ndeploy and use with no parameter tuning, and it is compatible with existing\nCaffe models. We quantify the dependence of the speedup obtained by SparkNet on\nthe number of machines, the communication frequency, and the cluster's\ncommunication overhead, and we benchmark our system's performance on the\nImageNet dataset.},\n archiveprefix = {arXiv},\n author = {Philipp Moritz and Robert Nishihara and Ion Stoica and Michael I. Jordan},\n eprint = {1511.06051v4},\n file = {1511.06051v4.pdf},\n month = {Dec},\n primaryclass = {stat.ML},\n title = {SparkNet: Training Deep Networks in Spark},\n url = {https://arxiv.org/abs/1511.06051v4},\n year = {2015}\n}\n\n", - "citation_id": "rmJZ2Aui" - }, - "url:https://repozitorij.uni-lj.si/IzpisGradiva.php?id=85515": { - "source": "url", - "identifer": "https://repozitorij.uni-lj.si/IzpisGradiva.php?id=85515", - "standard_citation": "url:https://repozitorij.uni-lj.si/IzpisGradiva.php?id=85515", - "citeproc": { - "type": "report", - "title": "Globoko ucenje na genomskih in filogenetskih podatkih", - "publisher": "Univerza v Ljubljani, Fakulteta za racunalništvo in informatiko", - "source": "repozitorij.uni-lj.si", - "URL": "https://repozitorij.uni-lj.si/IzpisGradiva.php?id=85515", - "language": "slv", - "author": [ - { - "family": "Mrzelj", - "given": "Nina" - } - ], - "issued": { - "date-parts": [ - [ - "2016" - ] - ] - }, - "accessed": { - "date-parts": [ - [ - "2017", - 5, - 23 - ] - ] - }, - "id": "iPIJrVVs" - }, - "citation_id": "iPIJrVVs" - }, - "arxiv:1702.02540": { - "source": "arxiv", - "identifer": "1702.02540", - "standard_citation": "arxiv:1702.02540", - "bibtex": "@article{10ViHstXn,\n abstract = {Although deep learning models have proven effective at solving problems in\nnatural language processing, the mechanism by which they come to their\nconclusions is often unclear. As a result, these models are generally treated\nas black boxes, yielding no insight of the underlying learned patterns. In this\npaper we consider Long Short Term Memory networks (LSTMs) and demonstrate a new\napproach for tracking the importance of a given input to the LSTM for a given\noutput. By identifying consistently important patterns of words, we are able to\ndistill state of the art LSTMs on sentiment analysis and question answering\ninto a set of representative phrases. This representation is then\nquantitatively validated by using the extracted phrases to construct a simple, rule-based classifier which approximates the output of the LSTM.},\n archiveprefix = {arXiv},\n author = {W. James Murdoch and Arthur Szlam},\n eprint = {1702.02540v2},\n file = {1702.02540v2.pdf},\n month = {Feb},\n primaryclass = {cs.CL},\n title = {Automatic Rule Extraction from Long Short Term Memory Networks},\n url = {https://arxiv.org/abs/1702.02540v2},\n year = {2017}\n}\n\n", - "citation_id": "10ViHstXn" - }, - "doi:10.1109/embc.2016.7591355": { - "source": "doi", - "identifer": "10.1109/embc.2016.7591355", - "standard_citation": "doi:10.1109/embc.2016.7591355", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 11 - ] - ], - "date-time": "2017-09-11T12:02:15Z", - "timestamp": 1505131335022 - }, - "reference-count": 18, - "publisher": "IEEE", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 8 - ] - ] - }, - "DOI": "10.1109/embc.2016.7591355", - "type": "paper-conference", - "created": { - "date-parts": [ - [ - 2016, - 10, - 20 - ] - ], - "date-time": "2016-10-20T20:57:43Z", - "timestamp": 1476997063000 - }, - "source": "Crossref", - "is-referenced-by-count": 1, - "title": "Optimal medication dosing from suboptimal clinical examples: A deep reinforcement learning approach", - "prefix": "10.1109", - "author": [ - { - "given": "Shamim", - "family": "Nemati", - "affiliation": [] - }, - { - "given": "Mohammad M.", - "family": "Ghassemi", - "affiliation": [] - }, - { - "given": "Gari D.", - "family": "Clifford", - "affiliation": [] - } - ], - "member": "263", - "container-title": "2016 38th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC)", - "original-title": [], - "link": [ - { - "URL": "http://xplorestaging.ieee.org/ielx7/7580725/7590615/07591355.pdf?arnumber=7591355", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 11, - 2 - ] - ], - "date-time": "2016-11-02T07:01:33Z", - "timestamp": 1478070093000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 8 - ] - ] - }, - "references-count": 18, - "URL": "https://doi.org/10.1109/embc.2016.7591355", - "relation": {}, - "id": "16OQvsRqJ" - }, - "citation_id": "16OQvsRqJ" - }, - "url:https://ai.stanford.edu/~ang/papers/icml11-MultimodalDeepLearning.pdf": { - "source": "url", - "identifer": "https://ai.stanford.edu/~ang/papers/icml11-MultimodalDeepLearning.pdf", - "standard_citation": "url:https://ai.stanford.edu/~ang/papers/icml11-MultimodalDeepLearning.pdf", - "citeproc": { - "type": "paper-conference", - "title": "Multimodal Deep Learning", - "container-title": "Proceedings of the 28th International Conference on Machine Learning", - "source": "Google Scholar", - "URL": "https://ccrma.stanford.edu/~juhan/pubs/NgiamKhoslaKimNamLeeNg2011.pdf", - "author": [ - { - "family": "Ngiam", - "given": "Jiquan" - }, - { - "family": "Khosla", - "given": "Aditya" - }, - { - "family": "Kim", - "given": "Mingyu" - }, - { - "family": "Nam", - "given": "Juhan" - }, - { - "family": "Lee", - "given": "Honglak" - }, - { - "family": "Ng", - "given": "Andrew Y." - } - ], - "issued": { - "date-parts": [ - [ - "2011" - ] - ] - }, - "accessed": { - "date-parts": [ - [ - "2017", - 5, - 23 - ] - ] - }, - "id": "1eN66lwn" - }, - "citation_id": "1eN66lwn" - }, - "arxiv:1412.1897v4": { - "source": "arxiv", - "identifer": "1412.1897v4", - "standard_citation": "arxiv:1412.1897v4", - "bibtex": "@article{1AkF8Wsv7,\n abstract = {Deep neural networks (DNNs) have recently been achieving state-of-the-art\nperformance on a variety of pattern-recognition tasks, most notably visual\nclassification problems. Given that DNNs are now able to classify objects in\nimages with near-human-level performance, questions naturally arise as to what\ndifferences remain between computer and human vision. A recent study revealed\nthat changing an image (e.g. of a lion) in a way imperceptible to humans can\ncause a DNN to label the image as something else entirely (e.g. mislabeling a\nlion a library). Here we show a related result: it is easy to produce images\nthat are completely unrecognizable to humans, but that state-of-the-art DNNs\nbelieve to be recognizable objects with 99.99% confidence (e.g. labeling with\ncertainty that white noise static is a lion). Specifically, we take\nconvolutional neural networks trained to perform well on either the ImageNet or\nMNIST datasets and then find images with evolutionary algorithms or gradient\nascent that DNNs label with high confidence as belonging to each dataset class.\nIt is possible to produce images totally unrecognizable to human eyes that DNNs\nbelieve with near certainty are familiar objects, which we call \"fooling\nimages\" (more generally, fooling examples). Our results shed light on\ninteresting differences between human vision and current DNNs, and raise\nquestions about the generality of DNN computer vision.},\n archiveprefix = {arXiv},\n author = {Anh Nguyen and Jason Yosinski and Jeff Clune},\n eprint = {1412.1897v4},\n file = {1412.1897v4.pdf},\n month = {12},\n primaryclass = {cs.CV},\n title = {Deep Neural Networks are Easily Fooled: High Confidence Predictions for\nUnrecognizable Images},\n url = {https://arxiv.org/abs/1412.1897v4},\n year = {2014}\n}\n\n", - "citation_id": "1AkF8Wsv7" - }, - "doi:10.1007/978-3-319-46723-8_25": { - "source": "doi", - "identifer": "10.1007/978-3-319-46723-8_25", - "standard_citation": "doi:10.1007/978-3-319-46723-8_25", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T04:22:16Z", - "timestamp": 1507004536538 - }, - "publisher-location": "Cham", - "reference-count": 11, - "publisher": "Springer International Publishing", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2016, - 1, - 1 - ] - ], - "date-time": "2016-01-01T00:00:00Z", - "timestamp": 1451606400000 - }, - "delay-in-days": 0, - "content-version": "vor" - }, - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2016, - 1, - 1 - ] - ], - "date-time": "2016-01-01T00:00:00Z", - "timestamp": 1451606400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016 - ] - ] - }, - "DOI": "10.1007/978-3-319-46723-8_25", - "type": "chapter", - "created": { - "date-parts": [ - [ - 2016, - 10, - 1 - ] - ], - "date-time": "2016-10-01T03:01:21Z", - "timestamp": 1475290881000 - }, - "page": "212-220", - "source": "Crossref", - "is-referenced-by-count": 2, - "title": "3D Deep Learning for Multi-modal Imaging-Guided Survival Time Prediction of Brain Tumor Patients", - "prefix": "10.1007", - "author": [ - { - "given": "Dong", - "family": "Nie", - "affiliation": [] - }, - { - "given": "Han", - "family": "Zhang", - "affiliation": [] - }, - { - "given": "Ehsan", - "family": "Adeli", - "affiliation": [] - }, - { - "given": "Luyan", - "family": "Liu", - "affiliation": [] - }, - { - "given": "Dinggang", - "family": "Shen", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2016, - 10, - 2 - ] - ] - }, - "container-title": "Medical Image Computing and Computer-Assisted Intervention – MICCAI 2016", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1007/978-3-319-46723-8_25", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T00:08:45Z", - "timestamp": 1498349325000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016 - ] - ] - }, - "references-count": 11, - "URL": "https://doi.org/10.1007/978-3-319-46723-8_25", - "relation": { - "cites": [] - }, - "id": "18EpaZ7QB" - }, - "citation_id": "18EpaZ7QB" - }, - "url:https://www.nigms.nih.gov/Education/Documents/curiosity.pdf": { - "source": "url", - "identifer": "https://www.nigms.nih.gov/Education/Documents/curiosity.pdf", - "standard_citation": "url:https://www.nigms.nih.gov/Education/Documents/curiosity.pdf", - "citeproc": { - "URL": "https://www.nigms.nih.gov/Education/Documents/curiosity.pdf", - "title": "Curiosity Creates Cures: The Value and Impact of Basic Research", - "issued": { - "date-parts": [ - [ - 2012, - 5 - ] - ] - }, - "author": [ - { - "family": "NIH" - } - ], - "type": "webpage", - "id": "ru0hjGeQ" - }, - "citation_id": "ru0hjGeQ" - }, - "arxiv:1704.07555": { - "source": "arxiv", - "identifer": "1704.07555", - "standard_citation": "arxiv:1704.07555", - "bibtex": "@article{1EayJRsI,\n abstract = {This work introduces a method to tune a sequence-based generative model for\nmolecular de novo design that through augmented episodic likelihood can learn\nto generate structures with certain specified desirable properties. We\ndemonstrate how this model can execute a range of tasks such as generating\nanalogues to a query structure and generating compounds predicted to be active\nagainst a biological target. As a proof of principle, the model is first\ntrained to generate molecules that do not contain sulphur. As a second example, the model is trained to generate analogues to the drug Celecoxib, a technique\nthat could be used for scaffold hopping or library expansion starting from a\nsingle molecule. Finally, when tuning the model towards generating compounds\npredicted to be active against the dopamine receptor type 2, the model\ngenerates structures of which more than 95% are predicted to be active, including experimentally confirmed actives that have not been included in\neither the generative model nor the activity prediction model.},\n archiveprefix = {arXiv},\n author = {Marcus Olivecrona and Thomas Blaschke and Ola Engkvist and Hongming Chen},\n eprint = {1704.07555v2},\n file = {1704.07555v2.pdf},\n month = {Apr},\n primaryclass = {cs.AI},\n title = {Molecular De Novo Design through Deep Reinforcement Learning},\n url = {https://arxiv.org/abs/1704.07555v2},\n year = {2017}\n}\n\n", - "citation_id": "1EayJRsI" - }, - "url:https://openreview.net/forum?id=HkwoSDPgg": { - "source": "url", - "identifer": "https://openreview.net/forum?id=HkwoSDPgg", - "standard_citation": "url:https://openreview.net/forum?id=HkwoSDPgg", - "citeproc": { - "URL": "https://openreview.net/forum?id=HkwoSDPgg", - "title": "Semi-supervised Knowledge Transfer for Deep Learning from Private Training Data", - "issued": { - "date-parts": [ - [ - 2016, - 11, - 2 - ] - ] - }, - "author": [ - { - "family": "Papernot", - "given": "Nicolas" - }, - { - "family": "Abadi", - "given": "Martín" - }, - { - "family": "Erlingsson", - "given": "Úlfar" - }, - { - "family": "Goodfellow", - "given": "Ian" - }, - { - "family": "Talwar", - "given": "Kunal" - } - ], - "greycite-status": "Scanned", - "greycite-scanned": "2017-05-09 00:13:05", - "type": "webpage", - "id": "b8DJ1u6W" - }, - "citation_id": "b8DJ1u6W" - }, - "arxiv:1605.00017": { - "source": "arxiv", - "identifer": "1605.00017", - "standard_citation": "arxiv:1605.00017", - "bibtex": "@article{1TeyWffV,\n abstract = {Since microRNAs (miRNAs) play a crucial role in post-transcriptional gene\nregulation, miRNA identification is one of the most essential problems in\ncomputational biology. miRNAs are usually short in length ranging between 20\nand 23 base pairs. It is thus often difficult to distinguish miRNA-encoding\nsequences from other non-coding RNAs and pseudo miRNAs that have a similar\nlength, and most previous studies have recommended using precursor miRNAs\ninstead of mature miRNAs for robust detection. A great number of conventional\nmachine-learning-based classification methods have been proposed, but they\noften have the serious disadvantage of requiring manual feature engineering, and their performance is limited as well. In this paper, we propose a novel\nmiRNA precursor prediction algorithm, deepMiRGene, based on recurrent neural\nnetworks, specifically long short-term memory networks. deepMiRGene\nautomatically learns suitable features from the data themselves without manual\nfeature engineering and constructs a model that can successfully reflect\nstructural characteristics of precursor miRNAs. For the performance evaluation\nof our approach, we have employed several widely used evaluation metrics on\nthree recent benchmark datasets and verified that deepMiRGene delivered\ncomparable performance among the current state-of-the-art tools.},\n archiveprefix = {arXiv},\n author = {Seunghyun Park and Seonwoo Min and Hyunsoo Choi and Sungroh Yoon},\n eprint = {1605.00017v1},\n file = {1605.00017v1.pdf},\n month = {Apr},\n primaryclass = {cs.LG},\n title = {deepMiRGene: Deep Neural Network based Precursor microRNA Prediction},\n url = {https://arxiv.org/abs/1605.00017v1},\n year = {2016}\n}\n\n", - "citation_id": "1TeyWffV" - }, - "doi:10.1534/g3.116.033654": { - "source": "doi", - "identifer": "10.1534/g3.116.033654", - "standard_citation": "doi:10.1534/g3.116.033654", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 31 - ] - ], - "date-time": "2017-08-31T16:02:17Z", - "timestamp": 1504195337886 - }, - "reference-count": 0, - "publisher": "Genetics Society of America", - "issue": "5", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2017, - 5 - ] - ] - }, - "DOI": "10.1534/g3.116.033654", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2017, - 4, - 9 - ] - ], - "date-time": "2017-04-09T00:20:13Z", - "timestamp": 1491697213000 - }, - "page": "1385-1392", - "source": "Crossref", - "is-referenced-by-count": 1, - "title": "Accurate Classification of Protein Subcellular Localization from High-Throughput Microscopy Images Using Deep Learning", - "prefix": "10.1534", - "volume": "7", - "author": [ - { - "given": "Tanel", - "family": "Pärnamaa", - "affiliation": [] - }, - { - "given": "Leopold", - "family": "Parts", - "affiliation": [] - } - ], - "member": "414", - "published-online": { - "date-parts": [ - [ - 2017, - 4, - 8 - ] - ] - }, - "container-title": "G3: Genes|Genomes|Genetics", - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1534/g3.116.033654", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 8 - ] - ], - "date-time": "2017-06-08T03:10:18Z", - "timestamp": 1496891418000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 4, - 8 - ] - ] - }, - "references-count": 0, - "alternative-id": [ - "10.1534/g3.116.033654" - ], - "URL": "https://doi.org/10.1534/g3.116.033654", - "relation": { - "has-preprint": [ - { - "id-type": "doi", - "id": "10.1101/050757", - "asserted-by": "object" - } - ] - }, - "subject": [ - "Genetics(clinical)", - "Genetics", - "Molecular Biology" - ], - "container-title-short": "G3", - "id": "2a7MHtAx" - }, - "citation_id": "2a7MHtAx" - }, - "doi:10.1021/acs.jcim.6b00355": { - "source": "doi", - "identifer": "10.1021/acs.jcim.6b00355", - "standard_citation": "doi:10.1021/acs.jcim.6b00355", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 19 - ] - ], - "date-time": "2017-09-19T16:02:28Z", - "timestamp": 1505836948897 - }, - "reference-count": 45, - "publisher": "American Chemical Society (ACS)", - "issue": "12", - "funder": [ - { - "DOI": "10.13039/501100006507", - "name": "Fundação Oswaldo Cruz", - "doi-asserted-by": "publisher", - "award": [] - }, - { - "DOI": "10.13039/501100003593", - "name": "Conselho Nacional de Desenvolvimento Científico e Tecnológico", - "doi-asserted-by": "publisher", - "award": [ - "407741/2012-7" - ] - }, - { - "DOI": "10.13039/501100004586", - "name": "Fundação Carlos Chagas Filho de Amparo à Pesquisa do Estado do Rio de Janeiro", - "doi-asserted-by": "publisher", - "award": [ - "E-26/111.401/2013" - ] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 12, - 27 - ] - ] - }, - "DOI": "10.1021/acs.jcim.6b00355", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 11, - 4 - ] - ], - "date-time": "2016-11-04T17:52:43Z", - "timestamp": 1478281963000 - }, - "page": "2495-2506", - "source": "Crossref", - "is-referenced-by-count": 6, - "title": "Boosting Docking-Based Virtual Screening with Deep Learning", - "prefix": "10.1021", - "volume": "56", - "author": [ - { - "given": "Janaina Cruz", - "family": "Pereira", - "affiliation": [] - }, - { - "given": "Ernesto Raúl", - "family": "Caffarena", - "affiliation": [] - }, - { - "given": "Cicero Nogueira", - "family": "dos Santos", - "affiliation": [] - } - ], - "member": "316", - "container-title": "Journal of Chemical Information and Modeling", - "original-title": [], - "link": [ - { - "URL": "http://pubs.acs.org/doi/pdf/10.1021/acs.jcim.6b00355", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 12, - 27 - ] - ], - "date-time": "2016-12-27T09:26:59Z", - "timestamp": 1482830819000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 12, - 27 - ] - ] - }, - "references-count": 45, - "alternative-id": [ - "10.1021/acs.jcim.6b00355" - ], - "URL": "https://doi.org/10.1021/acs.jcim.6b00355", - "relation": {}, - "subject": [ - "General Chemistry", - "General Chemical Engineering", - "Library and Information Sciences", - "Computer Science Applications" - ], - "container-title-short": "J. Chem. Inf. Model.", - "id": "Gue0c5Gb" - }, - "citation_id": "Gue0c5Gb" - }, - "doi:10.1007/978-3-319-40126-3_2": { - "source": "doi", - "identifer": "10.1007/978-3-319-40126-3_2", - "standard_citation": "doi:10.1007/978-3-319-40126-3_2", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T11:43:15Z", - "timestamp": 1502278995219 - }, - "publisher-location": "Cham", - "reference-count": 27, - "publisher": "Springer International Publishing", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2016, - 1, - 1 - ] - ], - "date-time": "2016-01-01T00:00:00Z", - "timestamp": 1451606400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016 - ] - ] - }, - "DOI": "10.1007/978-3-319-40126-3_2", - "type": "chapter", - "created": { - "date-parts": [ - [ - 2016, - 5, - 31 - ] - ], - "date-time": "2016-05-31T07:50:23Z", - "timestamp": 1464681023000 - }, - "page": "13-22", - "source": "Crossref", - "is-referenced-by-count": 1, - "title": "Virtual Screening: A Challenge for Deep Learning", - "prefix": "10.1007", - "author": [ - { - "given": "Javier", - "family": "Pérez-Sianes", - "affiliation": [] - }, - { - "given": "Horacio", - "family": "Pérez-Sánchez", - "affiliation": [] - }, - { - "given": "Fernando", - "family": "Díaz", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2016, - 6, - 1 - ] - ] - }, - "container-title": "Advances in Intelligent Systems and Computing", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1007/978-3-319-40126-3_2", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T15:23:05Z", - "timestamp": 1498317785000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016 - ] - ] - }, - "references-count": 27, - "URL": "https://doi.org/10.1007/978-3-319-40126-3_2", - "relation": { - "cites": [] - }, - "id": "1DTUK3YyI" - }, - "citation_id": "1DTUK3YyI" - }, - "doi:10.1101/092890": { - "source": "doi", - "identifer": "10.1101/092890", - "standard_citation": "doi:10.1101/092890", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T05:20:36Z", - "timestamp": 1502342436194 - }, - "posted": { - "date-parts": [ - [ - 2016, - 12, - 14 - ] - ] - }, - "group-title": "Genomics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2016, - 12, - 21 - ] - ] - }, - "abstract": "Next-generation sequencing (NGS) is a rapidly evolving set of technologies that can be used to determine the sequence of an individual's genome by calling genetic variants present in an individual using billions of short, errorful sequence reads. Despite more than a decade of effort and thousands of dedicated researchers, the hand-crafted and parameterized statistical models used for variant calling still produce thousands of errors and missed variants in each genome. Here we show that a deep convolutional neural network can call genetic variation in aligned next-generation sequencing read data by learning statistical relationships (likelihoods) between images of read pileups around putative variant sites and ground-truth genotype calls. This approach, called DeepVariant, outperforms existing tools, even winning the \"highest performance\" award for SNPs in a FDA-administered variant calling challenge. The learned model generalizes across genome builds and even to other species, allowing non-human sequencing projects to benefit from the wealth of human ground truth data. We further show that, unlike existing tools which perform well on only a specific technology, DeepVariant can learn to call variants in a variety of sequencing technologies and experimental designs, from deep whole genomes from 10X Genomics to Ion Ampliseq exomes. DeepVariant represents a significant step from expert-driven statistical modeling towards more automatic deep learning approaches for developing software to interpret biological instrumentation data.", - "DOI": "10.1101/092890", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2016, - 12, - 15 - ] - ], - "date-time": "2016-12-15T06:10:40Z", - "timestamp": 1481782240000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Creating a universal SNP and small indel variant caller with deep neural networks", - "prefix": "10.1101", - "author": [ - { - "given": "Ryan", - "family": "Poplin", - "affiliation": [] - }, - { - "given": "Dan", - "family": "Newburger", - "affiliation": [] - }, - { - "given": "Jojo", - "family": "Dijamco", - "affiliation": [] - }, - { - "given": "Nam", - "family": "Nguyen", - "affiliation": [] - }, - { - "given": "Dion", - "family": "Loy", - "affiliation": [] - }, - { - "given": "Sam S.", - "family": "Gross", - "affiliation": [] - }, - { - "given": "Cory Y.", - "family": "McLean", - "affiliation": [] - }, - { - "given": "Mark A.", - "family": "DePristo", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/092890", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 1, - 4 - ] - ], - "date-time": "2017-01-04T06:17:53Z", - "timestamp": 1483510673000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 12, - 14 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/092890", - "relation": {}, - "id": "FVfZESYP" - }, - "citation_id": "FVfZESYP" - }, - "doi:10.1016/j.procs.2016.07.014": { - "source": "doi", - "identifer": "10.1016/j.procs.2016.07.014", - "standard_citation": "doi:10.1016/j.procs.2016.07.014", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T02:22:22Z", - "timestamp": 1506997342204 - }, - "reference-count": 20, - "publisher": "Elsevier BV", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2016, - 1, - 1 - ] - ], - "date-time": "2016-01-01T00:00:00Z", - "timestamp": 1451606400000 - }, - "delay-in-days": 0, - "content-version": "tdm" - }, - { - "URL": "http://creativecommons.org/licenses/by-nc-nd/4.0/", - "start": { - "date-parts": [ - [ - 2016, - 7, - 12 - ] - ], - "date-time": "2016-07-12T00:00:00Z", - "timestamp": 1468281600000 - }, - "delay-in-days": 193, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "elsevier.com", - "sciencedirect.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2016 - ] - ] - }, - "DOI": "10.1016/j.procs.2016.07.014", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 7, - 25 - ] - ], - "date-time": "2016-07-25T18:00:18Z", - "timestamp": 1469469618000 - }, - "page": "200-205", - "update-policy": "http://dx.doi.org/10.1016/elsevier_cm_policy", - "source": "Crossref", - "is-referenced-by-count": 5, - "title": "Convolutional Neural Networks for Diabetic Retinopathy", - "prefix": "10.1016", - "volume": "90", - "author": [ - { - "given": "Harry", - "family": "Pratt", - "affiliation": [] - }, - { - "given": "Frans", - "family": "Coenen", - "affiliation": [] - }, - { - "given": "Deborah M.", - "family": "Broadbent", - "affiliation": [] - }, - { - "given": "Simon P.", - "family": "Harding", - "affiliation": [] - }, - { - "given": "Yalin", - "family": "Zheng", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Procedia Computer Science", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S1877050916311929?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S1877050916311929?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T19:07:19Z", - "timestamp": 1498331239000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016 - ] - ] - }, - "references-count": 20, - "alternative-id": [ - "S1877050916311929" - ], - "URL": "https://doi.org/10.1016/j.procs.2016.07.014", - "relation": {}, - "container-title-short": "Procedia Computer Science", - "assertion": [ - { - "value": "Elsevier", - "name": "publisher", - "label": "This article is maintained by" - }, - { - "value": "Convolutional Neural Networks for Diabetic Retinopathy", - "name": "articletitle", - "label": "Article Title" - }, - { - "value": "Procedia Computer Science", - "name": "journaltitle", - "label": "Journal Title" - }, - { - "value": "http://dx.doi.org/10.1016/j.procs.2016.07.014", - "name": "articlelink", - "label": "CrossRef DOI link to publisher maintained version" - }, - { - "value": "article", - "name": "content_type", - "label": "Content Type" - }, - { - "value": "© 2016 The Author(s). Published by Elsevier B.V.", - "name": "copyright", - "label": "Copyright" - } - ], - "id": "ayTsooEM" - }, - "citation_id": "ayTsooEM" - }, - "doi:10.1371/journal.pcbi.1005403": { - "source": "doi", - "identifer": "10.1371/journal.pcbi.1005403", - "standard_citation": "doi:10.1371/journal.pcbi.1005403", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 11 - ] - ], - "date-time": "2017-08-11T00:13:46Z", - "timestamp": 1502410426834 - }, - "update-to": [ - { - "updated": { - "date-parts": [ - [ - 2017, - 3, - 10 - ] - ], - "date-time": "2017-03-10T00:00:00Z", - "timestamp": 1489104000000 - }, - "DOI": "10.1371/journal.pcbi.1005403", - "type": "new_version", - "label": "New version" - } - ], - "reference-count": 42, - "publisher": "Public Library of Science (PLoS)", - "issue": "2", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2017, - 2, - 24 - ] - ], - "date-time": "2017-02-24T00:00:00Z", - "timestamp": 1487894400000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/501100001809", - "name": "National Natural Science Foundation of China", - "doi-asserted-by": "publisher", - "award": [ - "31301096", - "31329003" - ] - } - ], - "content-domain": { - "domain": [ - "www.ploscompbiol.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pcbi.1005403", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2017, - 2, - 24 - ] - ], - "date-time": "2017-02-24T18:30:22Z", - "timestamp": 1487961022000 - }, - "page": "e1005403", - "update-policy": "http://dx.doi.org/10.1371/journal.pcbi.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 2, - "title": "Imputation for transcription factor binding predictions based on deep learning", - "prefix": "10.1371", - "volume": "13", - "author": [ - { - "given": "Qian", - "family": "Qin", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-4180-3323", - "authenticated-orcid": true, - "given": "Jianxing", - "family": "Feng", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2017, - 2, - 24 - ] - ] - }, - "container-title": "PLOS Computational Biology", - "original-title": [], - "link": [ - { - "URL": "http://dx.plos.org/10.1371/journal.pcbi.1005403", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T11:21:26Z", - "timestamp": 1498389686000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Ilya", - "family": "Ioshikhes", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 2, - 24 - ] - ] - }, - "references-count": 42, - "URL": "https://doi.org/10.1371/journal.pcbi.1005403", - "relation": { - "cites": [] - }, - "subject": [ - "Ecology", - "Modelling and Simulation", - "Computational Theory and Mathematics", - "Genetics", - "Ecology, Evolution, Behavior and Systematics", - "Molecular Biology", - "Cellular and Molecular Neuroscience" - ], - "container-title-short": "PLoS Comput Biol", - "id": "Qbtqlmhf" - }, - "citation_id": "Qbtqlmhf" - }, - "doi:10.1101/110668": { - "source": "doi", - "identifer": "10.1101/110668", - "standard_citation": "doi:10.1101/110668", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 6 - ] - ], - "date-time": "2017-09-06T21:42:16Z", - "timestamp": 1504734136160 - }, - "posted": { - "date-parts": [ - [ - 2017, - 2, - 21 - ] - ] - }, - "group-title": "Genomics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2017, - 2, - 21 - ] - ] - }, - "abstract": "Organizing single cells along a developmental trajectory has emerged as a powerful tool for understanding how gene regulation governs cell fate decisions. However, learning the structure of complex single-cell trajectories with two or more branches remains a challenging computational problem. We present Monocle 2, which uses reversed graph embedding to reconstruct single-cell trajectories in a fully unsupervised manner. Monocle 2 learns an explicit principal graph to describe the data, greatly improving the robustness and accuracy of its trajectories compared to other algorithms. Monocle 2 uncovered a new, alternative cell fate in what we previously reported to be a linear trajectory for differentiating myoblasts. We also reconstruct branched trajectories for two studies of blood development, and show that loss of function mutations in key lineage transcription factors diverts cells to alternative branches on the a trajectory. Monocle 2 is thus a powerful tool for analyzing cell fate decisions with single-cell genomics.", - "DOI": "10.1101/110668", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2017, - 2, - 22 - ] - ], - "date-time": "2017-02-22T06:10:18Z", - "timestamp": 1487743818000 - }, - "source": "Crossref", - "is-referenced-by-count": 2, - "title": "Reversed graph embedding resolves complex single-cell developmental trajectories", - "prefix": "10.1101", - "author": [ - { - "given": "Xiaojie", - "family": "Qiu", - "affiliation": [] - }, - { - "given": "Qi", - "family": "Mao", - "affiliation": [] - }, - { - "given": "Ying", - "family": "Tang", - "affiliation": [] - }, - { - "given": "Li", - "family": "Wang", - "affiliation": [] - }, - { - "given": "Raghav", - "family": "Chawla", - "affiliation": [] - }, - { - "given": "Hannah", - "family": "Pliner", - "affiliation": [] - }, - { - "given": "Cole", - "family": "Trapnell", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/110668", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 2, - 22 - ] - ], - "date-time": "2017-02-22T06:10:32Z", - "timestamp": 1487743832000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 2, - 21 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/110668", - "relation": {}, - "id": "Oljj2W96" - }, - "citation_id": "Oljj2W96" - }, - "doi:10.1145/1721654.1721672": { - "source": "doi", - "identifer": "10.1145/1721654.1721672", - "standard_citation": "doi:10.1145/1721654.1721672", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T10:02:18Z", - "timestamp": 1507111338758 - }, - "reference-count": 0, - "publisher": "Association for Computing Machinery (ACM)", - "issue": "4", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2010, - 4, - 1 - ] - ] - }, - "DOI": "10.1145/1721654.1721672", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2010, - 3, - 30 - ] - ], - "date-time": "2010-03-30T12:32:23Z", - "timestamp": 1269952343000 - }, - "page": "50", - "source": "Crossref", - "is-referenced-by-count": 2850, - "title": "A view of cloud computing", - "prefix": "10.1145", - "volume": "53", - "author": [ - { - "given": "Michael", - "family": "Armbrust", - "affiliation": [] - }, - { - "given": "Ion", - "family": "Stoica", - "affiliation": [] - }, - { - "given": "Matei", - "family": "Zaharia", - "affiliation": [] - }, - { - "given": "Armando", - "family": "Fox", - "affiliation": [] - }, - { - "given": "Rean", - "family": "Griffith", - "affiliation": [] - }, - { - "given": "Anthony D.", - "family": "Joseph", - "affiliation": [] - }, - { - "given": "Randy", - "family": "Katz", - "affiliation": [] - }, - { - "given": "Andy", - "family": "Konwinski", - "affiliation": [] - }, - { - "given": "Gunho", - "family": "Lee", - "affiliation": [] - }, - { - "given": "David", - "family": "Patterson", - "affiliation": [] - }, - { - "given": "Ariel", - "family": "Rabkin", - "affiliation": [] - } - ], - "member": "320", - "container-title": "Communications of the ACM", - "original-title": [], - "link": [ - { - "URL": "http://dl.acm.org/ft_gateway.cfm?id=1721672&ftid=757661&dwn=1", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 12, - 7 - ] - ], - "date-time": "2016-12-07T04:30:42Z", - "timestamp": 1481085042000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2010, - 4, - 1 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1145/1721654.1721672", - "relation": {}, - "subject": [ - "General Computer Science" - ], - "container-title-short": "Commun. ACM", - "id": "ObFN78yp" - }, - "citation_id": "ObFN78yp" - }, - "arxiv:1612.02751": { - "source": "arxiv", - "identifer": "1612.02751", - "standard_citation": "arxiv:1612.02751", - "bibtex": "@article{bNBiIiTt,\n abstract = {Computational approaches to drug discovery can reduce the time and cost\nassociated with experimental assays and enable the screening of novel\nchemotypes. Structure-based drug design methods rely on scoring functions to\nrank and predict binding affinities and poses. The ever-expanding amount of\nprotein-ligand binding and structural data enables the use of deep machine\nlearning techniques for protein-ligand scoring.\nWe describe convolutional neural network (CNN) scoring functions that take as\ninput a comprehensive 3D representation of a protein-ligand interaction. A CNN\nscoring function automatically learns the key features of protein-ligand\ninteractions that correlate with binding. We train and optimize our CNN scoring\nfunctions to discriminate between correct and incorrect binding poses and known\nbinders and non-binders. We find that our CNN scoring function outperforms the\nAutoDock Vina scoring function when ranking poses both for pose prediction and\nvirtual screening.},\n archiveprefix = {arXiv},\n author = {Matthew Ragoza and Joshua Hochuli and Elisa Idrobo and Jocelyn Sunseri and David Ryan Koes},\n eprint = {1612.02751v1},\n file = {1612.02751v1.pdf},\n month = {12},\n primaryclass = {stat.ML},\n title = {Protein-Ligand Scoring with Convolutional Neural Networks},\n url = {https://arxiv.org/abs/1612.02751v1},\n year = {2016}\n}\n\n", - "citation_id": "bNBiIiTt" - }, - "doi:10.1145/1553374.1553486": { - "source": "doi", - "identifer": "10.1145/1553374.1553486", - "standard_citation": "doi:10.1145/1553374.1553486", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 23 - ] - ], - "date-time": "2017-09-23T18:22:11Z", - "timestamp": 1506190931474 - }, - "publisher-location": "New York, New York, USA", - "reference-count": 0, - "publisher": "ACM Press", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2009 - ] - ] - }, - "DOI": "10.1145/1553374.1553486", - "type": "paper-conference", - "created": { - "date-parts": [ - [ - 2009, - 6, - 16 - ] - ], - "date-time": "2009-06-16T13:34:36Z", - "timestamp": 1245159276000 - }, - "source": "Crossref", - "is-referenced-by-count": 86, - "title": "Large-scale deep unsupervised learning using graphics processors", - "prefix": "10.1145", - "author": [ - { - "given": "Rajat", - "family": "Raina", - "affiliation": [] - }, - { - "given": "Anand", - "family": "Madhavan", - "affiliation": [] - }, - { - "given": "Andrew Y.", - "family": "Ng", - "affiliation": [] - } - ], - "member": "320", - "container-title": "Proceedings of the 26th Annual International Conference on Machine Learning - ICML '09", - "original-title": [], - "link": [ - { - "URL": "http://dl.acm.org/ft_gateway.cfm?id=1553486&ftid=640570&dwn=1", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 12, - 6 - ] - ], - "date-time": "2016-12-06T20:55:09Z", - "timestamp": 1481057709000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2009 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1145/1553374.1553486", - "relation": {}, - "id": "F3e4wfzQ" - }, - "citation_id": "F3e4wfzQ" - }, - "arxiv:1502.02072": { - "source": "arxiv", - "identifer": "1502.02072", - "standard_citation": "arxiv:1502.02072", - "bibtex": "@article{yAoN5gTU,\n abstract = {Massively multitask neural architectures provide a learning framework for\ndrug discovery that synthesizes information from many distinct biological\nsources. To train these architectures at scale, we gather large amounts of data\nfrom public sources to create a dataset of nearly 40 million measurements\nacross more than 200 biological targets. We investigate several aspects of the\nmultitask framework by performing a series of empirical studies and obtain some\ninteresting results: (1) massively multitask networks obtain predictive\naccuracies significantly better than single-task methods, (2) the predictive\npower of multitask networks improves as additional tasks and data are added, (3) the total amount of data and the total number of tasks both contribute\nsignificantly to multitask improvement, and (4) multitask networks afford\nlimited transferability to tasks not in the training set. Our results\nunderscore the need for greater data sharing and further algorithmic innovation\nto accelerate the drug discovery process.},\n archiveprefix = {arXiv},\n author = {Bharath Ramsundar and Steven Kearnes and Patrick Riley and Dale Webster and David Konerding and Vijay Pande},\n eprint = {1502.02072v1},\n file = {1502.02072v1.pdf},\n month = {Feb},\n primaryclass = {stat.ML},\n title = {Massively Multitask Networks for Drug Discovery},\n url = {https://arxiv.org/abs/1502.02072v1},\n year = {2015}\n}\n\n", - "citation_id": "yAoN5gTU" - }, - "arxiv:1602.04938": { - "source": "arxiv", - "identifer": "1602.04938", - "standard_citation": "arxiv:1602.04938", - "bibtex": "@article{QwXSJhr0,\n abstract = {Despite widespread adoption, machine learning models remain mostly black\nboxes. Understanding the reasons behind predictions is, however, quite\nimportant in assessing trust, which is fundamental if one plans to take action\nbased on a prediction, or when choosing whether to deploy a new model. Such\nunderstanding also provides insights into the model, which can be used to\ntransform an untrustworthy model or prediction into a trustworthy one. In this\nwork, we propose LIME, a novel explanation technique that explains the\npredictions of any classifier in an interpretable and faithful manner, by\nlearning an interpretable model locally around the prediction. We also propose\na method to explain models by presenting representative individual predictions\nand their explanations in a non-redundant way, framing the task as a submodular\noptimization problem. We demonstrate the flexibility of these methods by\nexplaining different models for text (e.g. random forests) and image\nclassification (e.g. neural networks). We show the utility of explanations via\nnovel experiments, both simulated and with human subjects, on various scenarios\nthat require trust: deciding if one should trust a prediction, choosing between\nmodels, improving an untrustworthy classifier, and identifying why a classifier\nshould not be trusted.},\n archiveprefix = {arXiv},\n author = {Marco Tulio Ribeiro and Sameer Singh and Carlos Guestrin},\n eprint = {1602.04938v3},\n file = {1602.04938v3.pdf},\n month = {Feb},\n primaryclass = {cs.LG},\n title = {\"Why Should I Trust You?\": Explaining the Predictions of Any Classifier},\n url = {https://arxiv.org/abs/1602.04938v3},\n year = {2016}\n}\n\n", - "citation_id": "QwXSJhr0" - }, - "doi:10.1021/ci100050t": { - "source": "doi", - "identifer": "10.1021/ci100050t", - "standard_citation": "doi:10.1021/ci100050t", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T13:42:15Z", - "timestamp": 1507038135944 - }, - "reference-count": 0, - "publisher": "American Chemical Society (ACS)", - "issue": "5", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2010, - 5, - 24 - ] - ] - }, - "DOI": "10.1021/ci100050t", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2010, - 4, - 28 - ] - ], - "date-time": "2010-04-28T16:44:59Z", - "timestamp": 1272473099000 - }, - "page": "742-754", - "source": "Crossref", - "is-referenced-by-count": 674, - "title": "Extended-Connectivity Fingerprints", - "prefix": "10.1021", - "volume": "50", - "author": [ - { - "given": "David", - "family": "Rogers", - "affiliation": [] - }, - { - "given": "Mathew", - "family": "Hahn", - "affiliation": [] - } - ], - "member": "316", - "container-title": "Journal of Chemical Information and Modeling", - "original-title": [], - "link": [ - { - "URL": "http://pubs.acs.org/doi/pdf/10.1021/ci100050t", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 9, - 1 - ] - ], - "date-time": "2016-09-01T12:23:18Z", - "timestamp": 1472732598000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2010, - 5, - 24 - ] - ] - }, - "references-count": 0, - "alternative-id": [ - "10.1021/ci100050t" - ], - "URL": "https://doi.org/10.1021/ci100050t", - "relation": {}, - "subject": [ - "General Chemistry", - "General Chemical Engineering", - "Library and Information Sciences", - "Computer Science Applications" - ], - "container-title-short": "J. Chem. Inf. Model.", - "id": "QnZ7V9Rd" - }, - "citation_id": "QnZ7V9Rd" - }, - "url:https://openreview.net/pdf?id=Sk-oDY9ge": { - "source": "url", - "identifer": "https://openreview.net/pdf?id=Sk-oDY9ge", - "standard_citation": "url:https://openreview.net/pdf?id=Sk-oDY9ge", - "citeproc": { - "type": "article-journal", - "title": "Diet Networks: Thin Parameters for Fat Genomics", - "container-title": "International Conference on Learning Representations 2017", - "source": "openreview.net", - "URL": "https://openreview.net/forum?id=Sk-oDY9ge¬eId=Sk-oDY9ge", - "shortTitle": "Diet Networks", - "journalAbbreviation": "ICLR 2017", - "author": [ - { - "family": "Romero", - "given": "Adriana" - }, - { - "family": "Carrier", - "given": "Pierre Luc" - }, - { - "family": "Erraqabi", - "given": "Akram" - }, - { - "family": "Sylvain", - "given": "Tristan" - }, - { - "family": "Auvolat", - "given": "Alex" - }, - { - "family": "Dejoie", - "given": "Etienne" - }, - { - "family": "Legault", - "given": "Marc-André" - }, - { - "family": "Dubé", - "given": "Marie-Pierre" - }, - { - "family": "Hussin", - "given": "Julie G." - }, - { - "family": "Bengio", - "given": "Yoshua" - } - ], - "issued": { - "date-parts": [ - [ - "2016", - 11, - 4 - ] - ] - }, - "accessed": { - "date-parts": [ - [ - "2017", - 5, - 23 - ] - ] - }, - "id": "15JUKBg9y" - }, - "citation_id": "15JUKBg9y" - }, - "doi:10.1016/j.cell.2015.09.054": { - "source": "doi", - "identifer": "10.1016/j.cell.2015.09.054", - "standard_citation": "doi:10.1016/j.cell.2015.09.054", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 18 - ] - ], - "date-time": "2017-09-18T17:42:13Z", - "timestamp": 1505756533615 - }, - "reference-count": 46, - "publisher": "Elsevier BV", - "issue": "3", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2015, - 10, - 1 - ] - ], - "date-time": "2015-10-01T00:00:00Z", - "timestamp": 1443657600000 - }, - "delay-in-days": 0, - "content-version": "tdm" - }, - { - "URL": "http://www.elsevier.com/open-access/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2016, - 10, - 22 - ] - ], - "date-time": "2016-10-22T00:00:00Z", - "timestamp": 1477094400000 - }, - "delay-in-days": 387, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "cell.com", - "elsevier.com", - "sciencedirect.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2015, - 10 - ] - ] - }, - "DOI": "10.1016/j.cell.2015.09.054", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 10, - 23 - ] - ], - "date-time": "2015-10-23T02:13:44Z", - "timestamp": 1445566424000 - }, - "page": "698-711", - "update-policy": "http://dx.doi.org/10.1016/elsevier_cm_policy", - "source": "Crossref", - "is-referenced-by-count": 18, - "title": "Learning the Sequence Determinants of Alternative Splicing from Millions of Random Sequences", - "prefix": "10.1016", - "volume": "163", - "author": [ - { - "given": "Alexander B.", - "family": "Rosenberg", - "affiliation": [] - }, - { - "given": "Rupali P.", - "family": "Patwardhan", - "affiliation": [] - }, - { - "given": "Jay", - "family": "Shendure", - "affiliation": [] - }, - { - "given": "Georg", - "family": "Seelig", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Cell", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S0092867415012714?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S0092867415012714?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 23 - ] - ], - "date-time": "2017-06-23T21:50:54Z", - "timestamp": 1498254654000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 10 - ] - ] - }, - "references-count": 46, - "alternative-id": [ - "S0092867415012714" - ], - "URL": "https://doi.org/10.1016/j.cell.2015.09.054", - "relation": {}, - "subject": [ - "General Biochemistry, Genetics and Molecular Biology" - ], - "container-title-short": "Cell", - "assertion": [ - { - "value": "Elsevier", - "name": "publisher", - "label": "This article is maintained by" - }, - { - "value": "Learning the Sequence Determinants of Alternative Splicing from Millions of Random Sequences", - "name": "articletitle", - "label": "Article Title" - }, - { - "value": "Cell", - "name": "journaltitle", - "label": "Journal Title" - }, - { - "value": "http://dx.doi.org/10.1016/j.cell.2015.09.054", - "name": "articlelink", - "label": "CrossRef DOI link to publisher maintained version" - }, - { - "value": "article", - "name": "content_type", - "label": "Content Type" - }, - { - "value": "Copyright © 2015 Elsevier Inc. All rights reserved.", - "name": "copyright", - "label": "Copyright" - } - ], - "id": "mlqKTlZY" - }, - "citation_id": "mlqKTlZY" - }, - "doi:10.1109/tmi.2015.2482920": { - "source": "doi", - "identifer": "10.1109/tmi.2015.2482920", - "standard_citation": "doi:10.1109/tmi.2015.2482920", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T04:02:42Z", - "timestamp": 1507003362713 - }, - "reference-count": 60, - "publisher": "Institute of Electrical and Electronics Engineers (IEEE)", - "issue": "5", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 5 - ] - ] - }, - "DOI": "10.1109/tmi.2015.2482920", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 9, - 28 - ] - ], - "date-time": "2015-09-28T18:37:52Z", - "timestamp": 1443465472000 - }, - "page": "1170-1181", - "source": "Crossref", - "is-referenced-by-count": 32, - "title": "Improving Computer-Aided Detection Using_newlineConvolutional Neural Networks and Random View Aggregation", - "prefix": "10.1109", - "volume": "35", - "author": [ - { - "ORCID": "http://orcid.org/0000-0002-3662-8743", - "authenticated-orcid": false, - "given": "Holger R.", - "family": "Roth", - "affiliation": [] - }, - { - "given": "Le", - "family": "Lu", - "affiliation": [] - }, - { - "given": "Jiamin", - "family": "Liu", - "affiliation": [] - }, - { - "given": "Jianhua", - "family": "Yao", - "affiliation": [] - }, - { - "given": "Ari", - "family": "Seff", - "affiliation": [] - }, - { - "given": "Kevin", - "family": "Cherry", - "affiliation": [] - }, - { - "given": "Lauren", - "family": "Kim", - "affiliation": [] - }, - { - "given": "Ronald M.", - "family": "Summers", - "affiliation": [] - } - ], - "member": "263", - "container-title": "IEEE Transactions on Medical Imaging", - "original-title": [], - "link": [ - { - "URL": "http://xplorestaging.ieee.org/ielx7/42/7463083/07279156.pdf?arnumber=7279156", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 9, - 29 - ] - ], - "date-time": "2016-09-29T20:08:27Z", - "timestamp": 1475179707000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 5 - ] - ] - }, - "references-count": 60, - "URL": "https://doi.org/10.1109/tmi.2015.2482920", - "relation": {}, - "subject": [ - "Electrical and Electronic Engineering", - "Radiological and Ultrasound Technology", - "Software", - "Computer Science Applications" - ], - "container-title-short": "IEEE Trans. Med. Imaging", - "id": "KseoWN2w" - }, - "citation_id": "KseoWN2w" - }, - "doi:10.1007/s11263-015-0816-y": { - "source": "doi", - "identifer": "10.1007/s11263-015-0816-y", - "standard_citation": "doi:10.1007/s11263-015-0816-y", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T09:22:22Z", - "timestamp": 1507108942046 - }, - "reference-count": 102, - "publisher": "Springer Nature", - "issue": "3", - "content-domain": { - "domain": [ - "link.springer.com" - ], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015, - 12 - ] - ] - }, - "DOI": "10.1007/s11263-015-0816-y", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 4, - 10 - ] - ], - "date-time": "2015-04-10T04:47:45Z", - "timestamp": 1428641265000 - }, - "page": "211-252", - "update-policy": "http://dx.doi.org/10.1007/springer_crossmark_policy", - "source": "Crossref", - "is-referenced-by-count": 1075, - "title": "ImageNet Large Scale Visual Recognition Challenge", - "prefix": "10.1007", - "volume": "115", - "author": [ - { - "given": "Olga", - "family": "Russakovsky", - "affiliation": [] - }, - { - "given": "Jia", - "family": "Deng", - "affiliation": [] - }, - { - "given": "Hao", - "family": "Su", - "affiliation": [] - }, - { - "given": "Jonathan", - "family": "Krause", - "affiliation": [] - }, - { - "given": "Sanjeev", - "family": "Satheesh", - "affiliation": [] - }, - { - "given": "Sean", - "family": "Ma", - "affiliation": [] - }, - { - "given": "Zhiheng", - "family": "Huang", - "affiliation": [] - }, - { - "given": "Andrej", - "family": "Karpathy", - "affiliation": [] - }, - { - "given": "Aditya", - "family": "Khosla", - "affiliation": [] - }, - { - "given": "Michael", - "family": "Bernstein", - "affiliation": [] - }, - { - "given": "Alexander C.", - "family": "Berg", - "affiliation": [] - }, - { - "given": "Li", - "family": "Fei-Fei", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2015, - 4, - 11 - ] - ] - }, - "container-title": "International Journal of Computer Vision", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1007/s11263-015-0816-y", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 23 - ] - ], - "date-time": "2017-06-23T09:04:48Z", - "timestamp": 1498208688000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 4, - 11 - ] - ] - }, - "references-count": 102, - "alternative-id": [ - "816" - ], - "URL": "https://doi.org/10.1007/s11263-015-0816-y", - "relation": { - "cites": [] - }, - "subject": [ - "Software", - "Artificial Intelligence", - "Computer Vision and Pattern Recognition" - ], - "container-title-short": "Int J Comput Vis", - "id": "cBVeXnZx" - }, - "citation_id": "cBVeXnZx" - }, - "arxiv:1506.06438": { - "source": "arxiv", - "identifer": "1506.06438", - "standard_citation": "arxiv:1506.06438", - "bibtex": "@article{w6CoVmFK,\n abstract = {Stochastic gradient descent (SGD) is a ubiquitous algorithm for a variety of\nmachine learning problems. Researchers and industry have developed several\ntechniques to optimize SGD's runtime performance, including asynchronous\nexecution and reduced precision. Our main result is a martingale-based analysis\nthat enables us to capture the rich noise models that may arise from such\ntechniques. Specifically, we use our new analysis in three ways: (1) we derive\nconvergence rates for the convex case (Hogwild!) with relaxed assumptions on\nthe sparsity of the problem; (2) we analyze asynchronous SGD algorithms for\nnon-convex matrix problems including matrix completion; and (3) we design and\nanalyze an asynchronous SGD algorithm, called Buckwild!, that uses\nlower-precision arithmetic. We show experimentally that our algorithms run\nefficiently for a variety of problems on modern hardware.},\n archiveprefix = {arXiv},\n author = {Christopher De Sa and Ce Zhang and Kunle Olukotun and Christopher Ré},\n eprint = {1506.06438v2},\n file = {1506.06438v2.pdf},\n month = {Jun},\n primaryclass = {cs.LG},\n title = {Taming the Wild: A Unified Analysis of Hogwild!-Style Algorithms},\n url = {https://arxiv.org/abs/1506.06438v2},\n year = {2015}\n}\n\n", - "citation_id": "w6CoVmFK" - }, - "doi:10.1186/1471-2105-11-544": { - "source": "doi", - "identifer": "10.1186/1471-2105-11-544", - "standard_citation": "doi:10.1186/1471-2105-11-544", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 23 - ] - ], - "date-time": "2017-09-23T07:22:14Z", - "timestamp": 1506151334602 - }, - "reference-count": 0, - "publisher": "Springer Nature", - "issue": "1", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2010 - ] - ] - }, - "DOI": "10.1186/1471-2105-11-544", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2010, - 11, - 3 - ] - ], - "date-time": "2010-11-03T07:41:19Z", - "timestamp": 1288770079000 - }, - "page": "544", - "source": "Crossref", - "is-referenced-by-count": 35, - "title": "Clustering metagenomic sequences with interpolated Markov models", - "prefix": "10.1186", - "volume": "11", - "author": [ - { - "given": "David R", - "family": "Kelley", - "affiliation": [] - }, - { - "given": "Steven L", - "family": "Salzberg", - "affiliation": [] - } - ], - "member": "297", - "container-title": "BMC Bioinformatics", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2016, - 5, - 16 - ] - ], - "date-time": "2016-05-16T12:50:09Z", - "timestamp": 1463403009000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2010 - ] - ] - }, - "references-count": 0, - "alternative-id": [ - "1471-2105-11-544" - ], - "URL": "https://doi.org/10.1186/1471-2105-11-544", - "relation": {}, - "subject": [ - "Biochemistry", - "Applied Mathematics", - "Molecular Biology", - "Structural Biology", - "Computer Science Applications" - ], - "container-title-short": "BMC Bioinformatics", - "id": "c4rnN1wo" - }, - "citation_id": "c4rnN1wo" - }, - "doi:10.1038/nbt0710-691": { - "source": "doi", - "identifer": "10.1038/nbt0710-691", - "standard_citation": "doi:10.1038/nbt0710-691", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 25 - ] - ], - "date-time": "2017-09-25T20:42:13Z", - "timestamp": 1506372133848 - }, - "reference-count": 24, - "publisher": "Springer Nature", - "issue": "7", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2010, - 7 - ] - ] - }, - "DOI": "10.1038/nbt0710-691", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2010, - 7, - 12 - ] - ], - "date-time": "2010-07-12T01:35:21Z", - "timestamp": 1278898521000 - }, - "page": "691-693", - "source": "Crossref", - "is-referenced-by-count": 135, - "title": "Cloud computing and the DNA data race", - "prefix": "10.1038", - "volume": "28", - "author": [ - { - "given": "Michael C", - "family": "Schatz", - "affiliation": [] - }, - { - "given": "Ben", - "family": "Langmead", - "affiliation": [] - }, - { - "given": "Steven L", - "family": "Salzberg", - "affiliation": [] - } - ], - "member": "339", - "container-title": "Nature Biotechnology", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 19 - ] - ], - "date-time": "2017-06-19T10:40:45Z", - "timestamp": 1497868845000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2010, - 7 - ] - ] - }, - "references-count": 24, - "alternative-id": [ - "nbt0710-691" - ], - "URL": "https://doi.org/10.1038/nbt0710-691", - "relation": { - "cites": [] - }, - "subject": [ - "Biotechnology", - "Molecular Medicine", - "Applied Microbiology and Biotechnology", - "Bioengineering", - "Biomedical Engineering" - ], - "container-title-short": "Nat Biotechnol", - "id": "B6g0qKf4" - }, - "citation_id": "B6g0qKf4" - }, - "doi:10.1016/j.neunet.2014.09.003": { - "source": "doi", - "identifer": "10.1016/j.neunet.2014.09.003", - "standard_citation": "doi:10.1016/j.neunet.2014.09.003", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T10:22:21Z", - "timestamp": 1507112541015 - }, - "reference-count": 888, - "publisher": "Elsevier BV", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 2015, - 1, - 1 - ] - ], - "date-time": "2015-01-01T00:00:00Z", - "timestamp": 1420070400000 - }, - "delay-in-days": 0, - "content-version": "tdm" - } - ], - "funder": [ - { - "DOI": "10.13039/501100000780", - "name": "European Commission", - "doi-asserted-by": "crossref", - "award": [] - }, - { - "name": "SNF", - "award": [] - }, - { - "name": "DFG", - "award": [] - } - ], - "content-domain": { - "domain": [ - "elsevier.com", - "sciencedirect.com" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2015, - 1 - ] - ] - }, - "DOI": "10.1016/j.neunet.2014.09.003", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 10, - 13 - ] - ], - "date-time": "2014-10-13T17:04:15Z", - "timestamp": 1413219855000 - }, - "page": "85-117", - "update-policy": "http://dx.doi.org/10.1016/elsevier_cm_policy", - "source": "Crossref", - "is-referenced-by-count": 813, - "title": "Deep learning in neural networks: An overview", - "prefix": "10.1016", - "volume": "61", - "author": [ - { - "given": "Jürgen", - "family": "Schmidhuber", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Neural Networks", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S0893608014002135?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S0893608014002135?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 22 - ] - ], - "date-time": "2017-06-22T22:19:32Z", - "timestamp": 1498169972000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 1 - ] - ] - }, - "references-count": 888, - "alternative-id": [ - "S0893608014002135" - ], - "URL": "https://doi.org/10.1016/j.neunet.2014.09.003", - "relation": {}, - "subject": [ - "Cognitive Neuroscience", - "Artificial Intelligence" - ], - "container-title-short": "Neural Networks", - "assertion": [ - { - "value": "Elsevier", - "name": "publisher", - "label": "This article is maintained by" - }, - { - "value": "Deep learning in neural networks: An overview", - "name": "articletitle", - "label": "Article Title" - }, - { - "value": "Neural Networks", - "name": "journaltitle", - "label": "Journal Title" - }, - { - "value": "http://dx.doi.org/10.1016/j.neunet.2014.09.003", - "name": "articlelink", - "label": "CrossRef DOI link to publisher maintained version" - }, - { - "value": "article", - "name": "content_type", - "label": "Content Type" - }, - { - "value": "Copyright © 2014 Published by Elsevier Ltd.", - "name": "copyright", - "label": "Copyright" - } - ], - "id": "BQS8ClV0" - }, - "citation_id": "BQS8ClV0" - }, - "doi:10.1038/nrg.2015.3": { - "source": "doi", - "identifer": "10.1038/nrg.2015.3", - "standard_citation": "doi:10.1038/nrg.2015.3", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T14:02:16Z", - "timestamp": 1507039336011 - }, - "reference-count": 141, - "publisher": "Springer Nature", - "issue": "1", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/nrg.2015.3", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 11, - 23 - ] - ], - "date-time": "2015-11-23T06:33:30Z", - "timestamp": 1448260410000 - }, - "page": "19-32", - "source": "Crossref", - "is-referenced-by-count": 77, - "title": "RNA mis-splicing in disease", - "prefix": "10.1038", - "volume": "17", - "author": [ - { - "given": "Marina M.", - "family": "Scotti", - "affiliation": [] - }, - { - "given": "Maurice S.", - "family": "Swanson", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2015, - 11, - 23 - ] - ] - }, - "container-title": "Nature Reviews Genetics", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T00:10:08Z", - "timestamp": 1498263008000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 11, - 23 - ] - ] - }, - "references-count": 141, - "alternative-id": [ - "nrg.2015.3" - ], - "URL": "https://doi.org/10.1038/nrg.2015.3", - "relation": { - "cites": [] - }, - "subject": [ - "Genetics(clinical)", - "Genetics", - "Molecular Biology" - ], - "container-title-short": "Nat Rev Genet", - "id": "QFK6GapR" - }, - "citation_id": "QFK6GapR" - }, - "doi:10.1371/journal.pcbi.1004977": { - "source": "doi", - "identifer": "10.1371/journal.pcbi.1004977", - "standard_citation": "doi:10.1371/journal.pcbi.1004977", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T07:02:17Z", - "timestamp": 1507100537918 - }, - "reference-count": 66, - "publisher": "Public Library of Science (PLoS)", - "issue": "7", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2016, - 7, - 11 - ] - ], - "date-time": "2016-07-11T00:00:00Z", - "timestamp": 1468195200000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/501100003407", - "name": "Ministero dell’Istruzione, dell’Università e della Ricerca", - "doi-asserted-by": "publisher", - "award": [ - "RBFR13EWWI" - ] - }, - { - "DOI": "10.13039/100000001", - "name": "National Science Foundation", - "doi-asserted-by": "publisher", - "award": [ - "CNS-958379", - "CNS-0855217", - "ACI-1126113" - ] - }, - { - "DOI": "10.13039/100000060", - "name": "National Institute of Allergy and Infectious Diseases", - "doi-asserted-by": "publisher", - "award": [ - "1R21AI121784-01" - ] - }, - { - "name": "Seventh Framework Programme (BE)", - "award": [ - "PCIG13-618833" - ] - }, - { - "name": "Fondazione Caritro", - "award": [ - "Rif.Int.2013.0239" - ] - }, - { - "name": "Terme di Comano", - "award": [] - }, - { - "name": "City University of New York High Performance Computing Center", - "award": [] - } - ], - "content-domain": { - "domain": [ - "www.ploscompbiol.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pcbi.1004977", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 7, - 11 - ] - ], - "date-time": "2016-07-11T17:28:34Z", - "timestamp": 1468258114000 - }, - "page": "e1004977", - "update-policy": "http://dx.doi.org/10.1371/journal.pcbi.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 13, - "title": "Machine Learning Meta-analysis of Large Metagenomic Datasets: Tools and Biological Insights", - "prefix": "10.1371", - "volume": "12", - "author": [ - { - "given": "Edoardo", - "family": "Pasolli", - "affiliation": [] - }, - { - "given": "Duy Tin", - "family": "Truong", - "affiliation": [] - }, - { - "given": "Faizan", - "family": "Malik", - "affiliation": [] - }, - { - "given": "Levi", - "family": "Waldron", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-1583-5794", - "authenticated-orcid": true, - "given": "Nicola", - "family": "Segata", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2016, - 7, - 11 - ] - ] - }, - "container-title": "PLOS Computational Biology", - "original-title": [], - "link": [ - { - "URL": "http://dx.plos.org/10.1371/journal.pcbi.1004977", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T18:12:26Z", - "timestamp": 1498327946000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Jonathan A.", - "family": "Eisen", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 7, - 11 - ] - ] - }, - "references-count": 66, - "URL": "https://doi.org/10.1371/journal.pcbi.1004977", - "relation": { - "cites": [] - }, - "subject": [ - "Ecology", - "Modelling and Simulation", - "Computational Theory and Mathematics", - "Genetics", - "Ecology, Evolution, Behavior and Systematics", - "Molecular Biology", - "Cellular and Molecular Neuroscience" - ], - "container-title-short": "PLoS Comput Biol", - "id": "y9s5irW" - }, - "citation_id": "y9s5irW" - }, - "arxiv:1701.01329": { - "source": "arxiv", - "identifer": "1701.01329", - "standard_citation": "arxiv:1701.01329", - "bibtex": "@article{8LWFFeYg,\n abstract = {In de novo drug design, computational strategies are used to generate novel\nmolecules with good affinity to the desired biological target. In this work, we\nshow that recurrent neural networks can be trained as generative models for\nmolecular structures, similar to statistical language models in natural\nlanguage processing. We demonstrate that the properties of the generated\nmolecules correlate very well with the properties of the molecules used to\ntrain the model. In order to enrich libraries with molecules active towards a\ngiven biological target, we propose to fine-tune the model with small sets of\nmolecules, which are known to be active against that target.\nAgainst Staphylococcus aureus, the model reproduced 14% of 6051 hold-out test\nmolecules that medicinal chemists designed, whereas against Plasmodium\nfalciparum (Malaria) it reproduced 28% of 1240 test molecules. When coupled\nwith a scoring function, our model can perform the complete de novo drug design\ncycle to generate large sets of novel molecules for drug discovery.},\n archiveprefix = {arXiv},\n author = {Marwin H. S. Segler and Thierry Kogej and Christian Tyrchan and Mark P. Waller},\n eprint = {1701.01329v1},\n file = {1701.01329v1.pdf},\n month = {Jan},\n primaryclass = {cs.NE},\n title = {Generating Focussed Molecule Libraries for Drug Discovery with Recurrent\nNeural Networks},\n url = {https://arxiv.org/abs/1701.01329v1},\n year = {2017}\n}\n\n", - "citation_id": "8LWFFeYg" - }, - "doi:10.1109/icassp.2014.6853593": { - "source": "doi", - "identifer": "10.1109/icassp.2014.6853593", - "standard_citation": "doi:10.1109/icassp.2014.6853593", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T04:02:37Z", - "timestamp": 1507003357711 - }, - "reference-count": 23, - "publisher": "IEEE", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2014, - 5 - ] - ] - }, - "DOI": "10.1109/icassp.2014.6853593", - "type": "paper-conference", - "created": { - "date-parts": [ - [ - 2014, - 7, - 29 - ] - ], - "date-time": "2014-07-29T19:23:23Z", - "timestamp": 1406661803000 - }, - "source": "Crossref", - "is-referenced-by-count": 14, - "title": "On parallelizability of stochastic gradient descent for speech DNNS", - "prefix": "10.1109", - "author": [ - { - "given": "Frank", - "family": "Seide", - "affiliation": [] - }, - { - "given": "Hao", - "family": "Fu", - "affiliation": [] - }, - { - "given": "Jasha", - "family": "Droppo", - "affiliation": [] - }, - { - "given": "Gang", - "family": "Li", - "affiliation": [] - }, - { - "given": "Dong", - "family": "Yu", - "affiliation": [] - } - ], - "member": "263", - "container-title": "2014 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)", - "original-title": [], - "link": [ - { - "URL": "http://xplorestaging.ieee.org/ielx7/6844297/6853544/06853593.pdf?arnumber=6853593", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 3, - 23 - ] - ], - "date-time": "2017-03-23T18:36:57Z", - "timestamp": 1490294217000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 5 - ] - ] - }, - "references-count": 23, - "URL": "https://doi.org/10.1109/icassp.2014.6853593", - "relation": {}, - "id": "IULiPa6L" - }, - "citation_id": "IULiPa6L" - }, - "arxiv:1610.02391": { - "source": "arxiv", - "identifer": "1610.02391", - "standard_citation": "arxiv:1610.02391", - "bibtex": "@article{RZsNSRDS,\n abstract = {We propose a technique for producing \"visual explanations\" for decisions from\na large class of CNN-based models, making them more transparent. Our approach -\nGradient-weighted Class Activation Mapping (Grad-CAM), uses the gradients of\nany target concept, flowing into the final convolutional layer to produce a\ncoarse localization map highlighting the important regions in the image for\npredicting the concept. Unlike previous approaches, GradCAM is applicable to a\nwide variety of CNN model-families: (1) CNNs with fully-connected layers (e.g.\nVGG), (2) CNNs used for structured outputs (e.g. captioning), (3) CNNs used in\ntasks with multimodal inputs (e.g. VQA) or reinforcement learning, without any\narchitectural changes or re-training. We combine GradCAM with fine-grained\nvisualizations to create a high-resolution class-discriminative visualization\nand apply it to off-the-shelf image classification, captioning, and visual\nquestion answering (VQA) models, including ResNet-based architectures. In the\ncontext of image classification models, our visualizations (a) lend insights\ninto their failure modes (showing that seemingly unreasonable predictions have\nreasonable explanations), (b) are robust to adversarial images, (c) outperform\nprevious methods on weakly-supervised localization, (d) are more faithful to\nthe underlying model and (e) help achieve generalization by identifying dataset\nbias. For captioning and VQA, our visualizations show that even non-attention\nbased models can localize inputs. Finally, we conduct human studies to measure\nif GradCAM explanations help users establish trust in predictions from deep\nnetworks and show that GradCAM helps untrained users successfully discern a\n\"stronger\" deep network from a \"weaker\" one. Our code is available at\nhttps://github.com/ramprs/grad-cam. A demo and a video of the demo can be found\nat http://gradcam.cloudcv.org and youtu.be/COjUB9Izk6E.},\n archiveprefix = {arXiv},\n author = {Ramprasaath R. Selvaraju and Michael Cogswell and Abhishek Das and Ramakrishna Vedantam and Devi Parikh and Dhruv Batra},\n eprint = {1610.02391v3},\n file = {1610.02391v3.pdf},\n month = {Nov},\n primaryclass = {cs.CV},\n title = {Grad-CAM: Visual Explanations from Deep Networks via Gradient-based\nLocalization},\n url = {https://arxiv.org/abs/1610.02391v3},\n year = {2016}\n}\n\n", - "citation_id": "RZsNSRDS" - }, - "doi:10.1371/journal.pcbi.1004271": { - "source": "doi", - "identifer": "10.1371/journal.pcbi.1004271", - "standard_citation": "doi:10.1371/journal.pcbi.1004271", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 27 - ] - ], - "date-time": "2017-09-27T20:42:17Z", - "timestamp": 1506544937243 - }, - "reference-count": 49, - "publisher": "Public Library of Science (PLoS)", - "issue": "5", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2015, - 5, - 27 - ] - ], - "date-time": "2015-05-27T00:00:00Z", - "timestamp": 1432684800000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "www.ploscompbiol.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pcbi.1004271", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 5, - 28 - ] - ], - "date-time": "2015-05-28T10:44:43Z", - "timestamp": 1432809883000 - }, - "page": "e1004271", - "update-policy": "http://dx.doi.org/10.1371/journal.pcbi.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 9, - "title": "SeqGL Identifies Context-Dependent Binding Signals in Genome-Wide Regulatory Element Maps", - "prefix": "10.1371", - "volume": "11", - "author": [ - { - "given": "Manu", - "family": "Setty", - "affiliation": [] - }, - { - "given": "Christina S.", - "family": "Leslie", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2015, - 5, - 27 - ] - ] - }, - "container-title": "PLOS Computational Biology", - "original-title": [], - "link": [ - { - "URL": "http://dx.plos.org/10.1371/journal.pcbi.1004271", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 23 - ] - ], - "date-time": "2017-06-23T12:28:44Z", - "timestamp": 1498220924000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Zhiping", - "family": "Weng", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 5, - 27 - ] - ] - }, - "references-count": 49, - "URL": "https://doi.org/10.1371/journal.pcbi.1004271", - "relation": { - "cites": [] - }, - "subject": [ - "Ecology", - "Modelling and Simulation", - "Computational Theory and Mathematics", - "Genetics", - "Ecology, Evolution, Behavior and Systematics", - "Molecular Biology", - "Cellular and Molecular Neuroscience" - ], - "container-title-short": "PLoS Comput Biol", - "id": "138dgb9Ca" - }, - "citation_id": "138dgb9Ca" - }, - "arxiv:1610.04181": { - "source": "arxiv", - "identifer": "1610.04181", - "standard_citation": "arxiv:1610.04181", - "bibtex": "@article{T2Md9xLY,\n abstract = {Sources of variability in experimentally derived data include measurement\nerror in addition to the physical phenomena of interest. This measurement error\nis a combination of systematic components, originating from the measuring\ninstrument, and random measurement errors. Several novel biological\ntechnologies, such as mass cytometry and single-cell RNA-seq, are plagued with\nsystematic errors that may severely affect statistical analysis if the data is\nnot properly calibrated. We propose a novel deep learning approach for removing\nsystematic batch effects. Our method is based on a residual network, trained to\nminimize the Maximum Mean Discrepancy (MMD) between the multivariate\ndistributions of two replicates, measured in different batches. We apply our\nmethod to mass cytometry and single-cell RNA-seq datasets, and demonstrate that\nit effectively attenuates batch effects.},\n archiveprefix = {arXiv},\n author = {Uri Shaham and Kelly P. Stanton and Jun Zhao and Huamin Li and Khadir Raddassi and Ruth Montgomery and Yuval Kluger},\n doi = {10.1093/bioinformatics/btx196},\n eprint = {1610.04181v5},\n file = {1610.04181v5.pdf},\n month = {Nov},\n primaryclass = {stat.ML},\n title = {Removal of Batch Effects using Distribution-Matching Residual Networks},\n url = {https://arxiv.org/abs/1610.04181v5},\n year = {2016}\n}\n\n", - "citation_id": "T2Md9xLY" - }, - "doi:10.1515/9781400881970-018": { - "source": "doi", - "identifer": "10.1515/9781400881970-018", - "standard_citation": "doi:10.1515/9781400881970-018", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 5, - 25 - ] - ], - "date-time": "2017-05-25T12:37:20Z", - "timestamp": 1495715840346 - }, - "publisher-location": "Princeton", - "reference-count": 0, - "publisher": "Princeton University Press", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "short-container-title": [], - "DOI": "10.1515/9781400881970-018", - "type": "chapter", - "created": { - "date-parts": [ - [ - 2016, - 5, - 25 - ] - ], - "date-time": "2016-05-25T10:10:30Z", - "timestamp": 1464171030000 - }, - "issued": { - "date-parts": [ - [ - 1953 - ] - ] - }, - "source": "Crossref", - "is-referenced-by-count": 82, - "title": "17. A Value for n-Person Games", - "prefix": "10.1515", - "author": [ - { - "given": "L. S.", - "family": "Shapley", - "affiliation": [] - } - ], - "member": "374", - "container-title": "Contributions to the Theory of Games (AM-28), Volume II", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 5, - 18 - ] - ], - "date-time": "2017-05-18T21:46:44Z", - "timestamp": 1495144004000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "references-count": 0, - "URL": "https://doi.org/10.1515/9781400881970-018", - "relation": {}, - "id": "YBJdA6LJ" - }, - "citation_id": "YBJdA6LJ" - }, - "doi:10.1146/annurev-bioeng-071516-044442": { - "source": "doi", - "identifer": "10.1146/annurev-bioeng-071516-044442", - "standard_citation": "doi:10.1146/annurev-bioeng-071516-044442", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T04:02:42Z", - "timestamp": 1507003362713 - }, - "reference-count": 121, - "publisher": "Annual Reviews", - "issue": "1", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2017, - 6, - 21 - ] - ] - }, - "DOI": "10.1146/annurev-bioeng-071516-044442", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2017, - 3, - 16 - ] - ], - "date-time": "2017-03-16T22:54:53Z", - "timestamp": 1489704893000 - }, - "page": "221-248", - "source": "Crossref", - "is-referenced-by-count": 8, - "title": "Deep Learning in Medical Image Analysis", - "prefix": "10.1146", - "volume": "19", - "author": [ - { - "given": "Dinggang", - "family": "Shen", - "affiliation": [ - { - "name": "Department of Radiology, University of North Carolina, Chapel Hill, North Carolina 27599;" - }, - { - "name": "Department of Brain and Cognitive Engineering, Korea University, Seoul 02841, Republic of Korea;" - } - ] - }, - { - "given": "Guorong", - "family": "Wu", - "affiliation": [ - { - "name": "Department of Radiology, University of North Carolina, Chapel Hill, North Carolina 27599;" - } - ] - }, - { - "given": "Heung-Il", - "family": "Suk", - "affiliation": [ - { - "name": "Department of Brain and Cognitive Engineering, Korea University, Seoul 02841, Republic of Korea;" - } - ] - } - ], - "member": "22", - "container-title": "Annual Review of Biomedical Engineering", - "original-title": [], - "link": [ - { - "URL": "http://www.annualreviews.org/doi/pdf/10.1146/annurev-bioeng-071516-044442", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 21 - ] - ], - "date-time": "2017-06-21T00:03:47Z", - "timestamp": 1498003427000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 6, - 21 - ] - ] - }, - "references-count": 121, - "alternative-id": [ - "10.1146/annurev-bioeng-071516-044442" - ], - "URL": "https://doi.org/10.1146/annurev-bioeng-071516-044442", - "relation": {}, - "subject": [ - "Medicine (miscellaneous)", - "Biomedical Engineering" - ], - "container-title-short": "Annu. Rev. Biomed. Eng.", - "id": "yEstnIOT" - }, - "citation_id": "yEstnIOT" - }, - "doi:10.1109/tmi.2016.2528162": { - "source": "doi", - "identifer": "10.1109/tmi.2016.2528162", - "standard_citation": "doi:10.1109/tmi.2016.2528162", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T04:22:16Z", - "timestamp": 1507004536515 - }, - "reference-count": 73, - "publisher": "Institute of Electrical and Electronics Engineers (IEEE)", - "issue": "5", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 5 - ] - ] - }, - "DOI": "10.1109/tmi.2016.2528162", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 2, - 11 - ] - ], - "date-time": "2016-02-11T12:46:43Z", - "timestamp": 1455194803000 - }, - "page": "1285-1298", - "source": "Crossref", - "is-referenced-by-count": 128, - "title": "Deep Convolutional Neural Networks for Computer-Aided Detection: CNN Architectures, Dataset Characteristics and Transfer Learning", - "prefix": "10.1109", - "volume": "35", - "author": [ - { - "given": "Hoo-Chang", - "family": "Shin", - "affiliation": [] - }, - { - "given": "Holger R.", - "family": "Roth", - "affiliation": [] - }, - { - "given": "Mingchen", - "family": "Gao", - "affiliation": [] - }, - { - "given": "Le", - "family": "Lu", - "affiliation": [] - }, - { - "given": "Ziyue", - "family": "Xu", - "affiliation": [] - }, - { - "given": "Isabella", - "family": "Nogues", - "affiliation": [] - }, - { - "given": "Jianhua", - "family": "Yao", - "affiliation": [] - }, - { - "given": "Daniel", - "family": "Mollura", - "affiliation": [] - }, - { - "given": "Ronald M.", - "family": "Summers", - "affiliation": [] - } - ], - "member": "263", - "container-title": "IEEE Transactions on Medical Imaging", - "original-title": [], - "link": [ - { - "URL": "http://xplorestaging.ieee.org/ielx7/42/7463083/07404017.pdf?arnumber=7404017", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 9, - 24 - ] - ], - "date-time": "2016-09-24T14:17:09Z", - "timestamp": 1474726629000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 5 - ] - ] - }, - "references-count": 73, - "URL": "https://doi.org/10.1109/tmi.2016.2528162", - "relation": {}, - "subject": [ - "Electrical and Electronic Engineering", - "Radiological and Ultrasound Technology", - "Software", - "Computer Science Applications" - ], - "container-title-short": "IEEE Trans. Med. Imaging", - "id": "1GAyqYBNZ" - }, - "citation_id": "1GAyqYBNZ" - }, - "arxiv:1704.02685": { - "source": "arxiv", - "identifer": "1704.02685", - "standard_citation": "arxiv:1704.02685", - "bibtex": "@article{zhmq9ktJ,\n abstract = {The purported \"black box\"' nature of neural networks is a barrier to adoption\nin applications where interpretability is essential. Here we present DeepLIFT\n(Deep Learning Important FeaTures), a method for decomposing the output\nprediction of a neural network on a specific input by backpropagating the\ncontributions of all neurons in the network to every feature of the input.\nDeepLIFT compares the activation of each neuron to its 'reference activation'\nand assigns contribution scores according to the difference. By optionally\ngiving separate consideration to positive and negative contributions, DeepLIFT\ncan also reveal dependencies which are missed by other approaches. Scores can\nbe computed efficiently in a single backward pass. We apply DeepLIFT to models\ntrained on MNIST and simulated genomic data, and show significant advantages\nover gradient-based methods. A detailed video tutorial on the method is at\nhttp://goo.gl/qKb7pL and code is at http://goo.gl/RM8jvH.},\n archiveprefix = {arXiv},\n author = {Avanti Shrikumar and Peyton Greenside and Anshul Kundaje},\n eprint = {1704.02685v1},\n file = {1704.02685v1.pdf},\n month = {Apr},\n primaryclass = {cs.CV},\n title = {Learning Important Features Through Propagating Activation Differences},\n url = {https://arxiv.org/abs/1704.02685v1},\n year = {2017}\n}\n\n", - "citation_id": "zhmq9ktJ" - }, - "doi:10.1101/103663": { - "source": "doi", - "identifer": "10.1101/103663", - "standard_citation": "doi:10.1101/103663", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T21:27:33Z", - "timestamp": 1502400453800 - }, - "posted": { - "date-parts": [ - [ - 2017, - 1, - 27 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2017, - 1, - 27 - ] - ] - }, - "abstract": "Deep learning approaches that have produced breakthrough predictive models in computer vision, speech recognition and machine translation are now being successfully applied to problems in regulatory genomics. However, deep learning architectures used thus far in genomics are often directly ported from computer vision and natural language processing applications with few, if any, domain-specific modifications. In double-stranded DNA, the same pattern may appear identically on one strand and its reverse complement due to complementary base pairing. Here, we show that conventional deep learning models that do not explicitly model this property can produce substantially different predictions on forward and reverse-complement versions of the same DNA sequence. We present four new convolutional neural network layers that leverage the reverse-complement property of genomic DNA sequence by sharing parameters between forward and reverse-complement representations in the model. These layers guarantee that forward and reverse-complement sequences produce identical predictions within numerical precision. Using experiments on simulated and in vivo transcription factor binding data, we show that our proposed architectures lead to improved performance, faster learning and cleaner internal representations compared to conventional architectures trained on the same data.", - "DOI": "10.1101/103663", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2017, - 1, - 28 - ] - ], - "date-time": "2017-01-28T06:10:13Z", - "timestamp": 1485583813000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Reverse-complement parameter sharing improves deep learning models for genomics", - "prefix": "10.1101", - "author": [ - { - "given": "Avanti", - "family": "Shrikumar", - "affiliation": [] - }, - { - "given": "Peyton", - "family": "Greenside", - "affiliation": [] - }, - { - "given": "Anshul", - "family": "Kundaje", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/103663", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 1, - 28 - ] - ], - "date-time": "2017-01-28T06:10:25Z", - "timestamp": 1485583825000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 1, - 27 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/103663", - "relation": {}, - "id": "iEmvzeT8" - }, - "citation_id": "iEmvzeT8" - }, - "doi:10.1038/nature16961": { - "source": "doi", - "identifer": "10.1038/nature16961", - "standard_citation": "doi:10.1038/nature16961", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T09:42:13Z", - "timestamp": 1507110133732 - }, - "reference-count": 62, - "publisher": "Springer Nature", - "issue": "7587", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/nature16961", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 1, - 26 - ] - ], - "date-time": "2016-01-26T17:44:19Z", - "timestamp": 1453830259000 - }, - "page": "484-489", - "source": "Crossref", - "is-referenced-by-count": 488, - "title": "Mastering the game of Go with deep neural networks and tree search", - "prefix": "10.1038", - "volume": "529", - "author": [ - { - "given": "David", - "family": "Silver", - "affiliation": [] - }, - { - "given": "Aja", - "family": "Huang", - "affiliation": [] - }, - { - "given": "Chris J.", - "family": "Maddison", - "affiliation": [] - }, - { - "given": "Arthur", - "family": "Guez", - "affiliation": [] - }, - { - "given": "Laurent", - "family": "Sifre", - "affiliation": [] - }, - { - "given": "George", - "family": "van den Driessche", - "affiliation": [] - }, - { - "given": "Julian", - "family": "Schrittwieser", - "affiliation": [] - }, - { - "given": "Ioannis", - "family": "Antonoglou", - "affiliation": [] - }, - { - "given": "Veda", - "family": "Panneershelvam", - "affiliation": [] - }, - { - "given": "Marc", - "family": "Lanctot", - "affiliation": [] - }, - { - "given": "Sander", - "family": "Dieleman", - "affiliation": [] - }, - { - "given": "Dominik", - "family": "Grewe", - "affiliation": [] - }, - { - "given": "John", - "family": "Nham", - "affiliation": [] - }, - { - "given": "Nal", - "family": "Kalchbrenner", - "affiliation": [] - }, - { - "given": "Ilya", - "family": "Sutskever", - "affiliation": [] - }, - { - "given": "Timothy", - "family": "Lillicrap", - "affiliation": [] - }, - { - "given": "Madeleine", - "family": "Leach", - "affiliation": [] - }, - { - "given": "Koray", - "family": "Kavukcuoglu", - "affiliation": [] - }, - { - "given": "Thore", - "family": "Graepel", - "affiliation": [] - }, - { - "given": "Demis", - "family": "Hassabis", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2016, - 1, - 27 - ] - ] - }, - "container-title": "Nature", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 24 - ] - ], - "date-time": "2017-06-24T04:33:59Z", - "timestamp": 1498278839000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 1, - 27 - ] - ] - }, - "references-count": 62, - "alternative-id": [ - "nature16961" - ], - "URL": "https://doi.org/10.1038/nature16961", - "relation": { - "cites": [] - }, - "subject": [ - "Multidisciplinary" - ], - "container-title-short": "Nature", - "id": "2gn6PKkv" - }, - "citation_id": "2gn6PKkv" - }, - "arxiv:1312.6034": { - "source": "arxiv", - "identifer": "1312.6034", - "standard_citation": "arxiv:1312.6034", - "bibtex": "@article{1YcKYTvO,\n abstract = {This paper addresses the visualisation of image classification models, learnt\nusing deep Convolutional Networks (ConvNets). We consider two visualisation\ntechniques, based on computing the gradient of the class score with respect to\nthe input image. The first one generates an image, which maximises the class\nscore [Erhan et al., 2009], thus visualising the notion of the class, captured\nby a ConvNet. The second technique computes a class saliency map, specific to a\ngiven image and class. We show that such maps can be employed for weakly\nsupervised object segmentation using classification ConvNets. Finally, we\nestablish the connection between the gradient-based ConvNet visualisation\nmethods and deconvolutional networks [Zeiler et al., 2013].},\n archiveprefix = {arXiv},\n author = {Karen Simonyan and Andrea Vedaldi and Andrew Zisserman},\n eprint = {1312.6034v2},\n file = {1312.6034v2.pdf},\n month = {12},\n primaryclass = {cs.CV},\n title = {Deep Inside Convolutional Networks: Visualising Image Classification\nModels and Saliency Maps},\n url = {https://arxiv.org/abs/1312.6034v2},\n year = {2013}\n}\n\n", - "citation_id": "1YcKYTvO" - }, - "arxiv:1607.02078": { - "source": "arxiv", - "identifer": "1607.02078", - "standard_citation": "arxiv:1607.02078", - "bibtex": "@article{G10wkFHt,\n abstract = {Motivation: Histone modifications are among the most important factors that\ncontrol gene regulation. Computational methods that predict gene expression\nfrom histone modification signals are highly desirable for understanding their\ncombinatorial effects in gene regulation. This knowledge can help in developing\n'epigenetic drugs' for diseases like cancer. Previous studies for quantifying\nthe relationship between histone modifications and gene expression levels\neither failed to capture combinatorial effects or relied on multiple methods\nthat separate predictions and combinatorial analysis. This paper develops a\nunified discriminative framework using a deep convolutional neural network to\nclassify gene expression using histone modification data as input. Our system, called DeepChrome, allows automatic extraction of complex interactions among\nimportant features. To simultaneously visualize the combinatorial interactions\namong histone modifications, we propose a novel optimization-based technique\nthat generates feature pattern maps from the learnt deep model. This provides\nan intuitive description of underlying epigenetic mechanisms that regulate\ngenes. Results: We show that DeepChrome outperforms state-of-the-art models\nlike Support Vector Machines and Random Forests for gene expression\nclassification task on 56 different cell-types from REMC database. The output\nof our visualization technique not only validates the previous observations but\nalso allows novel insights about combinatorial interactions among histone\nmodification marks, some of which have recently been observed by experimental\nstudies.},\n archiveprefix = {arXiv},\n author = {Ritambhara Singh and Jack Lanchantin and Gabriel Robins and Yanjun Qi},\n eprint = {1607.02078v1},\n file = {1607.02078v1.pdf},\n month = {Jul},\n primaryclass = {cs.LG},\n title = {DeepChrome: Deep-learning for predicting gene expression from histone\nmodifications},\n url = {https://arxiv.org/abs/1607.02078v1},\n year = {2016}\n}\n\n", - "citation_id": "G10wkFHt" - }, - "arxiv:1503.01919": { - "source": "arxiv", - "identifer": "1503.01919", - "standard_citation": "arxiv:1503.01919", - "bibtex": "@article{81Cl5QSM,\n abstract = {Machine learning is widely used to analyze biological sequence data.\nNon-sequential models such as SVMs or feed-forward neural networks are often\nused although they have no natural way of handling sequences of varying length.\nRecurrent neural networks such as the long short term memory (LSTM) model on\nthe other hand are designed to handle sequences. In this study we demonstrate\nthat LSTM networks predict the subcellular location of proteins given only the\nprotein sequence with high accuracy (0.902) outperforming current state of the\nart algorithms. We further improve the performance by introducing convolutional\nfilters and experiment with an attention mechanism which lets the LSTM focus on\nspecific parts of the protein. Lastly we introduce new visualizations of both\nthe convolutional filters and the attention mechanisms and show how they can be\nused to extract biological relevant knowledge from the LSTM networks.},\n archiveprefix = {arXiv},\n author = {Søren Kaae Sønderby and Casper Kaae Sønderby and Henrik Nielsen and Ole Winther},\n doi = {10.1007/978-3-319-21233-3_6},\n eprint = {1503.01919v1},\n file = {1503.01919v1.pdf},\n month = {Mar},\n note = {Algorithms for Computational Biology 9199 (2015) 68},\n primaryclass = {q-bio.QM},\n title = {Convolutional LSTM Networks for Subcellular Localization of Proteins},\n url = {https://arxiv.org/abs/1503.01919v1},\n year = {2015}\n}\n\n", - "citation_id": "81Cl5QSM" - }, - "doi:10.1515/metgen-2016-0001": { - "source": "doi", - "identifer": "10.1515/metgen-2016-0001", - "standard_citation": "doi:10.1515/metgen-2016-0001", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T05:15:46Z", - "timestamp": 1502342146851 - }, - "reference-count": 101, - "publisher": "Walter de Gruyter GmbH", - "issue": "1", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by-nc-nd/3.0", - "start": { - "date-parts": [ - [ - 2017, - 1, - 1 - ] - ], - "date-time": "2017-01-01T00:00:00Z", - "timestamp": 1483228800000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2017, - 1, - 1 - ] - ] - }, - "abstract": "AbstractOwing to the complexity and variability of metagenomic studies, modern machine learning approaches have seen increased usage to answer a variety of question encompassing the full range of metagenomic NGS data analysis.We review here the contribution of machine learning techniques for the field of metagenomics, by presenting known successful approaches in a unified framework. This review focuses on five important metagenomic problems:OTU-clustering, binning, taxonomic proffiing and assignment, comparative metagenomics and gene prediction. For each of these problems, we identify the most prominent methods, summarize the machine learning approaches used and put them into perspective of similar methods.We conclude our review looking further ahead at the challenge posed by the analysis of interactions within microbial communities and different environments, in a field one could call “integrative metagenomics”.", - "DOI": "10.1515/metgen-2016-0001", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 12, - 14 - ] - ], - "date-time": "2016-12-14T10:04:57Z", - "timestamp": 1481709897000 - }, - "source": "Crossref", - "is-referenced-by-count": 1, - "title": "Machine learning for metagenomics: methods and tools", - "prefix": "10.1515", - "volume": "1", - "author": [ - { - "given": "Hayssam", - "family": "Soueidan", - "affiliation": [] - }, - { - "given": "Macha", - "family": "Nikolski", - "affiliation": [] - } - ], - "member": "374", - "container-title": "Metagenomics", - "original-title": [], - "link": [ - { - "URL": "http://www.degruyter.com/view/j/metgen.2016.1.issue-1/metgen-2016-0001/metgen-2016-0001.xml", - "content-type": "text/html", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://www.degruyter.com/view/j/metgen.2016.1.issue-1/metgen-2016-0001/metgen-2016-0001.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T05:53:17Z", - "timestamp": 1498369997000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 1, - 1 - ] - ] - }, - "references-count": 101, - "URL": "https://doi.org/10.1515/metgen-2016-0001", - "relation": { - "cites": [] - }, - "id": "yFOAeemA" - }, - "citation_id": "yFOAeemA" - }, - "url:http://www.businessinsider.com/ibm-edges-closer-to-human-speech-recognition-2017-3": { - "source": "url", - "identifer": "http://www.businessinsider.com/ibm-edges-closer-to-human-speech-recognition-2017-3", - "standard_citation": "url:http://www.businessinsider.com/ibm-edges-closer-to-human-speech-recognition-2017-3", - "citeproc": { - "URL": "http://www.businessinsider.com/ibm-edges-closer-to-human-speech-recognition-2017-3", - "greycite-canonical-uri": "http://uk.businessinsider.com/ibm-edges-closer-to-human-speech-recognition-2017-3", - "title": "IBM edges closer to human speech recognition", - "container-title": "Business Insider", - "issued": { - "date-parts": [ - [ - 2017, - 3, - 14 - ] - ] - }, - "author": [ - { - "family": "BI Intelligence", - "given": "" - } - ], - "redirects-to": "http://uk.businessinsider.com/ibm-edges-closer-to-human-speech-recognition-2017-3?r=US&IR=T", - "archives": [ - "http://wayback.archive.org/web/http://www.businessinsider.com/ibm-edges-closer-to-human-speech-recognition-2017-3" - ], - "type": "webpage", - "id": "nyjAIan4" - }, - "citation_id": "nyjAIan4" - }, - "arxiv:1412.6806": { - "source": "arxiv", - "identifer": "1412.6806", - "standard_citation": "arxiv:1412.6806", - "bibtex": "@article{f2L6isRj,\n abstract = {Most modern convolutional neural networks (CNNs) used for object recognition\nare built using the same principles: Alternating convolution and max-pooling\nlayers followed by a small number of fully connected layers. We re-evaluate the\nstate of the art for object recognition from small images with convolutional\nnetworks, questioning the necessity of different components in the pipeline. We\nfind that max-pooling can simply be replaced by a convolutional layer with\nincreased stride without loss in accuracy on several image recognition\nbenchmarks. Following this finding -- and building on other recent work for\nfinding simple network structures -- we propose a new architecture that\nconsists solely of convolutional layers and yields competitive or state of the\nart performance on several object recognition datasets (CIFAR-10, CIFAR-100, ImageNet). To analyze the network we introduce a new variant of the\n\"deconvolution approach\" for visualizing features learned by CNNs, which can be\napplied to a broader range of network structures than existing approaches.},\n archiveprefix = {arXiv},\n author = {Jost Tobias Springenberg and Alexey Dosovitskiy and Thomas Brox and Martin Riedmiller},\n eprint = {1412.6806v3},\n file = {1412.6806v3.pdf},\n month = {12},\n primaryclass = {cs.LG},\n title = {Striving for Simplicity: The All Convolutional Net},\n url = {https://arxiv.org/abs/1412.6806v3},\n year = {2014}\n}\n\n", - "citation_id": "f2L6isRj" - }, - "doi:10.1186/gb-2010-11-5-207": { - "source": "doi", - "identifer": "10.1186/gb-2010-11-5-207", - "standard_citation": "doi:10.1186/gb-2010-11-5-207", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T02:22:22Z", - "timestamp": 1506997342196 - }, - "reference-count": 0, - "publisher": "Springer Nature", - "issue": "5", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2010 - ] - ] - }, - "DOI": "10.1186/gb-2010-11-5-207", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2010, - 5, - 5 - ] - ], - "date-time": "2010-05-05T19:38:43Z", - "timestamp": 1273088323000 - }, - "page": "207", - "source": "Crossref", - "is-referenced-by-count": 222, - "title": "The case for cloud computing in genome informatics", - "prefix": "10.1186", - "volume": "11", - "author": [ - { - "given": "Lincoln D", - "family": "Stein", - "affiliation": [] - } - ], - "member": "297", - "container-title": "Genome Biology", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2016, - 5, - 16 - ] - ], - "date-time": "2016-05-16T19:24:54Z", - "timestamp": 1463426694000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2010 - ] - ] - }, - "references-count": 0, - "alternative-id": [ - "gb-2010-11-5-207" - ], - "URL": "https://doi.org/10.1186/gb-2010-11-5-207", - "relation": {}, - "container-title-short": "Genome Biol", - "id": "q0SsFrZd" - }, - "citation_id": "q0SsFrZd" - }, - "doi:10.1093/bioinformatics/16.1.16": { - "source": "doi", - "identifer": "10.1093/bioinformatics/16.1.16", - "standard_citation": "doi:10.1093/bioinformatics/16.1.16", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 22 - ] - ], - "date-time": "2017-09-22T04:42:42Z", - "timestamp": 1506055362414 - }, - "reference-count": 0, - "publisher": "Oxford University Press (OUP)", - "issue": "1", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2000, - 1, - 1 - ] - ] - }, - "DOI": "10.1093/bioinformatics/16.1.16", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2002, - 7, - 26 - ] - ], - "date-time": "2002-07-26T22:47:12Z", - "timestamp": 1027723632000 - }, - "page": "16-23", - "source": "Crossref", - "is-referenced-by-count": 583, - "title": "DNA binding sites: representation and discovery", - "prefix": "10.1093", - "volume": "16", - "author": [ - { - "given": "G. D.", - "family": "Stormo", - "affiliation": [] - } - ], - "member": "286", - "container-title": "Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/bioinformatics/article-pdf/16/1/16/669871/160016.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 22 - ] - ], - "date-time": "2017-08-22T15:12:48Z", - "timestamp": 1503414768000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2000, - 1, - 1 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1093/bioinformatics/16.1.16", - "relation": {}, - "subject": [ - "Statistics and Probability", - "Computational Theory and Mathematics", - "Biochemistry", - "Molecular Biology", - "Computational Mathematics", - "Computer Science Applications" - ], - "container-title-short": "Bioinformatics", - "id": "ywDQIvZJ" - }, - "citation_id": "ywDQIvZJ" - }, - "doi:10.1186/2049-2618-1-11": { - "source": "doi", - "identifer": "10.1186/2049-2618-1-11", - "standard_citation": "doi:10.1186/2049-2618-1-11", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T12:33:27Z", - "timestamp": 1502282007560 - }, - "reference-count": 19, - "publisher": "Springer Nature", - "issue": "1", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2013 - ] - ] - }, - "DOI": "10.1186/2049-2618-1-11", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2013, - 4, - 5 - ] - ], - "date-time": "2013-04-05T12:15:14Z", - "timestamp": 1365164114000 - }, - "page": "11", - "source": "Crossref", - "is-referenced-by-count": 20, - "title": "A comprehensive evaluation of multicategory classification methods for microbiomic data", - "prefix": "10.1186", - "volume": "1", - "author": [ - { - "given": "Alexander", - "family": "Statnikov", - "affiliation": [] - }, - { - "given": "Mikael", - "family": "Henaff", - "affiliation": [] - }, - { - "given": "Varun", - "family": "Narendra", - "affiliation": [] - }, - { - "given": "Kranti", - "family": "Konganti", - "affiliation": [] - }, - { - "given": "Zhiguo", - "family": "Li", - "affiliation": [] - }, - { - "given": "Liying", - "family": "Yang", - "affiliation": [] - }, - { - "given": "Zhiheng", - "family": "Pei", - "affiliation": [] - }, - { - "given": "Martin J", - "family": "Blaser", - "affiliation": [] - }, - { - "given": "Constantin F", - "family": "Aliferis", - "affiliation": [] - }, - { - "given": "Alexander V", - "family": "Alekseyenko", - "affiliation": [] - } - ], - "member": "297", - "container-title": "Microbiome", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2016, - 5, - 16 - ] - ], - "date-time": "2016-05-16T17:05:49Z", - "timestamp": 1463418349000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2013 - ] - ] - }, - "references-count": 19, - "alternative-id": [ - "2049-2618-1-11" - ], - "URL": "https://doi.org/10.1186/2049-2618-1-11", - "relation": { - "cites": [] - }, - "container-title-short": "Microbiome", - "id": "c5P9jHCg" - }, - "citation_id": "c5P9jHCg" - }, - "arxiv:1606.07461": { - "source": "arxiv", - "identifer": "1606.07461", - "standard_citation": "arxiv:1606.07461", - "bibtex": "@article{1Ad3UOefc,\n abstract = {Recurrent neural networks, and in particular long short-term memory networks\n(LSTMs), are a remarkably effective tool for sequence modeling that learn a\ndense black-box hidden representation of their sequential input. Researchers\ninterested in better understanding these models have studied the changes in\nhidden state representations over time and noticed some interpretable patterns\nbut also significant noise. In this work, we present LSTMVis a visual analysis\ntool for recurrent neural networks with a focus on understanding these hidden\nstate dynamics. The tool allows a user to select a hypothesis input range to\nfocus on local state changes, to match these states changes to similar patterns\nin a large data set, and to align these results with domain specific structural\nannotations. We further show several use cases of the tool for analyzing\nspecific hidden state properties on datasets containing nesting, phrase\nstructure, and chord progressions, and demonstrate how the tool can be used to\nisolate patterns for further statistical analysis.},\n archiveprefix = {arXiv},\n author = {Hendrik Strobelt and Sebastian Gehrmann and Bernd Huber and Hanspeter Pfister and Alexander M. Rush},\n eprint = {1606.07461v1},\n file = {1606.07461v1.pdf},\n month = {Jun},\n primaryclass = {cs.CL},\n title = {Visual Analysis of Hidden State Dynamics in Recurrent Neural Networks},\n url = {https://arxiv.org/abs/1606.07461v1},\n year = {2016}\n}\n\n", - "citation_id": "1Ad3UOefc" - }, - "arxiv:1507.01239": { - "source": "arxiv", - "identifer": "1507.01239", - "standard_citation": "arxiv:1507.01239", - "bibtex": "@article{aClNvbyM,\n abstract = {In this work we apply model averaging to parallel training of deep neural\nnetwork (DNN). Parallelization is done in a model averaging manner. Data is\npartitioned and distributed to different nodes for local model updates, and\nmodel averaging across nodes is done every few minibatches. We use multiple\nGPUs for data parallelization, and Message Passing Interface (MPI) for\ncommunication between nodes, which allows us to perform model averaging\nfrequently without losing much time on communication. We investigate the\neffectiveness of Natural Gradient Stochastic Gradient Descent (NG-SGD) and\nRestricted Boltzmann Machine (RBM) pretraining for parallel training in\nmodel-averaging framework, and explore the best setups in term of different\nlearning rate schedules, averaging frequencies and minibatch sizes. It is shown\nthat NG-SGD and RBM pretraining benefits parameter-averaging based model\ntraining. On the 300h Switchboard dataset, a 9.3 times speedup is achieved\nusing 16 GPUs and 17 times speedup using 32 GPUs with limited decoding accuracy\nloss.},\n archiveprefix = {arXiv},\n author = {Hang Su and Haoyu Chen},\n eprint = {1507.01239v2},\n file = {1507.01239v2.pdf},\n month = {Jul},\n primaryclass = {cs.LG},\n title = {Experiments on Parallel Training of Deep Neural Network using Model\nAveraging},\n url = {https://arxiv.org/abs/1507.01239v2},\n year = {2015}\n}\n\n", - "citation_id": "aClNvbyM" - }, - "doi:10.1021/acs.jcim.6b00290": { - "source": "doi", - "identifer": "10.1021/acs.jcim.6b00290", - "standard_citation": "doi:10.1021/acs.jcim.6b00290", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 27 - ] - ], - "date-time": "2017-09-27T11:42:20Z", - "timestamp": 1506512540065 - }, - "reference-count": 79, - "publisher": "American Chemical Society (ACS)", - "issue": "10", - "funder": [ - { - "DOI": "10.13039/100005883", - "name": "Hertz Foundation", - "doi-asserted-by": "publisher", - "award": [] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 10, - 24 - ] - ] - }, - "DOI": "10.1021/acs.jcim.6b00290", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 9, - 30 - ] - ], - "date-time": "2016-09-30T22:38:21Z", - "timestamp": 1475275101000 - }, - "page": "1936-1949", - "source": "Crossref", - "is-referenced-by-count": 4, - "title": "Computational Modeling of β-Secretase 1 (BACE-1) Inhibitors Using Ligand Based Approaches", - "prefix": "10.1021", - "volume": "56", - "author": [ - { - "given": "Govindan", - "family": "Subramanian", - "affiliation": [] - }, - { - "given": "Bharath", - "family": "Ramsundar", - "affiliation": [] - }, - { - "given": "Vijay", - "family": "Pande", - "affiliation": [] - }, - { - "given": "Rajiah Aldrin", - "family": "Denny", - "affiliation": [] - } - ], - "member": "316", - "container-title": "Journal of Chemical Information and Modeling", - "original-title": [], - "link": [ - { - "URL": "http://pubs.acs.org/doi/pdf/10.1021/acs.jcim.6b00290", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T00:06:58Z", - "timestamp": 1498349218000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 10, - 24 - ] - ] - }, - "references-count": 79, - "alternative-id": [ - "10.1021/acs.jcim.6b00290" - ], - "URL": "https://doi.org/10.1021/acs.jcim.6b00290", - "relation": {}, - "subject": [ - "General Chemistry", - "General Chemical Engineering", - "Library and Information Sciences", - "Computer Science Applications" - ], - "container-title-short": "J. Chem. Inf. Model.", - "id": "B4cL1o2P" - }, - "citation_id": "B4cL1o2P" - }, - "arxiv:1606.00575": { - "source": "arxiv", - "identifer": "1606.00575", - "standard_citation": "arxiv:1606.00575", - "bibtex": "@article{JUF9VoRD,\n abstract = {Parallelization framework has become a necessity to speed up the training of\ndeep neural networks (DNN) recently. Such framework typically employs the Model\nAverage approach, denoted as MA-DNN, in which parallel workers conduct\nrespective training based on their own local data while the parameters of local\nmodels are periodically communicated and averaged to obtain a global model\nwhich serves as the new start of local models. However, since DNN is a highly\nnon-convex model, averaging parameters cannot ensure that such global model can\nperform better than those local models. To tackle this problem, we introduce a\nnew parallel training framework called Ensemble-Compression, denoted as EC-DNN.\nIn this framework, we propose to aggregate the local models by ensemble, i.e., averaging the outputs of local models instead of the parameters. As most of\nprevalent loss functions are convex to the output of DNN, the performance of\nensemble-based global model is guaranteed to be at least as good as the average\nperformance of local models. However, a big challenge lies in the explosion of\nmodel size since each round of ensemble can give rise to multiple times size\nincrement. Thus, we carry out model compression after each ensemble, specialized by a distillation based method in this paper, to reduce the size of\nthe global model to be the same as the local ones. Our experimental results\ndemonstrate the prominent advantage of EC-DNN over MA-DNN in terms of both\naccuracy and speedup.},\n archiveprefix = {arXiv},\n author = {Shizhao Sun and Wei Chen and Jiang Bian and Xiaoguang Liu and Tie-Yan Liu},\n eprint = {1606.00575v2},\n file = {1606.00575v2.pdf},\n month = {Jun},\n primaryclass = {cs.DC},\n title = {Ensemble-Compression: A New Method for Parallel Training of Deep Neural\nNetworks},\n url = {https://arxiv.org/abs/1606.00575v2},\n year = {2016}\n}\n\n", - "citation_id": "JUF9VoRD" - }, - "arxiv:1703.01365": { - "source": "arxiv", - "identifer": "1703.01365", - "standard_citation": "arxiv:1703.01365", - "bibtex": "@article{WzFOJBiA,\n abstract = {We study the problem of attributing the prediction of a deep network to its\ninput features, a problem previously studied by several other works. We\nidentify two fundamental axioms---Sensitivity and Implementation Invariance\nthat attribution methods ought to satisfy. We show that they are not satisfied\nby most known attribution methods, which we consider to be a fundamental\nweakness of those methods. We use the axioms to guide the design of a new\nattribution method called Integrated Gradients. Our method requires no\nmodification to the original network and is extremely simple to implement; it\njust needs a few calls to the standard gradient operator. We apply this method\nto a couple of image models, a couple of text models and a chemistry model, demonstrating its ability to debug networks, to extract rules from a network, and to enable users to engage with models better.},\n archiveprefix = {arXiv},\n author = {Mukund Sundararajan and Ankur Taly and Qiqi Yan},\n eprint = {1703.01365v2},\n file = {1703.01365v2.pdf},\n month = {Mar},\n primaryclass = {cs.LG},\n title = {Axiomatic Attribution for Deep Networks},\n url = {https://arxiv.org/abs/1703.01365v2},\n year = {2017}\n}\n\n", - "citation_id": "WzFOJBiA" - }, - "arxiv:1409.3215": { - "source": "arxiv", - "identifer": "1409.3215", - "standard_citation": "arxiv:1409.3215", - "bibtex": "@article{2cMhMv5A,\n abstract = {Deep Neural Networks (DNNs) are powerful models that have achieved excellent\nperformance on difficult learning tasks. Although DNNs work well whenever large\nlabeled training sets are available, they cannot be used to map sequences to\nsequences. In this paper, we present a general end-to-end approach to sequence\nlearning that makes minimal assumptions on the sequence structure. Our method\nuses a multilayered Long Short-Term Memory (LSTM) to map the input sequence to\na vector of a fixed dimensionality, and then another deep LSTM to decode the\ntarget sequence from the vector. Our main result is that on an English to\nFrench translation task from the WMT'14 dataset, the translations produced by\nthe LSTM achieve a BLEU score of 34.8 on the entire test set, where the LSTM's\nBLEU score was penalized on out-of-vocabulary words. Additionally, the LSTM did\nnot have difficulty on long sentences. For comparison, a phrase-based SMT\nsystem achieves a BLEU score of 33.3 on the same dataset. When we used the LSTM\nto rerank the 1000 hypotheses produced by the aforementioned SMT system, its\nBLEU score increases to 36.5, which is close to the previous best result on\nthis task. The LSTM also learned sensible phrase and sentence representations\nthat are sensitive to word order and are relatively invariant to the active and\nthe passive voice. Finally, we found that reversing the order of the words in\nall source sentences (but not target sentences) improved the LSTM's performance\nmarkedly, because doing so introduced many short term dependencies between the\nsource and the target sentence which made the optimization problem easier.},\n archiveprefix = {arXiv},\n author = {Ilya Sutskever and Oriol Vinyals and Quoc V. Le},\n eprint = {1409.3215v3},\n file = {1409.3215v3.pdf},\n month = {Sep},\n primaryclass = {cs.CL},\n title = {Sequence to Sequence Learning with Neural Networks},\n url = {https://arxiv.org/abs/1409.3215v3},\n year = {2014}\n}\n\n", - "citation_id": "2cMhMv5A" - }, - "doi:10.1021/ci8004379": { - "source": "doi", - "identifer": "10.1021/ci8004379", - "standard_citation": "doi:10.1021/ci8004379", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 14 - ] - ], - "date-time": "2017-08-14T12:22:21Z", - "timestamp": 1502713341255 - }, - "reference-count": 0, - "publisher": "American Chemical Society (ACS)", - "issue": "4", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2009, - 4, - 27 - ] - ] - }, - "DOI": "10.1021/ci8004379", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2009, - 3, - 26 - ] - ], - "date-time": "2009-03-26T12:05:55Z", - "timestamp": 1238069155000 - }, - "page": "756-766", - "source": "Crossref", - "is-referenced-by-count": 19, - "title": "Influence Relevance Voting: An Accurate And Interpretable Virtual High Throughput Screening Method", - "prefix": "10.1021", - "volume": "49", - "author": [ - { - "given": "S. Joshua", - "family": "Swamidass", - "affiliation": [] - }, - { - "given": "Chloé-Agathe", - "family": "Azencott", - "affiliation": [] - }, - { - "given": "Ting-Wan", - "family": "Lin", - "affiliation": [] - }, - { - "given": "Hugo", - "family": "Gramajo", - "affiliation": [] - }, - { - "given": "Shiou-Chuan", - "family": "Tsai", - "affiliation": [] - }, - { - "given": "Pierre", - "family": "Baldi", - "affiliation": [] - } - ], - "member": "316", - "container-title": "Journal of Chemical Information and Modeling", - "original-title": [], - "link": [ - { - "URL": "http://pubs.acs.org/doi/pdf/10.1021/ci8004379", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 9, - 2 - ] - ], - "date-time": "2016-09-02T02:55:56Z", - "timestamp": 1472784956000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2009, - 4, - 27 - ] - ] - }, - "references-count": 0, - "alternative-id": [ - "10.1021/ci8004379" - ], - "URL": "https://doi.org/10.1021/ci8004379", - "relation": {}, - "subject": [ - "General Chemistry", - "General Chemical Engineering", - "Library and Information Sciences", - "Computer Science Applications" - ], - "container-title-short": "J. Chem. Inf. Model.", - "id": "cjj5vT3H" - }, - "citation_id": "cjj5vT3H" - }, - "doi:10.1142/s0219720012500151": { - "source": "doi", - "identifer": "10.1142/s0219720012500151", - "standard_citation": "doi:10.1142/s0219720012500151", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 1 - ] - ], - "date-time": "2017-09-01T02:02:16Z", - "timestamp": 1504231336923 - }, - "reference-count": 30, - "publisher": "World Scientific Pub Co Pte Lt", - "issue": "05", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2012, - 10 - ] - ] - }, - "DOI": "10.1142/s0219720012500151", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2012, - 6, - 6 - ] - ], - "date-time": "2012-06-06T00:51:43Z", - "timestamp": 1338943903000 - }, - "page": "1250015", - "source": "Crossref", - "is-referenced-by-count": 13, - "title": "METAGENOMIC TAXONOMIC CLASSIFICATION USING EXTREME LEARNING MACHINES", - "prefix": "10.1142", - "volume": "10", - "author": [ - { - "given": "ZEEHASHAM", - "family": "RASHEED", - "affiliation": [] - }, - { - "given": "HUZEFA", - "family": "RANGWALA", - "affiliation": [] - } - ], - "member": "219", - "container-title": "Journal of Bioinformatics and Computational Biology", - "original-title": [], - "link": [ - { - "URL": "http://www.worldscientific.com/doi/pdf/10.1142/S0219720012500151", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 12, - 4 - ] - ], - "date-time": "2016-12-04T12:56:44Z", - "timestamp": 1480856204000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2012, - 10 - ] - ] - }, - "references-count": 30, - "alternative-id": [ - "10.1142/S0219720012500151" - ], - "URL": "https://doi.org/10.1142/s0219720012500151", - "relation": { - "cites": [] - }, - "subject": [ - "Biochemistry", - "Molecular Biology", - "Computer Science Applications" - ], - "container-title-short": "J. Bioinform. Comput. Biol.", - "id": "Wz7VUS03" - }, - "citation_id": "Wz7VUS03" - }, - "doi:10.1128/msystems.00025-15": { - "source": "doi", - "identifer": "10.1128/msystems.00025-15", - "standard_citation": "doi:10.1128/msystems.00025-15", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 11 - ] - ], - "date-time": "2017-08-11T08:06:06Z", - "timestamp": 1502438766597 - }, - "reference-count": 0, - "publisher": "American Society for Microbiology", - "issue": "1", - "funder": [ - { - "DOI": "10.13039/100000002", - "name": "National Institutes of Health", - "doi-asserted-by": "publisher", - "award": [ - "AI091702", - "CA023108", - "GM106394", - "DK007301" - ] - }, - { - "DOI": "10.13039/100000936", - "name": "Gordon and Betty Moore Foundation", - "doi-asserted-by": "publisher", - "award": [ - "GBMF4552" - ] - }, - { - "DOI": "10.13039/100000897", - "name": "Cystic Fibrosis Foundation", - "doi-asserted-by": "publisher", - "award": [ - "STANTO07R0", - "STANTO15R0" - ] - }, - { - "name": "William H. Neukom Institute for Computational Science", - "award": [] - } - ], - "content-domain": { - "domain": [ - "asm.org" - ], - "crossmark-restriction": true - }, - "published-print": { - "date-parts": [ - [ - 2016, - 2, - 23 - ] - ] - }, - "abstract": "ABSTRACTThe increasing number of genome-wide assays of gene expression available from public databases presents opportunities for computational methods that facilitate hypothesis generation and biological interpretation of these data. We present an unsupervised machine learning approach, ADAGE (analysis usingdenoisingautoencoders ofgeneexpression), and apply it to the publicly available gene expression data compendium forPseudomonas aeruginosa. In this approach, the machine-learned ADAGE model contained 50 nodes which we predicted would correspond to gene expression patterns across the gene expression compendium. While no biological knowledge was used during model construction, cooperonic genes had similar weights across nodes, and genes with similar weights across nodes were significantly more likely to share KEGG pathways. By analyzing newly generated and previously published microarray and transcriptome sequencing data, the ADAGE model identified differences between strains, modeled the cellular response to low oxygen, and predicted the involvement of biological processes based on low-level gene expression differences. ADAGE compared favorably with traditional principal component analysis and independent component analysis approaches in its ability to extract validated patterns, and based on our analyses, we propose that these approaches differ in the types of patterns they preferentially identify. We provide the ADAGE model with analysis of all publicly availableP. aeruginosaGeneChip experiments and open source code for use with other species and settings. Extraction of consistent patterns across large-scale collections of genomic data using methods like ADAGE provides the opportunity to identify general principles and biologically important patterns in microbial biology. This approach will be particularly useful in less-well-studied microbial species.IMPORTANCEThe quantity and breadth of genome-scale data sets that examine RNA expression in diverse bacterial and eukaryotic species are increasing more rapidly than for curated knowledge. Our ADAGE method integrates such data without requiring gene function, gene pathway, or experiment labeling, making practical its application to any large gene expression compendium. We built aPseudomonas aeruginosaADAGE model from a diverse set of publicly available experiments without any prespecified biological knowledge, and this model was accurate and predictive. We provide ADAGE results for the completeP. aeruginosaGeneChip compendium for use by researchers studyingP. aeruginosaand source code that facilitates ADAGE’s application to other species and data types.", - "DOI": "10.1128/msystems.00025-15", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 1, - 13 - ] - ], - "date-time": "2016-01-13T17:45:18Z", - "timestamp": 1452707118000 - }, - "page": "e00025-15", - "update-policy": "http://dx.doi.org/10.1128/asmj-crossmark-policy-page", - "source": "Crossref", - "is-referenced-by-count": 3, - "title": "ADAGE-Based Integration of Publicly AvailablePseudomonas aeruginosaGene Expression Data with Denoising Autoencoders Illuminates Microbe-Host Interactions", - "prefix": "10.1128", - "volume": "1", - "author": [ - { - "given": "Jie", - "family": "Tan", - "affiliation": [] - }, - { - "given": "John H.", - "family": "Hammond", - "affiliation": [] - }, - { - "given": "Deborah A.", - "family": "Hogan", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0001-8713-9213", - "authenticated-orcid": false, - "given": "Casey S.", - "family": "Greene", - "affiliation": [] - } - ], - "member": "235", - "published-online": { - "date-parts": [ - [ - 2016, - 1, - 19 - ] - ] - }, - "container-title": "mSystems", - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1128/mSystems.00025-15", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 12, - 23 - ] - ], - "date-time": "2016-12-23T17:42:10Z", - "timestamp": 1482514930000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Jack A.", - "family": "Gilbert", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 1, - 19 - ] - ] - }, - "references-count": 0, - "alternative-id": [ - "10.1128/mSystems.00025-15" - ], - "URL": "https://doi.org/10.1128/msystems.00025-15", - "relation": { - "has-preprint": [ - { - "id-type": "doi", - "id": "10.1101/030650", - "asserted-by": "object" - } - ] - }, - "container-title-short": "mSystems", - "id": "1CFhfCyWN" - }, - "citation_id": "1CFhfCyWN" - }, - "doi:10.1101/078659": { - "source": "doi", - "identifer": "10.1101/078659", - "standard_citation": "doi:10.1101/078659", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T00:21:24Z", - "timestamp": 1502324484330 - }, - "posted": { - "date-parts": [ - [ - 2016, - 10, - 3 - ] - ] - }, - "group-title": "Systems Biology", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2017, - 4, - 10 - ] - ] - }, - "abstract": "Cross experiment comparisons in public data compendia are challenged by unmatched conditions and technical noise. The ADAGE method, which performs unsupervised integration with neural networks, can effectively identify biological patterns, but because ADAGE models, like many neural networks, are over-parameterized, different ADAGE models perform equally well. To enhance model robustness and better build signatures consistent with biological pathways, we developed an ensemble ADAGE (eADAGE) that integrated stable signatures across models. We applied eADAGE to a Pseudomonas aeruginosa compendium containing experiments performed in 78 media. eADAGE revealed a phosphate starvation response controlled by PhoB. While we expected PhoB activity in limiting phosphate conditions, our analyses found PhoB activity in other media with moderate phosphate and predicted that a second stimulus provided by the sensor kinase, KinB, is required for PhoB activation in this setting. We validated this relationship using both targeted and unbiased genetic approaches. eADAGE, which captures stable biological patterns, enables cross-experiment comparisons that can highlight measured but undiscovered relationships.", - "DOI": "10.1101/078659", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2016, - 10, - 4 - ] - ], - "date-time": "2016-10-04T05:19:54Z", - "timestamp": 1475558394000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Unsupervised extraction of stable expression signatures from public compendia with eADAGE", - "prefix": "10.1101", - "author": [ - { - "ORCID": "http://orcid.org/0000-0002-8893-4566", - "authenticated-orcid": false, - "given": "Jie", - "family": "Tan", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-0835-6955", - "authenticated-orcid": false, - "given": "Georgia", - "family": "Doing", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0003-3010-8453", - "authenticated-orcid": false, - "given": "Kimberley A", - "family": "Lewis", - "affiliation": [] - }, - { - "given": "Courtney E", - "family": "Price", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-7461-9530", - "authenticated-orcid": false, - "given": "Kathleen M", - "family": "Chen", - "affiliation": [] - }, - { - "given": "Kyle C", - "family": "Cady", - "affiliation": [] - }, - { - "given": "Barret", - "family": "Perchuk", - "affiliation": [] - }, - { - "given": "Michael T", - "family": "Laub", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0002-6366-2971", - "authenticated-orcid": false, - "given": "Deborah A", - "family": "Hogan", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0001-8713-9213", - "authenticated-orcid": false, - "given": "Casey S", - "family": "Greene", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/078659", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 7, - 20 - ] - ], - "date-time": "2017-07-20T18:55:34Z", - "timestamp": 1500576934000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 10, - 3 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/078659", - "relation": { - "is-preprint-of": [ - { - "id-type": "doi", - "id": "10.1016/j.cels.2017.06.003", - "asserted-by": "subject" - } - ] - }, - "id": "zuLdSQx3" - }, - "citation_id": "zuLdSQx3" - }, - "arxiv:1603.04467": { - "source": "arxiv", - "identifer": "1603.04467", - "standard_citation": "arxiv:1603.04467", - "bibtex": "@article{hOeUlCvS,\n abstract = {TensorFlow is an interface for expressing machine learning algorithms, and an\nimplementation for executing such algorithms. A computation expressed using\nTensorFlow can be executed with little or no change on a wide variety of\nheterogeneous systems, ranging from mobile devices such as phones and tablets\nup to large-scale distributed systems of hundreds of machines and thousands of\ncomputational devices such as GPU cards. The system is flexible and can be used\nto express a wide variety of algorithms, including training and inference\nalgorithms for deep neural network models, and it has been used for conducting\nresearch and for deploying machine learning systems into production across more\nthan a dozen areas of computer science and other fields, including speech\nrecognition, computer vision, robotics, information retrieval, natural language\nprocessing, geographic information extraction, and computational drug\ndiscovery. This paper describes the TensorFlow interface and an implementation\nof that interface that we have built at Google. The TensorFlow API and a\nreference implementation were released as an open-source package under the\nApache 2.0 license in November, 2015 and are available at www.tensorflow.org.},\n archiveprefix = {arXiv},\n author = {Martín Abadi and Ashish Agarwal and Paul Barham and Eugene Brevdo and Zhifeng Chen and Craig Citro and Greg S. Corrado and Andy Davis and Jeffrey Dean and Matthieu Devin and Sanjay Ghemawat and Ian Goodfellow and Andrew Harp and Geoffrey Irving and Michael Isard and Yangqing Jia and Rafal Jozefowicz and Lukasz Kaiser and Manjunath Kudlur and Josh Levenberg and Dan Mane and Rajat Monga and Sherry Moore and Derek Murray and Chris Olah and Mike Schuster and Jonathon Shlens and Benoit Steiner and Ilya Sutskever and Kunal Talwar and Paul Tucker and Vincent Vanhoucke and Vijay Vasudevan and Fernanda Viegas and Oriol Vinyals and Pete Warden and Martin Wattenberg and Martin Wicke and Yuan Yu and Xiaoqiang Zheng},\n eprint = {1603.04467v2},\n file = {1603.04467v2.pdf},\n month = {Mar},\n primaryclass = {cs.DC},\n title = {TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed\nSystems},\n url = {https://arxiv.org/abs/1603.04467v2},\n year = {2016}\n}\n\n", - "citation_id": "hOeUlCvS" - }, - "doi:10.1101/097469": { - "source": "doi", - "identifer": "10.1101/097469", - "standard_citation": "doi:10.1101/097469", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T13:55:37Z", - "timestamp": 1502373337966 - }, - "posted": { - "date-parts": [ - [ - 2016, - 12, - 30 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2016, - 12, - 30 - ] - ] - }, - "abstract": "We present an open source software toolkit for training deep learning models to call genotypes in high-throughput sequencing data. The software supports SAM, BAM, CRAM and Goby alignments and the training of models for a variety of experimental assays and analysis protocols. We evaluate this software in the Illumina platinum whole genome datasets and find that a deep learning model trained on 80% of the genome achieves a 0.986% accuracy on variants (genotype concordance) when trained with 10% of the data from a genome. The software is distributed at https://github.com/CampagneLaboratory/variationanalysis. The software makes it possible to train genotype calling models on consumer hardware with CPUs or GPU(s). It will enable individual investigators and small laboratories to train and evaluate their own models and to make open source contributions. We welcome contributions to extend this early prototype or evaluate its performance on other gold standard datasets.", - "DOI": "10.1101/097469", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2016, - 12, - 31 - ] - ], - "date-time": "2016-12-31T06:10:12Z", - "timestamp": 1483164612000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Training Genotype Callers with Neural Networks", - "prefix": "10.1101", - "author": [ - { - "given": "Rémi", - "family": "Torracinta", - "affiliation": [] - }, - { - "ORCID": "http://orcid.org/0000-0001-6237-3564", - "authenticated-orcid": false, - "given": "Fabien", - "family": "Campagne", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/097469", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 12, - 31 - ] - ], - "date-time": "2016-12-31T06:10:14Z", - "timestamp": 1483164614000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 12, - 30 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/097469", - "relation": {}, - "id": "GSLRw2L5" - }, - "citation_id": "GSLRw2L5" - }, - "doi:10.1101/079087": { - "source": "doi", - "identifer": "10.1101/079087", - "standard_citation": "doi:10.1101/079087", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T00:30:15Z", - "timestamp": 1502325015289 - }, - "posted": { - "date-parts": [ - [ - 2016, - 10, - 4 - ] - ] - }, - "group-title": "Bioinformatics", - "reference-count": 0, - "publisher": "Cold Spring Harbor Laboratory", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "accepted": { - "date-parts": [ - [ - 2016, - 10, - 4 - ] - ] - }, - "abstract": "A number of approaches have been developed to call somatic variation in high-throughput sequencing data. Here, we present an adaptive approach to calling somatic variations. Our approach trains a deep feed-forward neural network with semi-simulated data. Semi-simulated datasets are constructed by planting somatic mutations in real datasets where no mutations are expected. Using semi-simulated data makes it possible to train the models with millions of training examples, a usual requirement for successfully training deep learning models. We initially focus on calling variations in RNA-Seq data. We derive semi-simulated datasets from real RNA-Seq data, which offer a good representation of the data the models will be applied to. We test the models on independent semi-simulated data as well as pure simulations. On independent semi-simulated data, models achieve an AUC of 0.973. When tested on semi-simulated exome DNA datasets, we find that the models trained on RNA-Seq data remain predictive (sens ~0.4 & spec ~0.9 at cutoff of P>=0.9), albeit with lower overall performance (AUC=0.737). Interestingly, while the models generalize across assay, training on RNA-Seq data lowers the confidence for a group of mutations. Haloplex exome specific training was also performed, demonstrating that the approach can produce probabilistic models tuned for specific assays and protocols. We found that the method adapts to the characteristics of experimental protocol. We further illustrate these points by training a model for a trio somatic experimental design when germline DNA of both parents is available in addition to data about the individual. These models are distributed with Goby (http://goby.campagnelab.org).", - "DOI": "10.1101/079087", - "type": "manuscript", - "created": { - "date-parts": [ - [ - 2016, - 10, - 5 - ] - ], - "date-time": "2016-10-05T05:12:27Z", - "timestamp": 1475644347000 - }, - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Adaptive Somatic Mutations Calls with Deep Learning and Semi-Simulated Data", - "prefix": "10.1101", - "author": [ - { - "given": "Remi", - "family": "Torracinta", - "affiliation": [] - }, - { - "given": "Laurent", - "family": "Mesnard", - "affiliation": [] - }, - { - "given": "Susan", - "family": "Levine", - "affiliation": [] - }, - { - "given": "Rita", - "family": "Shaknovich", - "affiliation": [] - }, - { - "given": "Maureen", - "family": "Hanson", - "affiliation": [] - }, - { - "given": "Fabien", - "family": "Campagne", - "affiliation": [] - } - ], - "member": "246", - "container-title": [], - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1101/079087", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 1, - 4 - ] - ], - "date-time": "2017-01-04T06:15:35Z", - "timestamp": 1483510535000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 10, - 4 - ] - ] - }, - "references-count": 0, - "URL": "https://doi.org/10.1101/079087", - "relation": {}, - "id": "ECTm1SuA" - }, - "citation_id": "ECTm1SuA" - }, - "doi:10.1016/s0895-4356(96)00002-9": { - "source": "doi", - "identifer": "10.1016/s0895-4356(96)00002-9", - "standard_citation": "doi:10.1016/s0895-4356(96)00002-9", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T11:22:17Z", - "timestamp": 1507116137284 - }, - "reference-count": 37, - "publisher": "Elsevier BV", - "issue": "11", - "license": [ - { - "URL": "http://www.elsevier.com/tdm/userlicense/1.0/", - "start": { - "date-parts": [ - [ - 1996, - 11, - 1 - ] - ], - "date-time": "1996-11-01T00:00:00Z", - "timestamp": 846806400000 - }, - "delay-in-days": 0, - "content-version": "tdm" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 1996, - 11 - ] - ] - }, - "DOI": "10.1016/s0895-4356(96)00002-9", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2002, - 7, - 25 - ] - ], - "date-time": "2002-07-25T18:50:39Z", - "timestamp": 1027623039000 - }, - "page": "1225-1231", - "source": "Crossref", - "is-referenced-by-count": 404, - "title": "Advantages and disadvantages of using artificial neural networks versus logistic regression for predicting medical outcomes", - "prefix": "10.1016", - "volume": "49", - "author": [ - { - "given": "Jack V.", - "family": "Tu", - "affiliation": [] - } - ], - "member": "78", - "container-title": "Journal of Clinical Epidemiology", - "original-title": [], - "link": [ - { - "URL": "http://api.elsevier.com/content/article/PII:S0895435696000029?httpAccept=text/xml", - "content-type": "text/xml", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.elsevier.com/content/article/PII:S0895435696000029?httpAccept=text/plain", - "content-type": "text/plain", - "content-version": "vor", - "intended-application": "text-mining" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 14 - ] - ], - "date-time": "2017-06-14T23:38:27Z", - "timestamp": 1497483507000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 1996, - 11 - ] - ] - }, - "references-count": 37, - "alternative-id": [ - "S0895435696000029" - ], - "URL": "https://doi.org/10.1016/s0895-4356(96)00002-9", - "relation": {}, - "subject": [ - "Epidemiology" - ], - "container-title-short": "Journal of Clinical Epidemiology", - "id": "kL0B4m9d" - }, - "citation_id": "kL0B4m9d" - }, - "url:http://www.bioinf.at/publications/2014/NIPS2014a.pdf": { - "source": "url", - "identifer": "http://www.bioinf.at/publications/2014/NIPS2014a.pdf", - "standard_citation": "url:http://www.bioinf.at/publications/2014/NIPS2014a.pdf", - "citeproc": { - "type": "article-journal", - "title": "Deep learning as an opportunity in virtual screening", - "container-title": "Neural Information Processing Systems 2014: Deep Learning and Representation Learning Workshop", - "URL": "http://www.dlworkshop.org/23.pdf?attredirects=0", - "author": [ - { - "family": "Unterthiner", - "given": "Thomas" - }, - { - "family": "Mayr", - "given": "Andreas" - }, - { - "family": "Klambauer", - "given": "Günter" - }, - { - "family": "Steijaert", - "given": "Marvin" - }, - { - "family": "Wegner", - "given": "Jörg K." - }, - { - "family": "Ceulemans", - "given": "Hugo" - }, - { - "family": "Hochreiter", - "given": "Sepp" - } - ], - "issued": { - "date-parts": [ - [ - "2014" - ] - ] - }, - "accessed": { - "date-parts": [ - [ - "2017", - 5, - 19 - ] - ] - }, - "id": "F8fP2vAg" - }, - "citation_id": "F8fP2vAg" - }, - "url:https://research.google.com/pubs/pub37631.html": { - "source": "url", - "identifer": "https://research.google.com/pubs/pub37631.html", - "standard_citation": "url:https://research.google.com/pubs/pub37631.html", - "citeproc": { - "URL": "https://research.google.com/pubs/pub37631.html", - "title": "Improving the speed of neural networks on CPUs", - "issued": { - "date-parts": [ - [ - 2011 - ] - ] - }, - "author": [ - { - "family": "Vanhoucke", - "given": "Vincent" - }, - { - "family": "Senior", - "given": "Andrew" - }, - { - "family": "Mao", - "given": "Mark Z." - } - ], - "greycite-status": "Scanned", - "greycite-scanned": "2017-05-17 02:13:17", - "type": "webpage", - "id": "NSgduYNT" - }, - "citation_id": "NSgduYNT" - }, - "doi:10.1146/annurev-genet-120215-034854": { - "source": "doi", - "identifer": "10.1146/annurev-genet-120215-034854", - "standard_citation": "doi:10.1146/annurev-genet-120215-034854", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T04:15:31Z", - "timestamp": 1502338531614 - }, - "reference-count": 160, - "publisher": "Annual Reviews", - "issue": "1", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 11, - 23 - ] - ] - }, - "DOI": "10.1146/annurev-genet-120215-034854", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 11, - 29 - ] - ], - "date-time": "2016-11-29T00:34:19Z", - "timestamp": 1480379659000 - }, - "page": "267-291", - "source": "Crossref", - "is-referenced-by-count": 2, - "title": "Single-Cell and Single-Molecule Analysis of Gene Expression Regulation", - "prefix": "10.1146", - "volume": "50", - "author": [ - { - "given": "Maria", - "family": "Vera", - "affiliation": [ - { - "name": "Department of Anatomy and Structural Biology, Albert Einstein College of Medicine, New York, NY 10461; email: , , , " - } - ] - }, - { - "given": "Jeetayu", - "family": "Biswas", - "affiliation": [ - { - "name": "Department of Anatomy and Structural Biology, Albert Einstein College of Medicine, New York, NY 10461; email: , , , " - } - ] - }, - { - "given": "Adrien", - "family": "Senecal", - "affiliation": [ - { - "name": "Department of Anatomy and Structural Biology, Albert Einstein College of Medicine, New York, NY 10461; email: , , , " - } - ] - }, - { - "given": "Robert H.", - "family": "Singer", - "affiliation": [ - { - "name": "Department of Anatomy and Structural Biology, Albert Einstein College of Medicine, New York, NY 10461; email: , , , " - }, - { - "name": "Janelia Research Campus of the HHMI, Ashburn, Virginia 20147" - } - ] - }, - { - "given": "Hye Yoon", - "family": "Park", - "affiliation": [ - { - "name": "Department of Physics and Astronomy, Seoul National University, Seoul, 08826, Korea; email: " - }, - { - "name": "Institute of Molecular Biology and Genetics, Seoul National University, Seoul, 08826, Korea" - } - ] - } - ], - "member": "22", - "container-title": "Annual Review of Genetics", - "original-title": [], - "link": [ - { - "URL": "http://www.annualreviews.org/doi/pdf/10.1146/annurev-genet-120215-034854", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 1, - 16 - ] - ], - "date-time": "2017-01-16T12:30:17Z", - "timestamp": 1484569817000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 11, - 23 - ] - ] - }, - "references-count": 160, - "alternative-id": [ - "10.1146/annurev-genet-120215-034854" - ], - "URL": "https://doi.org/10.1146/annurev-genet-120215-034854", - "relation": {}, - "subject": [ - "Genetics" - ], - "container-title-short": "Annu. Rev. Genet.", - "id": "v97iPXDw" - }, - "citation_id": "v97iPXDw" - }, - "doi:10.1093/bioinformatics/btv683": { - "source": "doi", - "identifer": "10.1093/bioinformatics/btv683", - "standard_citation": "doi:10.1093/bioinformatics/btv683", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 24 - ] - ], - "date-time": "2017-08-24T16:12:04Z", - "timestamp": 1503591124041 - }, - "reference-count": 31, - "publisher": "Oxford University Press (OUP)", - "issue": "7", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 4, - 1 - ] - ] - }, - "DOI": "10.1093/bioinformatics/btv683", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 11, - 21 - ] - ], - "date-time": "2015-11-21T02:29:46Z", - "timestamp": 1448072986000 - }, - "page": "1023-1032", - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Large-scale machine learning for metagenomics sequence classification", - "prefix": "10.1093", - "volume": "32", - "author": [ - { - "given": "Kévin", - "family": "Vervier", - "affiliation": [] - }, - { - "given": "Pierre", - "family": "Mahé", - "affiliation": [] - }, - { - "given": "Maud", - "family": "Tournoud", - "affiliation": [] - }, - { - "given": "Jean-Baptiste", - "family": "Veyrieras", - "affiliation": [] - }, - { - "given": "Jean-Philippe", - "family": "Vert", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2015, - 11, - 20 - ] - ] - }, - "container-title": "Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/bioinformatics/article-pdf/32/7/1023/", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 24 - ] - ], - "date-time": "2017-08-24T02:29:02Z", - "timestamp": 1503541742000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 11, - 20 - ] - ] - }, - "references-count": 31, - "alternative-id": [ - "10.1093/bioinformatics/btv683" - ], - "URL": "https://doi.org/10.1093/bioinformatics/btv683", - "relation": {}, - "subject": [ - "Statistics and Probability", - "Computational Theory and Mathematics", - "Biochemistry", - "Molecular Biology", - "Computational Mathematics", - "Computer Science Applications" - ], - "container-title-short": "Bioinformatics", - "id": "8DLzxOEt" - }, - "citation_id": "8DLzxOEt" - }, - "doi:10.1056/nejm198509263131306": { - "source": "doi", - "identifer": "10.1056/nejm198509263131306", - "standard_citation": "doi:10.1056/nejm198509263131306", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 13 - ] - ], - "date-time": "2017-09-13T11:42:13Z", - "timestamp": 1505302933663 - }, - "reference-count": 54, - "publisher": "New England Journal of Medicine (NEJM/MMS)", - "issue": "13", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 1985, - 9, - 26 - ] - ] - }, - "DOI": "10.1056/nejm198509263131306", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2010, - 7, - 28 - ] - ], - "date-time": "2010-07-28T21:32:51Z", - "timestamp": 1280352771000 - }, - "page": "793-799", - "source": "Crossref", - "is-referenced-by-count": 592, - "title": "Clinical Prediction Rules", - "prefix": "10.1056", - "volume": "313", - "author": [ - { - "given": "John H.", - "family": "Wasson", - "affiliation": [] - }, - { - "given": "Harold C.", - "family": "Sox", - "affiliation": [] - }, - { - "given": "Raymond K.", - "family": "Neff", - "affiliation": [] - }, - { - "given": "Lee", - "family": "Goldman", - "affiliation": [] - } - ], - "member": "150", - "container-title": "New England Journal of Medicine", - "original-title": [], - "link": [ - { - "URL": "http://www.nejm.org/doi/pdf/10.1056/NEJM198509263131306", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 19 - ] - ], - "date-time": "2017-06-19T11:47:11Z", - "timestamp": 1497872831000 - }, - "score": 1.0, - "subtitle": [ - "Applications and Methodological Standards" - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 1985, - 9, - 26 - ] - ] - }, - "references-count": 54, - "alternative-id": [ - "10.1056/NEJM198509263131306" - ], - "URL": "https://doi.org/10.1056/nejm198509263131306", - "relation": {}, - "subject": [ - "General Medicine" - ], - "container-title-short": "N Engl J Med", - "id": "xX68eyvs" - }, - "citation_id": "xX68eyvs" - }, - "arxiv:1301.3781": { - "source": "arxiv", - "identifer": "1301.3781", - "standard_citation": "arxiv:1301.3781", - "bibtex": "@article{1GhHIDxuW,\n abstract = {We propose two novel model architectures for computing continuous vector\nrepresentations of words from very large data sets. The quality of these\nrepresentations is measured in a word similarity task, and the results are\ncompared to the previously best performing techniques based on different types\nof neural networks. We observe large improvements in accuracy at much lower\ncomputational cost, i.e. it takes less than a day to learn high quality word\nvectors from a 1.6 billion words data set. Furthermore, we show that these\nvectors provide state-of-the-art performance on our test set for measuring\nsyntactic and semantic word similarities.},\n archiveprefix = {arXiv},\n author = {Tomas Mikolov and Kai Chen and Greg Corrado and Jeffrey Dean},\n eprint = {1301.3781v3},\n file = {1301.3781v3.pdf},\n month = {Jan},\n primaryclass = {cs.CL},\n title = {Efficient Estimation of Word Representations in Vector Space},\n url = {https://arxiv.org/abs/1301.3781v3},\n year = {2013}\n}\n\n", - "citation_id": "1GhHIDxuW" - }, - "arxiv:1703.00564": { - "source": "arxiv", - "identifer": "1703.00564", - "standard_citation": "arxiv:1703.00564", - "bibtex": "@article{16OPHvAij,\n abstract = {Molecular machine learning has been maturing rapidly over the last few years.\nImproved methods and the presence of larger datasets have enabled machine\nlearning algorithms to make increasingly accurate predictions about molecular\nproperties. However, algorithmic progress has been limited due to the lack of a\nstandard benchmark to compare the efficacy of proposed methods; most new\nalgorithms are benchmarked on different datasets making it challenging to gauge\nthe quality of proposed methods. This work introduces MoleculeNet, a large\nscale benchmark for molecular machine learning. MoleculeNet curates multiple\npublic datasets, establishes metrics for evaluation, and offers high quality\nopen-source implementations of multiple previously proposed molecular\nfeaturization and learning algorithms (released as part of the DeepChem open\nsource library). MoleculeNet benchmarks demonstrate that learnable\nrepresentations, and in particular graph convolutional networks, are powerful\ntools for molecular machine learning and broadly offer the best performance.\nHowever, for quantum mechanical and biophysical datasets, the use of\nphysics-aware featurizations can be significantly more important than choice of\nparticular learning algorithm.},\n archiveprefix = {arXiv},\n author = {Zhenqin Wu and Bharath Ramsundar and Evan N. Feinberg and Joseph Gomes and Caleb Geniesse and Aneesh S. Pappu and Karl Leswing and Vijay Pande},\n eprint = {1703.00564v1},\n file = {1703.00564v1.pdf},\n month = {Mar},\n primaryclass = {cs.LG},\n title = {MoleculeNet: A Benchmark for Molecular Machine Learning},\n url = {https://arxiv.org/abs/1703.00564v1},\n year = {2017}\n}\n\n", - "citation_id": "16OPHvAij" - }, - "doi:10.1093/bioinformatics/btr444": { - "source": "doi", - "identifer": "10.1093/bioinformatics/btr444", - "standard_citation": "doi:10.1093/bioinformatics/btr444", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 22 - ] - ], - "date-time": "2017-08-22T16:40:54Z", - "timestamp": 1503420054440 - }, - "reference-count": 22, - "publisher": "Oxford University Press (OUP)", - "issue": "18", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2011, - 9, - 15 - ] - ] - }, - "DOI": "10.1093/bioinformatics/btr444", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2011, - 7, - 30 - ] - ], - "date-time": "2011-07-30T02:55:41Z", - "timestamp": 1311994541000 - }, - "page": "2554-2562", - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Bayesian prediction of tissue-regulated splicing using RNA sequence and cellular context", - "prefix": "10.1093", - "volume": "27", - "author": [ - { - "given": "Hui Yuan", - "family": "Xiong", - "affiliation": [] - }, - { - "given": "Yoseph", - "family": "Barash", - "affiliation": [] - }, - { - "given": "Brendan J.", - "family": "Frey", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2011, - 7, - 29 - ] - ] - }, - "container-title": "Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/bioinformatics/article-pdf/27/18/2554/16900923/btr444.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 22 - ] - ], - "date-time": "2017-08-22T16:27:31Z", - "timestamp": 1503419251000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2011, - 7, - 29 - ] - ] - }, - "references-count": 22, - "URL": "https://doi.org/10.1093/bioinformatics/btr444", - "relation": {}, - "subject": [ - "Statistics and Probability", - "Computational Theory and Mathematics", - "Biochemistry", - "Molecular Biology", - "Computational Mathematics", - "Computer Science Applications" - ], - "id": "8VPGUHcf" - }, - "citation_id": "8VPGUHcf" - }, - "doi:10.1126/science.1254806": { - "source": "doi", - "identifer": "10.1126/science.1254806", - "standard_citation": "doi:10.1126/science.1254806", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T08:42:15Z", - "timestamp": 1507020135227 - }, - "reference-count": 78, - "publisher": "American Association for the Advancement of Science (AAAS)", - "issue": "6218", - "license": [ - { - "URL": "http://www.sciencemag.org/about/science-licenses-journal-article-reuse", - "start": { - "date-parts": [ - [ - 2016, - 1, - 9 - ] - ], - "date-time": "2016-01-09T00:00:00Z", - "timestamp": 1452297600000 - }, - "delay-in-days": 387, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/100000002", - "name": "National Institutes of Health", - "doi-asserted-by": "publisher", - "award": [ - "R37-GM42699a" - ] - }, - { - "DOI": "10.13039/100000073", - "name": "Autism Speaks", - "doi-asserted-by": "publisher", - "award": [] - }, - { - "DOI": "10.13039/100008762", - "name": "Genome Canada", - "doi-asserted-by": "publisher", - "award": [] - }, - { - "DOI": "10.13039/501100000024", - "name": "Canadian Institutes of Health Research", - "doi-asserted-by": "publisher", - "award": [] - }, - { - "DOI": "10.13039/501100000038", - "name": "Natural Sciences and Engineering Research Council of Canada", - "doi-asserted-by": "publisher", - "award": [] - }, - { - "DOI": "10.13039/501100000092", - "name": "Ontario Genomics Institute", - "doi-asserted-by": "publisher", - "award": [] - }, - { - "name": "University of Toronto McLaughlin Centre", - "award": [] - }, - { - "name": "McLaughlin", - "award": [] - }, - { - "name": "Autism Research Training Fellowship", - "award": [] - }, - { - "name": "CIHR Banting Fellowship", - "award": [] - }, - { - "name": "NSERC Alexander Graham Bell Scholarship", - "award": [] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015, - 1, - 9 - ] - ] - }, - "DOI": "10.1126/science.1254806", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 12, - 19 - ] - ], - "date-time": "2014-12-19T10:32:44Z", - "timestamp": 1418985164000 - }, - "page": "1254806-1254806", - "source": "Crossref", - "is-referenced-by-count": 184, - "title": "The human splicing code reveals new insights into the genetic determinants of disease", - "prefix": "10.1126", - "volume": "347", - "author": [ - { - "given": "H. Y.", - "family": "Xiong", - "affiliation": [] - }, - { - "given": "B.", - "family": "Alipanahi", - "affiliation": [] - }, - { - "given": "L. J.", - "family": "Lee", - "affiliation": [] - }, - { - "given": "H.", - "family": "Bretschneider", - "affiliation": [] - }, - { - "given": "D.", - "family": "Merico", - "affiliation": [] - }, - { - "given": "R. K. C.", - "family": "Yuen", - "affiliation": [] - }, - { - "given": "Y.", - "family": "Hua", - "affiliation": [] - }, - { - "given": "S.", - "family": "Gueroussov", - "affiliation": [] - }, - { - "given": "H. S.", - "family": "Najafabadi", - "affiliation": [] - }, - { - "given": "T. R.", - "family": "Hughes", - "affiliation": [] - }, - { - "given": "Q.", - "family": "Morris", - "affiliation": [] - }, - { - "given": "Y.", - "family": "Barash", - "affiliation": [] - }, - { - "given": "A. R.", - "family": "Krainer", - "affiliation": [] - }, - { - "given": "N.", - "family": "Jojic", - "affiliation": [] - }, - { - "given": "S. W.", - "family": "Scherer", - "affiliation": [] - }, - { - "given": "B. J.", - "family": "Blencowe", - "affiliation": [] - }, - { - "given": "B. J.", - "family": "Frey", - "affiliation": [] - } - ], - "member": "221", - "published-online": { - "date-parts": [ - [ - 2014, - 12, - 18 - ] - ] - }, - "container-title": "Science", - "original-title": [], - "link": [ - { - "URL": "https://syndication.highwire.org/content/doi/10.1126/science.1254806", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 23 - ] - ], - "date-time": "2017-06-23T02:20:36Z", - "timestamp": 1498184436000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 12, - 18 - ] - ] - }, - "references-count": 78, - "URL": "https://doi.org/10.1126/science.1254806", - "relation": { - "cites": [] - }, - "subject": [ - "Multidisciplinary" - ], - "container-title-short": "Science", - "id": "17sgPdcMT" - }, - "citation_id": "17sgPdcMT" - }, - "arxiv:1502.03044": { - "source": "arxiv", - "identifer": "1502.03044", - "standard_citation": "arxiv:1502.03044", - "bibtex": "@article{yHn4SDRI,\n abstract = {Inspired by recent work in machine translation and object detection, we\nintroduce an attention based model that automatically learns to describe the\ncontent of images. We describe how we can train this model in a deterministic\nmanner using standard backpropagation techniques and stochastically by\nmaximizing a variational lower bound. We also show through visualization how\nthe model is able to automatically learn to fix its gaze on salient objects\nwhile generating the corresponding words in the output sequence. We validate\nthe use of attention with state-of-the-art performance on three benchmark\ndatasets: Flickr8k, Flickr30k and MS COCO.},\n archiveprefix = {arXiv},\n author = {Kelvin Xu and Jimmy Ba and Ryan Kiros and Kyunghyun Cho and Aaron Courville and Ruslan Salakhutdinov and Richard Zemel and Yoshua Bengio},\n eprint = {1502.03044v3},\n file = {1502.03044v3.pdf},\n month = {Feb},\n primaryclass = {cs.LG},\n title = {Show, Attend and Tell: Neural Image Caption Generation with Visual\nAttention},\n url = {https://arxiv.org/abs/1502.03044v3},\n year = {2015}\n}\n\n", - "citation_id": "yHn4SDRI" - }, - "doi:10.1002/minf.201600045": { - "source": "doi", - "identifer": "10.1002/minf.201600045", - "standard_citation": "doi:10.1002/minf.201600045", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 9 - ] - ], - "date-time": "2017-08-09T17:27:32Z", - "timestamp": 1502299652265 - }, - "reference-count": 30, - "publisher": "Wiley-Blackwell", - "issue": "1-2", - "license": [ - { - "URL": "http://doi.wiley.com/10.1002/tdm_license_1", - "start": { - "date-parts": [ - [ - 2016, - 8, - 12 - ] - ], - "date-time": "2016-08-12T00:00:00Z", - "timestamp": 1470960000000 - }, - "delay-in-days": 0, - "content-version": "tdm" - }, - { - "URL": "http://onlinelibrary.wiley.com/termsAndConditions", - "start": { - "date-parts": [ - [ - 2017, - 8, - 12 - ] - ], - "date-time": "2017-08-12T00:00:00Z", - "timestamp": 1502496000000 - }, - "delay-in-days": 365, - "content-version": "am" - }, - { - "URL": "http://onlinelibrary.wiley.com/termsAndConditions", - "start": { - "date-parts": [ - [ - 2016, - 8, - 12 - ] - ], - "date-time": "2016-08-12T00:00:00Z", - "timestamp": 1470960000000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/501100003382", - "name": "Core Research for Evolutional Science and Technology", - "doi-asserted-by": "crossref", - "award": [] - }, - { - "DOI": "10.13039/501100002241", - "name": "Japan Science and Technology Agency", - "doi-asserted-by": "publisher", - "award": [] - }, - { - "name": "Innovative Drug Discovery Infrastructure through Functional Control of Biomolecular Systems", - "award": [] - }, - { - "name": "Priority Issue 1 in Post-K Supercomputer Development", - "award": [ - "hp150272", - "hp160213" - ] - }, - { - "name": "FOCUS Establishing Supercomputing Center of Excellence", - "award": [] - }, - { - "name": "Subsidy for Kobe Biomedical Innovation Cluster", - "award": [] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2017, - 1 - ] - ] - }, - "DOI": "10.1002/minf.201600045", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 8, - 12 - ] - ], - "date-time": "2016-08-12T11:42:43Z", - "timestamp": 1471002163000 - }, - "page": "1600045", - "source": "Crossref", - "is-referenced-by-count": 2, - "title": "CGBVS-DNN: Prediction of Compound-protein Interactions Based on Deep Learning", - "prefix": "10.1002", - "volume": "36", - "author": [ - { - "given": "Masatoshi", - "family": "Hamanaka", - "affiliation": [ - { - "name": "Graduate School of Medicine; Kyoto University; Shogoin-kawaharacho, city/>Sakyo-ku Kyoto 606-8507 Japan" - } - ] - }, - { - "given": "Kei", - "family": "Taneishi", - "affiliation": [ - { - "name": "Advanced Institute for Computational Science, RIKEN; 7-1-28, Minatojima-minami-machi Chuo-ku, Kobe, Hyogo 650-0047 Japan" - } - ] - }, - { - "given": "Hiroaki", - "family": "Iwata", - "affiliation": [ - { - "name": "Foundation for Biomedical Research and Innovation; 1-6-5, Minatojima-Minamimachi Chuo-ku Kobe 650-0047 Japan" - } - ] - }, - { - "given": "Jun", - "family": "Ye", - "affiliation": [ - { - "name": "Software and Services Group, Intel Corporation; Shanghai China" - } - ] - }, - { - "given": "Jianguo", - "family": "Pei", - "affiliation": [ - { - "name": "Software and Services Group, Intel Corporation; Shanghai China" - } - ] - }, - { - "given": "Jinlong", - "family": "Hou", - "affiliation": [ - { - "name": "Software and Services Group, Intel Corporation; Shanghai China" - } - ] - }, - { - "given": "Yasushi", - "family": "Okuno", - "affiliation": [ - { - "name": "Graduate School of Medicine; Kyoto University; Shogoin-kawaharacho, city/>Sakyo-ku Kyoto 606-8507 Japan" - } - ] - } - ], - "member": "311", - "published-online": { - "date-parts": [ - [ - 2016, - 8, - 12 - ] - ] - }, - "container-title": "Molecular Informatics", - "original-title": [], - "link": [ - { - "URL": "https://api.wiley.com/onlinelibrary/tdm/v1/articles/10.1002%2Fminf.201600045", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "https://api.wiley.com/onlinelibrary/tdm/v1/articles/10.1002%2Fminf.201600045", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://api.wiley.com/onlinelibrary/chorus/v1/articles/10.1002%2Fminf.201600045", - "content-type": "application/pdf", - "content-version": "am", - "intended-application": "syndication" - }, - { - "URL": "http://onlinelibrary.wiley.com/wol1/doi/10.1002/minf.201600045/fullpdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 9 - ] - ], - "date-time": "2017-06-09T12:24:25Z", - "timestamp": 1497011065000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 8, - 12 - ] - ] - }, - "references-count": 30, - "URL": "https://doi.org/10.1002/minf.201600045", - "relation": { - "cites": [] - }, - "subject": [ - "Organic Chemistry", - "Molecular Medicine", - "Structural Biology", - "Drug Discovery", - "Computer Science Applications" - ], - "container-title-short": "Mol. Inf.", - "id": "x0M6vals" - }, - "citation_id": "x0M6vals" - }, - "doi:10.1007/978-3-319-47898-2_21": { - "source": "doi", - "identifer": "10.1007/978-3-319-47898-2_21", - "standard_citation": "doi:10.1007/978-3-319-47898-2_21", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T00:49:54Z", - "timestamp": 1502326194851 - }, - "publisher-location": "Cham", - "reference-count": 23, - "publisher": "Springer International Publishing", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2016, - 10, - 8 - ] - ], - "date-time": "2016-10-08T00:00:00Z", - "timestamp": 1475884800000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2017 - ] - ] - }, - "DOI": "10.1007/978-3-319-47898-2_21", - "type": "chapter", - "created": { - "date-parts": [ - [ - 2016, - 10, - 7 - ] - ], - "date-time": "2016-10-07T18:41:28Z", - "timestamp": 1475865688000 - }, - "page": "195-204", - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Multi-task Deep Neural Networks for Automated Extraction of Primary Site and Laterality Information from Cancer Pathology Reports", - "prefix": "10.1007", - "author": [ - { - "given": "Hong-Jun", - "family": "Yoon", - "affiliation": [] - }, - { - "given": "Arvind", - "family": "Ramanathan", - "affiliation": [] - }, - { - "given": "Georgia", - "family": "Tourassi", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2016, - 10, - 8 - ] - ] - }, - "container-title": "Advances in Big Data", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1007/978-3-319-47898-2_21", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 25 - ] - ], - "date-time": "2017-06-25T00:37:36Z", - "timestamp": 1498351056000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 10, - 8 - ] - ] - }, - "references-count": 23, - "URL": "https://doi.org/10.1007/978-3-319-47898-2_21", - "relation": { - "cites": [] - }, - "id": "yUgE09ve" - }, - "citation_id": "yUgE09ve" - }, - "arxiv:1506.06579": { - "source": "arxiv", - "identifer": "1506.06579", - "standard_citation": "arxiv:1506.06579", - "bibtex": "@article{17i18PMkR,\n abstract = {Recent years have produced great advances in training large, deep neural\nnetworks (DNNs), including notable successes in training convolutional neural\nnetworks (convnets) to recognize natural images. However, our understanding of\nhow these models work, especially what computations they perform at\nintermediate layers, has lagged behind. Progress in the field will be further\naccelerated by the development of better tools for visualizing and interpreting\nneural nets. We introduce two such tools here. The first is a tool that\nvisualizes the activations produced on each layer of a trained convnet as it\nprocesses an image or video (e.g. a live webcam stream). We have found that\nlooking at live activations that change in response to user input helps build\nvaluable intuitions about how convnets work. The second tool enables\nvisualizing features at each layer of a DNN via regularized optimization in\nimage space. Because previous versions of this idea produced less recognizable\nimages, here we introduce several new regularization methods that combine to\nproduce qualitatively clearer, more interpretable visualizations. Both tools\nare open source and work on a pre-trained convnet with minimal setup.},\n archiveprefix = {arXiv},\n author = {Jason Yosinski and Jeff Clune and Anh Nguyen and Thomas Fuchs and Hod Lipson},\n eprint = {1506.06579v1},\n file = {1506.06579v1.pdf},\n month = {Jun},\n primaryclass = {cs.CV},\n title = {Understanding Neural Networks Through Deep Visualization},\n url = {https://arxiv.org/abs/1506.06579v1},\n year = {2015}\n}\n\n", - "citation_id": "17i18PMkR" - }, - "url:https://papers.nips.cc/paper/5347-how-transferable-are-features-in-deep-neural-networks": { - "source": "url", - "identifer": "https://papers.nips.cc/paper/5347-how-transferable-are-features-in-deep-neural-networks", - "standard_citation": "url:https://papers.nips.cc/paper/5347-how-transferable-are-features-in-deep-neural-networks", - "citeproc": { - "URL": "https://papers.nips.cc/paper/5347-how-transferable-are-features-in-deep-neural-networks", - "title": "How transferable are features in deep neural networks?", - "issued": { - "date-parts": [ - [ - 2014 - ] - ] - }, - "author": [ - { - "family": "Yosinski", - "given": "Jason" - }, - { - "family": "Clune", - "given": "Jeff" - }, - { - "family": "Bengio", - "given": "Yoshua" - }, - { - "family": "Lipson", - "given": "Hod" - } - ], - "greycite-status": "Scanned", - "greycite-scanned": "2017-05-17 02:13:23", - "type": "webpage", - "id": "enhj7VT6" - }, - "citation_id": "enhj7VT6" - }, - "doi:10.1109/tmi.2016.2642839": { - "source": "doi", - "identifer": "10.1109/tmi.2016.2642839", - "standard_citation": "doi:10.1109/tmi.2016.2642839", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T02:40:14Z", - "timestamp": 1506998414062 - }, - "reference-count": 49, - "publisher": "Institute of Electrical and Electronics Engineers (IEEE)", - "issue": "4", - "funder": [ - { - "DOI": "10.13039/501100003453", - "name": "Natural Science Foundation of Guangdong Province", - "doi-asserted-by": "publisher", - "award": [ - "2016A030313047" - ] - }, - { - "name": "Research Grants Council of the Hong Kong Special Administrative Region", - "award": [ - "CUHK 14203115", - "CUHK 14202514" - ] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2017, - 4 - ] - ] - }, - "DOI": "10.1109/tmi.2016.2642839", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 12, - 21 - ] - ], - "date-time": "2016-12-21T21:34:13Z", - "timestamp": 1482356053000 - }, - "page": "994-1004", - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Automated Melanoma Recognition in Dermoscopy Images via Very Deep Residual Networks", - "prefix": "10.1109", - "volume": "36", - "author": [ - { - "given": "Lequan", - "family": "Yu", - "affiliation": [] - }, - { - "given": "Hao", - "family": "Chen", - "affiliation": [] - }, - { - "given": "Qi", - "family": "Dou", - "affiliation": [] - }, - { - "given": "Jing", - "family": "Qin", - "affiliation": [] - }, - { - "given": "Pheng-Ann", - "family": "Heng", - "affiliation": [] - } - ], - "member": "263", - "container-title": "IEEE Transactions on Medical Imaging", - "original-title": [], - "link": [ - { - "URL": "http://xplorestaging.ieee.org/ielx7/42/7891082/07792699.pdf?arnumber=7792699", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T02:12:22Z", - "timestamp": 1506996742000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2017, - 4 - ] - ] - }, - "references-count": 49, - "URL": "https://doi.org/10.1109/tmi.2016.2642839", - "relation": {}, - "subject": [ - "Electrical and Electronic Engineering", - "Radiological and Ultrasound Technology", - "Software", - "Computer Science Applications" - ], - "container-title-short": "IEEE Trans. Med. Imaging", - "id": "phRCihNB" - }, - "citation_id": "phRCihNB" - }, - "arxiv:1311.2901": { - "source": "arxiv", - "identifer": "1311.2901", - "standard_citation": "arxiv:1311.2901", - "bibtex": "@article{voh0OiT2,\n abstract = {Large Convolutional Network models have recently demonstrated impressive\nclassification performance on the ImageNet benchmark. However there is no clear\nunderstanding of why they perform so well, or how they might be improved. In\nthis paper we address both issues. We introduce a novel visualization technique\nthat gives insight into the function of intermediate feature layers and the\noperation of the classifier. We also perform an ablation study to discover the\nperformance contribution from different model layers. This enables us to find\nmodel architectures that outperform Krizhevsky \\etal on the ImageNet\nclassification benchmark. We show our ImageNet model generalizes well to other\ndatasets: when the softmax classifier is retrained, it convincingly beats the\ncurrent state-of-the-art results on Caltech-101 and Caltech-256 datasets.},\n archiveprefix = {arXiv},\n author = {Matthew D Zeiler and Rob Fergus},\n eprint = {1311.2901v3},\n file = {1311.2901v3.pdf},\n month = {Dec},\n primaryclass = {cs.CV},\n title = {Visualizing and Understanding Convolutional Networks},\n url = {https://arxiv.org/abs/1311.2901v3},\n year = {2013}\n}\n\n", - "citation_id": "voh0OiT2" - }, - "doi:10.1186/s12859-015-0553-9": { - "source": "doi", - "identifer": "10.1186/s12859-015-0553-9", - "standard_citation": "doi:10.1186/s12859-015-0553-9", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 11 - ] - ], - "date-time": "2017-09-11T16:42:10Z", - "timestamp": 1505148130881 - }, - "reference-count": 23, - "publisher": "Springer Nature", - "issue": "1", - "license": [ - { - "URL": "http://www.springer.com/tdm", - "start": { - "date-parts": [ - [ - 2015, - 5, - 7 - ] - ], - "date-time": "2015-05-07T00:00:00Z", - "timestamp": 1430956800000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "link.springer.com" - ], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015, - 12 - ] - ] - }, - "DOI": "10.1186/s12859-015-0553-9", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 5, - 6 - ] - ], - "date-time": "2015-05-06T07:36:22Z", - "timestamp": 1430897782000 - }, - "update-policy": "http://dx.doi.org/10.1007/springer_crossmark_policy", - "source": "Crossref", - "is-referenced-by-count": 10, - "title": "Deep convolutional neural networks for annotating gene expression patterns in the mouse brain", - "prefix": "10.1186", - "volume": "16", - "author": [ - { - "given": "Tao", - "family": "Zeng", - "affiliation": [] - }, - { - "given": "Rongjian", - "family": "Li", - "affiliation": [] - }, - { - "given": "Ravi", - "family": "Mukkamala", - "affiliation": [] - }, - { - "given": "Jieping", - "family": "Ye", - "affiliation": [] - }, - { - "given": "Shuiwang", - "family": "Ji", - "affiliation": [] - } - ], - "member": "297", - "published-online": { - "date-parts": [ - [ - 2015, - 5, - 7 - ] - ] - }, - "container-title": "BMC Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://link.springer.com/content/pdf/10.1186/s12859-015-0553-9.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://link.springer.com/article/10.1186/s12859-015-0553-9/fulltext.html", - "content-type": "text/html", - "content-version": "vor", - "intended-application": "text-mining" - }, - { - "URL": "http://link.springer.com/content/pdf/10.1186/s12859-015-0553-9.pdf", - "content-type": "application/pdf", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2016, - 11, - 11 - ] - ], - "date-time": "2016-11-11T12:19:24Z", - "timestamp": 1478866764000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 5, - 7 - ] - ] - }, - "references-count": 23, - "alternative-id": [ - "553" - ], - "URL": "https://doi.org/10.1186/s12859-015-0553-9", - "relation": { - "cites": [] - }, - "subject": [ - "Biochemistry", - "Applied Mathematics", - "Molecular Biology", - "Structural Biology", - "Computer Science Applications" - ], - "container-title-short": "BMC Bioinformatics", - "article-number": "147", - "id": "z3I2IudI" - }, - "citation_id": "z3I2IudI" - }, - "doi:10.1093/bioinformatics/btw255": { - "source": "doi", - "identifer": "10.1093/bioinformatics/btw255", - "standard_citation": "doi:10.1093/bioinformatics/btw255", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 25 - ] - ], - "date-time": "2017-08-25T10:41:28Z", - "timestamp": 1503657688021 - }, - "reference-count": 15, - "publisher": "Oxford University Press (OUP)", - "issue": "12", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2016, - 6, - 15 - ] - ] - }, - "DOI": "10.1093/bioinformatics/btw255", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2016, - 6, - 15 - ] - ], - "date-time": "2016-06-15T15:43:52Z", - "timestamp": 1466005432000 - }, - "page": "i121-i127", - "source": "Crossref", - "is-referenced-by-count": 0, - "title": "Convolutional neural network architectures for predicting DNA–protein binding", - "prefix": "10.1093", - "volume": "32", - "author": [ - { - "given": "Haoyang", - "family": "Zeng", - "affiliation": [] - }, - { - "given": "Matthew D.", - "family": "Edwards", - "affiliation": [] - }, - { - "given": "Ge", - "family": "Liu", - "affiliation": [] - }, - { - "given": "David K.", - "family": "Gifford", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2016, - 6, - 15 - ] - ] - }, - "container-title": "Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/bioinformatics/article-pdf/32/12/i121/", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 24 - ] - ], - "date-time": "2017-08-24T09:59:46Z", - "timestamp": 1503568786000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2016, - 6, - 15 - ] - ] - }, - "references-count": 15, - "alternative-id": [ - "10.1093/bioinformatics/btw255" - ], - "URL": "https://doi.org/10.1093/bioinformatics/btw255", - "relation": {}, - "subject": [ - "Statistics and Probability", - "Computational Theory and Mathematics", - "Biochemistry", - "Molecular Biology", - "Computational Mathematics", - "Computer Science Applications" - ], - "container-title-short": "Bioinformatics", - "id": "182UhQqzp" - }, - "citation_id": "182UhQqzp" - }, - "doi:10.1145/2783258.2783304": { - "source": "doi", - "identifer": "10.1145/2783258.2783304", - "standard_citation": "doi:10.1145/2783258.2783304", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 3 - ] - ], - "date-time": "2017-09-03T20:02:18Z", - "timestamp": 1504468938538 - }, - "publisher-location": "New York, New York, USA", - "reference-count": 28, - "publisher": "ACM Press", - "license": [ - { - "URL": "http://www.acm.org/publications/policies/copyright_policy#Background", - "start": { - "date-parts": [ - [ - 2015, - 8, - 10 - ] - ], - "date-time": "2015-08-10T00:00:00Z", - "timestamp": 1439164800000 - }, - "delay-in-days": 221, - "content-version": "vor" - } - ], - "funder": [ - { - "DOI": "10.13039/100000002", - "name": "National Institutes of Health", - "doi-asserted-by": "publisher", - "award": [ - "R01 LM010730,HG002516-09" - ] - }, - { - "DOI": "10.13039/100000001", - "name": "National Science Foundation", - "doi-asserted-by": "publisher", - "award": [ - "IIS-0953662,DBI-1147134,DBI-1350258" - ] - } - ], - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2015 - ] - ] - }, - "DOI": "10.1145/2783258.2783304", - "type": "paper-conference", - "created": { - "date-parts": [ - [ - 2015, - 8, - 7 - ] - ], - "date-time": "2015-08-07T15:38:27Z", - "timestamp": 1438961907000 - }, - "source": "Crossref", - "is-referenced-by-count": 13, - "title": "Deep Model Based Transfer and Multi-Task Learning for Biological Image Analysis", - "prefix": "10.1145", - "author": [ - { - "given": "Wenlu", - "family": "Zhang", - "affiliation": [] - }, - { - "given": "Rongjian", - "family": "Li", - "affiliation": [] - }, - { - "given": "Tao", - "family": "Zeng", - "affiliation": [] - }, - { - "given": "Qian", - "family": "Sun", - "affiliation": [] - }, - { - "given": "Sudhir", - "family": "Kumar", - "affiliation": [] - }, - { - "given": "Jieping", - "family": "Ye", - "affiliation": [] - }, - { - "given": "Shuiwang", - "family": "Ji", - "affiliation": [] - } - ], - "member": "320", - "container-title": "Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining - KDD '15", - "original-title": [], - "link": [ - { - "URL": "http://dl.acm.org/ft_gateway.cfm?id=2783304&ftid=1613097&dwn=1", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 23 - ] - ], - "date-time": "2017-06-23T17:14:15Z", - "timestamp": 1498238055000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015 - ] - ] - }, - "references-count": 28, - "URL": "https://doi.org/10.1145/2783258.2783304", - "relation": { - "cites": [] - }, - "id": "HlDY7trA" - }, - "citation_id": "HlDY7trA" - }, - "doi:10.1038/nmeth.3547": { - "source": "doi", - "identifer": "10.1038/nmeth.3547", - "standard_citation": "doi:10.1038/nmeth.3547", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 3 - ] - ], - "date-time": "2017-10-03T14:02:16Z", - "timestamp": 1507039336022 - }, - "reference-count": 26, - "publisher": "Springer Nature", - "issue": "10", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "DOI": "10.1038/nmeth.3547", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2015, - 8, - 24 - ] - ], - "date-time": "2015-08-24T15:20:38Z", - "timestamp": 1440429638000 - }, - "page": "931-934", - "source": "Crossref", - "is-referenced-by-count": 96, - "title": "Predicting effects of noncoding variants with deep learning–based sequence model", - "prefix": "10.1038", - "volume": "12", - "author": [ - { - "given": "Jian", - "family": "Zhou", - "affiliation": [] - }, - { - "given": "Olga G", - "family": "Troyanskaya", - "affiliation": [] - } - ], - "member": "339", - "published-online": { - "date-parts": [ - [ - 2015, - 8, - 24 - ] - ] - }, - "container-title": "Nature Methods", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2015, - 9, - 29 - ] - ], - "date-time": "2015-09-29T15:07:31Z", - "timestamp": 1443539251000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2015, - 8, - 24 - ] - ] - }, - "references-count": 26, - "alternative-id": [ - "nmeth.3547" - ], - "URL": "https://doi.org/10.1038/nmeth.3547", - "relation": { - "cites": [] - }, - "subject": [ - "Biotechnology", - "Cell Biology", - "Biochemistry", - "Molecular Biology" - ], - "container-title-short": "Nat Meth", - "id": "2UI1BZuD" - }, - "citation_id": "2UI1BZuD" - }, - "arxiv:1702.04595": { - "source": "arxiv", - "identifer": "1702.04595", - "standard_citation": "arxiv:1702.04595", - "bibtex": "@article{Kk20paR7,\n abstract = {This article presents the prediction difference analysis method for\nvisualizing the response of a deep neural network to a specific input. When\nclassifying images, the method highlights areas in a given input image that\nprovide evidence for or against a certain class. It overcomes several\nshortcoming of previous methods and provides great additional insight into the\ndecision making process of classifiers. Making neural network decisions\ninterpretable through visualization is important both to improve models and to\naccelerate the adoption of black-box classifiers in application areas such as\nmedicine. We illustrate the method in experiments on natural images (ImageNet\ndata), as well as medical images (MRI brain scans).},\n archiveprefix = {arXiv},\n author = {Luisa M Zintgraf and Taco S Cohen and Tameem Adel and Max Welling},\n eprint = {1702.04595v1},\n file = {1702.04595v1.pdf},\n month = {Feb},\n primaryclass = {cs.CV},\n title = {Visualizing Deep Neural Network Decisions: Prediction Difference\nAnalysis},\n url = {https://arxiv.org/abs/1702.04595v1},\n year = {2017}\n}\n\n", - "citation_id": "Kk20paR7" - }, - "doi:10.1093/bioinformatics/btt389": { - "source": "doi", - "identifer": "10.1093/bioinformatics/btt389", - "standard_citation": "doi:10.1093/bioinformatics/btt389", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T07:02:12Z", - "timestamp": 1507100532877 - }, - "reference-count": 25, - "publisher": "Oxford University Press (OUP)", - "issue": "18", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2013, - 9, - 15 - ] - ] - }, - "DOI": "10.1093/bioinformatics/btt389", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2013, - 7, - 5 - ] - ], - "date-time": "2013-07-05T01:56:51Z", - "timestamp": 1372989411000 - }, - "page": "2253-2260", - "source": "Crossref", - "is-referenced-by-count": 43, - "title": "Scalable metagenomic taxonomy classification using a reference genome database", - "prefix": "10.1093", - "volume": "29", - "author": [ - { - "given": "Sasha K.", - "family": "Ames", - "affiliation": [] - }, - { - "given": "David A.", - "family": "Hysom", - "affiliation": [] - }, - { - "given": "Shea N.", - "family": "Gardner", - "affiliation": [] - }, - { - "given": "G. Scott", - "family": "Lloyd", - "affiliation": [] - }, - { - "given": "Maya B.", - "family": "Gokhale", - "affiliation": [] - }, - { - "given": "Jonathan E.", - "family": "Allen", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2013, - 7, - 4 - ] - ] - }, - "container-title": "Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/bioinformatics/article-pdf/29/18/2253/17128159/btt389.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 22 - ] - ], - "date-time": "2017-08-22T14:54:06Z", - "timestamp": 1503413646000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2013, - 7, - 4 - ] - ] - }, - "references-count": 25, - "URL": "https://doi.org/10.1093/bioinformatics/btt389", - "relation": {}, - "subject": [ - "Statistics and Probability", - "Computational Theory and Mathematics", - "Biochemistry", - "Molecular Biology", - "Computational Mathematics", - "Computer Science Applications" - ], - "id": "189TQrQA9" - }, - "citation_id": "189TQrQA9" - }, - "doi:10.1093/bioinformatics/btq619": { - "source": "doi", - "identifer": "10.1093/bioinformatics/btq619", - "standard_citation": "doi:10.1093/bioinformatics/btq619", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 9, - 21 - ] - ], - "date-time": "2017-09-21T14:02:11Z", - "timestamp": 1506002531436 - }, - "reference-count": 10, - "publisher": "Oxford University Press (OUP)", - "issue": "1", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2011, - 1, - 1 - ] - ] - }, - "DOI": "10.1093/bioinformatics/btq619", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2010, - 11, - 10 - ] - ], - "date-time": "2010-11-10T01:25:36Z", - "timestamp": 1289352336000 - }, - "page": "127-129", - "source": "Crossref", - "is-referenced-by-count": 85, - "title": "NBC: the Naive Bayes Classification tool webserver for taxonomic classification of metagenomic reads", - "prefix": "10.1093", - "volume": "27", - "author": [ - { - "given": "G. L.", - "family": "Rosen", - "affiliation": [] - }, - { - "given": "E. R.", - "family": "Reichenberger", - "affiliation": [] - }, - { - "given": "A. M.", - "family": "Rosenfeld", - "affiliation": [] - } - ], - "member": "286", - "published-online": { - "date-parts": [ - [ - 2010, - 11, - 8 - ] - ] - }, - "container-title": "Bioinformatics", - "original-title": [], - "link": [ - { - "URL": "http://academic.oup.com/bioinformatics/article-pdf/27/1/127/520591/btq619.pdf", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 8, - 22 - ] - ], - "date-time": "2017-08-22T15:05:26Z", - "timestamp": 1503414326000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2010, - 11, - 8 - ] - ] - }, - "references-count": 10, - "URL": "https://doi.org/10.1093/bioinformatics/btq619", - "relation": {}, - "subject": [ - "Statistics and Probability", - "Computational Theory and Mathematics", - "Biochemistry", - "Molecular Biology", - "Computational Mathematics", - "Computer Science Applications" - ], - "container-title-short": "Bioinformatics", - "id": "1HtJuEkb2" - }, - "citation_id": "1HtJuEkb2" - }, - "doi:10.1371/journal.pone.0091784": { - "source": "doi", - "identifer": "10.1371/journal.pone.0091784", - "standard_citation": "doi:10.1371/journal.pone.0091784", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 10, - 4 - ] - ], - "date-time": "2017-10-04T11:22:17Z", - "timestamp": 1507116137281 - }, - "reference-count": 19, - "publisher": "Public Library of Science (PLoS)", - "issue": "3", - "license": [ - { - "URL": "http://creativecommons.org/licenses/by/4.0/", - "start": { - "date-parts": [ - [ - 2014, - 3, - 13 - ] - ], - "date-time": "2014-03-13T00:00:00Z", - "timestamp": 1394668800000 - }, - "delay-in-days": 0, - "content-version": "vor" - } - ], - "content-domain": { - "domain": [ - "www.plosone.org" - ], - "crossmark-restriction": false - }, - "DOI": "10.1371/journal.pone.0091784", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2014, - 3, - 13 - ] - ], - "date-time": "2014-03-13T21:12:48Z", - "timestamp": 1394745168000 - }, - "page": "e91784", - "update-policy": "http://dx.doi.org/10.1371/journal.pone.corrections_policy", - "source": "Crossref", - "is-referenced-by-count": 14, - "title": "WGSQuikr: Fast Whole-Genome Shotgun Metagenomic Classification", - "prefix": "10.1371", - "volume": "9", - "author": [ - { - "given": "David", - "family": "Koslicki", - "affiliation": [] - }, - { - "given": "Simon", - "family": "Foucart", - "affiliation": [] - }, - { - "given": "Gail", - "family": "Rosen", - "affiliation": [] - } - ], - "member": "340", - "published-online": { - "date-parts": [ - [ - 2014, - 3, - 13 - ] - ] - }, - "container-title": "PLoS ONE", - "original-title": [], - "link": [ - { - "URL": "http://dx.plos.org/10.1371/journal.pone.0091784", - "content-type": "unspecified", - "content-version": "vor", - "intended-application": "similarity-checking" - } - ], - "deposited": { - "date-parts": [ - [ - 2017, - 6, - 22 - ] - ], - "date-time": "2017-06-22T09:34:26Z", - "timestamp": 1498124066000 - }, - "score": 1.0, - "subtitle": [], - "editor": [ - { - "given": "Mark R.", - "family": "Liles", - "affiliation": [] - } - ], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2014, - 3, - 13 - ] - ] - }, - "references-count": 19, - "URL": "https://doi.org/10.1371/journal.pone.0091784", - "relation": { - "cites": [] - }, - "subject": [ - "General Biochemistry, Genetics and Molecular Biology", - "General Agricultural and Biological Sciences", - "General Medicine" - ], - "container-title-short": "PLoS ONE", - "id": "RqhGD9c7" - }, - "citation_id": "RqhGD9c7" - }, - "doi:10.1186/1471-2105-12-20": { - "source": "doi", - "identifer": "10.1186/1471-2105-12-20", - "standard_citation": "doi:10.1186/1471-2105-12-20", - "citeproc": { - "indexed": { - "date-parts": [ - [ - 2017, - 8, - 10 - ] - ], - "date-time": "2017-08-10T17:19:37Z", - "timestamp": 1502385577714 - }, - "reference-count": 0, - "publisher": "Springer Nature", - "issue": "1", - "content-domain": { - "domain": [], - "crossmark-restriction": false - }, - "published-print": { - "date-parts": [ - [ - 2011 - ] - ] - }, - "DOI": "10.1186/1471-2105-12-20", - "type": "article-journal", - "created": { - "date-parts": [ - [ - 2011, - 1, - 19 - ] - ], - "date-time": "2011-01-19T11:10:25Z", - "timestamp": 1295435425000 - }, - "page": "20", - "source": "Crossref", - "is-referenced-by-count": 26, - "title": "Combining gene prediction methods to improve metagenomic gene annotation", - "prefix": "10.1186", - "volume": "12", - "author": [ - { - "given": "Non G", - "family": "Yok", - "affiliation": [] - }, - { - "given": "Gail L", - "family": "Rosen", - "affiliation": [] - } - ], - "member": "297", - "container-title": "BMC Bioinformatics", - "original-title": [], - "deposited": { - "date-parts": [ - [ - 2016, - 5, - 16 - ] - ], - "date-time": "2016-05-16T12:51:59Z", - "timestamp": 1463403119000 - }, - "score": 1.0, - "subtitle": [], - "short-title": [], - "issued": { - "date-parts": [ - [ - 2011 - ] - ] - }, - "references-count": 0, - "alternative-id": [ - "1471-2105-12-20" - ], - "URL": "https://doi.org/10.1186/1471-2105-12-20", - "relation": {}, - "subject": [ - "Biochemistry", - "Applied Mathematics", - "Molecular Biology", - "Structural Biology", - "Computer Science Applications" - ], - "container-title-short": "BMC Bioinformatics", - "id": "qUGH5CX8" - }, - "citation_id": "qUGH5CX8" - }, - "url:http://ana.blogs.com/maestros/2006/11/data_is_the_new.html": { - "source": "url", - "identifer": "http://ana.blogs.com/maestros/2006/11/data_is_the_new.html", - "standard_citation": "url:http://ana.blogs.com/maestros/2006/11/data_is_the_new.html", - "citeproc": { - "URL": "http://ana.blogs.com/maestros/2006/11/data_is_the_new.html", - "title": "Data is the New Oil", - "container-title": "ANA Marketing Maestros", - "issued": { - "date-parts": [ - [ - 2006, - 11 - ] - ] - }, - "author": [ - { - "family": "Palmer", - "given": "Michael" - } - ], - "archives": [ - "http://wayback.archive.org/web/http://ana.blogs.com/maestros/2006/11/data_is_the_new.html" - ], - "type": "webpage", - "id": "6fE0Vrba" - }, - "citation_id": "6fE0Vrba" - }, - "url:http://blog.dhimmel.com/irreproducible-timestamps/": { - "source": "url", - "identifer": "http://blog.dhimmel.com/irreproducible-timestamps/", - "standard_citation": "url:http://blog.dhimmel.com/irreproducible-timestamps/", - "citeproc": { - "URL": "http://blog.dhimmel.com/irreproducible-timestamps/", - "title": "The most interesting case of scientific irreproducibility?", - "container-title": "Satoshi Village", - "issued": { - "date-parts": [ - [ - 2017, - 3, - 8 - ] - ] - }, - "author": [ - { - "family": "Himmelstein", - "given": "Daniel" - } - ], - "greycite-status": "Scanned", - "greycite-scanned": "2017-05-16 19:23:57", - "archives": [ - "http://wayback.archive.org/web/http://blog.dhimmel.com/irreproducible-timestamps/" - ], - "type": "webpage", - "id": "QBWMEuxW" - }, - "citation_id": "QBWMEuxW" - }, - "url:http://hazyresearch.github.io/snorkel/blog/weak_supervision.html": { - "source": "url", - "identifer": "http://hazyresearch.github.io/snorkel/blog/weak_supervision.html", - "standard_citation": "url:http://hazyresearch.github.io/snorkel/blog/weak_supervision.html", - "citeproc": { - "URL": "http://hazyresearch.github.io/snorkel/blog/weak_supervision.html", - "title": "Data Programming: Machine Learning with Weak Supervision", - "issued": { - "date-parts": [ - [ - 2016, - 9, - 19 - ] - ] - }, - "author": [ - { - "given": "Alex", - "family": "Ratner" - }, - { - "given": "Stephen", - "family": "Bach" - }, - { - "given": "Chris", - "family": "Ré" - } - ], - "type": "webpage", - "id": "hfcf5Hmi" - }, - "citation_id": "hfcf5Hmi" - }, - "url:http://proceedings.mlr.press/v54/mcmahan17a.html": { - "source": "url", - "identifer": "http://proceedings.mlr.press/v54/mcmahan17a.html", - "standard_citation": "url:http://proceedings.mlr.press/v54/mcmahan17a.html", - "citeproc": { - "URL": "http://proceedings.mlr.press/v54/mcmahan17a.html", - "title": "Communication-Efficient Learning of Deep Networks from Decentralized Data", - "issued": { - "date-parts": [ - [ - 2017, - 4, - 10 - ] - ] - }, - "author": [ - { - "family": "McMahan", - "given": "Brendan" - }, - { - "family": "Moore", - "given": "Eider" - }, - { - "family": "Ramage", - "given": "Daniel" - }, - { - "family": "Hampson", - "given": "Seth" - }, - { - "family": "Aguera y Arcas", - "given": "Blaise" - } - ], - "greycite-status": "Scanned", - "greycite-scanned": "2017-05-17 02:13:30", - "type": "webpage", - "id": "U0ySdznJ" - }, - "citation_id": "U0ySdznJ" - }, - "url:http://research.google.com/archive/unsupervised_icml2012.html": { - "source": "url", - "identifer": "http://research.google.com/archive/unsupervised_icml2012.html", - "standard_citation": "url:http://research.google.com/archive/unsupervised_icml2012.html", - "citeproc": { - "URL": "http://research.google.com/archive/unsupervised_icml2012.html", - "title": "Google Research Publication: Building High-level Features Using Large Scale Unsupervised Learning", - "issued": { - "date-parts": [ - [ - 2016, - 12, - 15 - ] - ] - }, - "greycite-status": "Scanned", - "greycite-scanned": "2017-05-17 02:13:33", - "redirects-to": "https://research.google.com/archive/unsupervised_icml2012.html", - "type": "webpage", - "id": "IiNJE32f" - }, - "citation_id": "IiNJE32f" - }, - "url:http://www.cameo3d.org/": { - "source": "url", - "identifer": "http://www.cameo3d.org/", - "standard_citation": "url:http://www.cameo3d.org/", - "citeproc": { - "URL": "http://www.cameo3d.org/", - "title": "CAMEO - Continuous Automated Model Evaluation", - "issued": { - "date-parts": [ - [ - 2017 - ] - ] - }, - "redirects-to": "https://www.cameo3d.org/", - "type": "webpage", - "id": "u9uApoaB" - }, - "citation_id": "u9uApoaB" - }, - "url:http://www.deeplearningbook.org/": { - "source": "url", - "identifer": "http://www.deeplearningbook.org/", - "standard_citation": "url:http://www.deeplearningbook.org/", - "citeproc": { - "URL": "http://www.deeplearningbook.org/", - "title": "Deep Learning", - "issued": { - "date-parts": [ - [ - 2016 - ] - ] - }, - "author": [ - { - "given": "Ian", - "family": "Goodfellow" - }, - { - "given": "Yoshua", - "family": "Bengio" - }, - { - "given": "Aaron", - "family": "Courville" - } - ], - "type": "webpage", - "id": "yg8NW0K7" - }, - "citation_id": "yg8NW0K7" - }, - "url:http://www.intel.com/pressroom/archive/speeches/ag080998.htm": { - "source": "url", - "identifer": "http://www.intel.com/pressroom/archive/speeches/ag080998.htm", - "standard_citation": "url:http://www.intel.com/pressroom/archive/speeches/ag080998.htm", - "citeproc": { - "URL": "http://www.intel.com/pressroom/archive/speeches/ag080998.htm", - "title": "Academy of Management: Andrew S. Grove", - "issued": { - "date-parts": [ - [ - 1998, - 8, - 9 - ] - ] - }, - "author": [ - { - "family": "Grove", - "given": "Andrew S." - } - ], - "type": "webpage", - "id": "mAXsmd43" - }, - "citation_id": "mAXsmd43" - }, - "url:http://www.predictioncenter.org/casp12/rrc_avrg_results.cgi": { - "source": "url", - "identifer": "http://www.predictioncenter.org/casp12/rrc_avrg_results.cgi", - "standard_citation": "url:http://www.predictioncenter.org/casp12/rrc_avrg_results.cgi", - "citeproc": { - "URL": "http://www.predictioncenter.org/casp12/rrc_avrg_results.cgi", - "title": "RR Results - CASP12", - "issued": { - "date-parts": [ - [ - 2017 - ] - ] - }, - "type": "webpage", - "id": "zScWGveU" - }, - "citation_id": "zScWGveU" - }, - "url:https://eprint.iacr.org/2017/281.pdf": { - "source": "url", - "identifer": "https://eprint.iacr.org/2017/281.pdf", - "standard_citation": "url:https://eprint.iacr.org/2017/281.pdf", - "citeproc": { - "type": "report", - "title": "Practical Secure Aggregation for Privacy Preserving Machine Learning", - "genre": "Cryptology ePrint Archive", - "source": "ePrint IACR", - "abstract": "We design a novel, communication-efficient, failure-robust protocol for secure aggregation of high-dimensional data. Our protocol allows a server to compute the sum of large, user-held data vectors from mobile devices in a secure manner (i.e. without learning each user's individual contribution), and can be used, for example, in a federated learning setting, to aggregate user-provided model updates for a deep neural network. We prove the security of our protocol in the honest-but-curious and malicious settings, and show that security is maintained even if an arbitrarily chosen subset of users drop out at any time. We evaluate the efficiency of our protocol and show, by complexity analysis and a concrete implementation, that its runtime and communication overhead remain low even on large data sets and client pools. For 16-bit input values, our protocol offerscommunication expansion forusers and-dimensional vectors, andexpansion for\n\nusers and-dimensional vectors over sending data in the clear.", - "URL": "https://eprint.iacr.org/2017/281", - "number": "281", - "author": [ - { - "family": "Bonawitz", - "given": "Keith" - }, - { - "family": "Ivanov", - "given": "Vladimir" - }, - { - "family": "Kreuter", - "given": "Ben" - }, - { - "family": "Marcedone", - "given": "Antonio" - }, - { - "family": "McMahan", - "given": "H. Brendan" - }, - { - "family": "Patel", - "given": "Sarvar" - }, - { - "family": "Ramage", - "given": "Daniel" - }, - { - "family": "Segal", - "given": "Aaron" - }, - { - "family": "Seth", - "given": "Karn" - } - ], - "issued": { - "date-parts": [ - [ - "2017" - ] - ] - }, - "accessed": { - "date-parts": [ - [ - "2017", - 5, - 23 - ] - ] - }, - "id": "1GprsH3DV" - }, - "citation_id": "1GprsH3DV" - }, - "url:https://github.com/greenelab/deep-review": { - "source": "url", - "identifer": "https://github.com/greenelab/deep-review", - "standard_citation": "url:https://github.com/greenelab/deep-review", - "citeproc": { - "URL": "https://github.com/greenelab/deep-review", - "title": "greenelab/deep-review", - "container-title": "GitHub", - "issued": { - "date-parts": [ - [ - 2017 - ] - ] - }, - "type": "webpage", - "id": "yLr4pV4G" - }, - "citation_id": "yLr4pV4G" - }, - "url:https://medium.com/twenty-one-hundred/data-is-the-new-oil-a-ludicrous-proposition-1d91bba4f294": { - "source": "url", - "identifer": "https://medium.com/twenty-one-hundred/data-is-the-new-oil-a-ludicrous-proposition-1d91bba4f294", - "standard_citation": "url:https://medium.com/twenty-one-hundred/data-is-the-new-oil-a-ludicrous-proposition-1d91bba4f294", - "citeproc": { - "URL": "https://medium.com/twenty-one-hundred/data-is-the-new-oil-a-ludicrous-proposition-1d91bba4f294", - "title": "“Data is the New Oil” — A Ludicrous Proposition – Twenty One Hundred – Medium", - "container-title": "Medium", - "issued": { - "date-parts": [ - [ - 2016, - 5, - 2 - ] - ] - }, - "author": [ - { - "family": "Haupt", - "given": "Michael" - } - ], - "greycite-status": "Scanned", - "greycite-scanned": "2017-05-17 02:14:56", - "archives": [ - "http://wayback.archive.org/web/https://medium.com/twenty-one-hundred/data-is-the-new-oil-a-ludicrous-proposition-1d91bba4f294" - ], - "type": "webpage", - "id": "o8mib4CN" - }, - "citation_id": "o8mib4CN" - }, - "url:https://nihcc.app.box.com/v/ChestXray-NIHCC": { - "source": "url", - "identifer": "https://nihcc.app.box.com/v/ChestXray-NIHCC", - "standard_citation": "url:https://nihcc.app.box.com/v/ChestXray-NIHCC", - "citeproc": { - "URL": "https://nihcc.app.box.com/v/ChestXray-NIHCC", - "title": "NIH Chest X-ray Dataset", - "issued": { - "date-parts": [ - [ - 2017, - 9, - 7 - ] - ] - }, - "author": [ - { - "family": "NIH Clinical Center", - "given": "" - } - ], - "type": "webpage", - "id": "odFR7ptt" - }, - "citation_id": "odFR7ptt" - }, - "url:https://opentimestamps.org/": { - "source": "url", - "identifer": "https://opentimestamps.org/", - "standard_citation": "url:https://opentimestamps.org/", - "citeproc": { - "URL": "https://opentimestamps.org/", - "title": "OpenTimestamps: a timestamping proof standard", - "issued": { - "date-parts": [ - [ - 2017, - 5, - 16 - ] - ] - }, - "greycite-status": "Scanned", - "greycite-scanned": "2017-05-19 17:01:28", - "type": "webpage", - "id": "igA8jklc" - }, - "citation_id": "igA8jklc" - }, - "url:https://phekb.org/implementations": { - "source": "url", - "identifer": "https://phekb.org/implementations", - "standard_citation": "url:https://phekb.org/implementations", - "citeproc": { - "URL": "https://phekb.org/implementations", - "title": "Implementations by Phenotype | PheKB", - "issued": { - "date-parts": [ - [ - 2017, - 5, - 17 - ] - ] - }, - "greycite-status": "Scanned", - "greycite-scanned": "2017-05-17 02:05:33", - "type": "webpage", - "id": "ziudr6hx" - }, - "citation_id": "ziudr6hx" - }, - "url:https://www.bgcarlisle.com/blog/2014/08/25/proof-of-prespecified-endpoints-in-medical-research-with-the-bitcoin-blockchain/": { - "source": "url", - "identifer": "https://www.bgcarlisle.com/blog/2014/08/25/proof-of-prespecified-endpoints-in-medical-research-with-the-bitcoin-blockchain/", - "standard_citation": "url:https://www.bgcarlisle.com/blog/2014/08/25/proof-of-prespecified-endpoints-in-medical-research-with-the-bitcoin-blockchain/", - "citeproc": { - "URL": "https://www.bgcarlisle.com/blog/2014/08/25/proof-of-prespecified-endpoints-in-medical-research-with-the-bitcoin-blockchain/", - "title": "The Grey Literature — Proof of prespecified endpoints in medical research with the bitcoin blockchain", - "issued": { - "date-parts": [ - [ - 2014, - 8, - 25 - ] - ] - }, - "author": [ - { - "family": "Carlisle", - "given": "Benjamin Gregory" - } - ], - "greycite-status": "Scanned", - "greycite-scanned": "2017-05-18 17:00:44", - "type": "webpage", - "id": "6MR50hyY" - }, - "citation_id": "6MR50hyY" - } -} \ No newline at end of file diff --git a/processed-citations.tsv b/citations.tsv similarity index 98% rename from processed-citations.tsv rename to citations.tsv index 5586f8a5..e6a30d10 100644 --- a/processed-citations.tsv +++ b/citations.tsv @@ -1,108 +1,230 @@ -text citation standard_citation citation_id +string citation standard_citation citation_id @arxiv:1106.5730 arxiv:1106.5730 arxiv:1106.5730 3qm8sXnB @arxiv:1206.7051 arxiv:1206.7051 arxiv:1206.7051 8RAYEOPl @arxiv:1212.0901v2 arxiv:1212.0901v2 arxiv:1212.0901v2 g2vvbB91 +@tag:Word2Vec arxiv:1301.3781 arxiv:1301.3781 1GhHIDxuW @arxiv:1308.0850 arxiv:1308.0850 arxiv:1308.0850 15y7iq6HF +@tag:Zeiler2013_visualizing arxiv:1311.2901 arxiv:1311.2901 voh0OiT2 +@tag:Simonyan2013_deep arxiv:1312.6034 arxiv:1312.6034 1YcKYTvO +@tag:Caruana2014_need arxiv:1312.6184 arxiv:1312.6184 1AhGoHZP9 @arxiv:1312.6199 arxiv:1312.6199 arxiv:1312.6199 1Fel6Bdb8 @arxiv:1403.1347 arxiv:1403.1347 arxiv:1403.1347 8t43CQ9m +@tag:Krizhevsky2014_weird_trick arxiv:1404.5997 arxiv:1404.5997 ZSVsnPVO +@tag:Dahl2014_multi_qsar arxiv:1406.1231 arxiv:1406.1231 1Dzz0P0qr +@tag:Bahdanu2014_neural arxiv:1409.0473 arxiv:1409.0473 haHzVaaz +@tag:Sutskever arxiv:1409.3215 arxiv:1409.3215 2cMhMv5A +@tag:CudNN arxiv:1410.0759 arxiv:1410.0759 YwdqeYZi @arxiv:1411.2581v1 arxiv:1411.2581v1 arxiv:1411.2581v1 pxdeuhMS +@tag:Mahendran2014_understanding arxiv:1412.0035 arxiv:1412.0035 19mGl6pfy +@tag:Nguyen2014_adversarial arxiv:1412.1897v4 arxiv:1412.1897v4 1AkF8Wsv7 @arxiv:1412.6572 arxiv:1412.6572 arxiv:1412.6572 UtcyntjF +@tag:Springenberg2014_striving arxiv:1412.6806 arxiv:1412.6806 f2L6isRj +@tag:Bengio2015_prec arxiv:1412.7024 arxiv:1412.7024 1G3owNNps +@tag:Ramsundar2015_multitask_drug arxiv:1502.02072 arxiv:1502.02072 yAoN5gTU +@tag:Gupta2015_prec arxiv:1502.02551 arxiv:1502.02551 CKcJuj03 +@tag:Xu2015_show arxiv:1502.03044 arxiv:1502.03044 yHn4SDRI +@tag:Sonderby arxiv:1503.01919 arxiv:1503.01919 81Cl5QSM +@tag:Hinton2015_dark_knowledge arxiv:1503.02531 arxiv:1503.02531 1CRF3gAV +@tag:Hadjas2015_cc arxiv:1504.04343 arxiv:1504.04343 13KjSCKB2 +@tag:Chen2015_hashing arxiv:1504.04788 arxiv:1504.04788 15lYGmZpY +@tag:Meng2016_mllib arxiv:1505.06807 arxiv:1505.06807 rZnxDitd +@tag:Karpathy2015_visualizing arxiv:1506.02078 arxiv:1506.02078 2cpYveR4 +@tag:Gal2015_dropout arxiv:1506.02142 arxiv:1506.02142 1FDihfnM +@tag:Sa2015_buckwild arxiv:1506.06438 arxiv:1506.06438 w6CoVmFK +@tag:Yosinksi2015_understanding arxiv:1506.06579 arxiv:1506.06579 17i18PMkR +@tag:Su2015_gpu arxiv:1507.01239 arxiv:1507.01239 aClNvbyM @arxiv:1510.02855 arxiv:1510.02855 arxiv:1510.02855 Z7fd0BYf +@tag:Wallach2015_atom_net arxiv:1510.02855 arxiv:1510.02855 Z7fd0BYf +@tag:Lipton2015_lstm arxiv:1510.07641 arxiv:1510.07641 glyI7H6F @arxiv:1511.02386 arxiv:1511.02386 arxiv:1511.02386 15lbUf0as +@tag:Moritz2015_sparknet arxiv:1511.06051 arxiv:1511.06051 rmJZ2Aui +@tag:He2015_images arxiv:1512.03385 arxiv:1512.03385 j7KrVyi8 +@tag:Che2015_distill arxiv:1512.03542 arxiv:1512.03542 14DAmZTDg @arxiv:1602.00357 arxiv:1602.00357 arxiv:1602.00357 HRXii6Ni +@tag:Lacey2016_dl_fpga arxiv:1602.04283 arxiv:1602.04283 9NKsJjSw +@tag:Ribeiro2016_lime arxiv:1602.04938 arxiv:1602.04938 QwXSJhr0 +@tag:TensorFlow arxiv:1603.04467 arxiv:1603.04467 hOeUlCvS +@tag:Lee2016_deeptarget arxiv:1603.09123v2 arxiv:1603.09123v2 1GwC1ll6h +@tag:Boza arxiv:1603.09195 arxiv:1603.09195 1BTJ1KqRa +@tag:Liu2016_towards arxiv:1604.07043 arxiv:1604.07043 AEc66xxR +@tag:Park2016_deepmirgene arxiv:1605.00017 arxiv:1605.00017 1TeyWffV @arxiv:1605.03661 arxiv:1605.03661 arxiv:1605.03661 173ftiSzF +@tag:Deming2016_genetic arxiv:1605.07156 arxiv:1605.07156 SAvEOARL @arxiv:1605.07723 arxiv:1605.07723 arxiv:1605.07723 5Il3kN32 +@tag:Sun2016_ensemble arxiv:1606.00575 arxiv:1606.00575 JUF9VoRD @arxiv:1606.00931 arxiv:1606.00931 arxiv:1606.00931 1FE0F2pQ +@tag:Katzman2016_deepsurv arxiv:1606.00931 arxiv:1606.00931 1FE0F2pQ +@tag:Che2016_rnn arxiv:1606.01865 arxiv:1606.01865 O7Vbecm2 +@tag:Lipton2016_missing arxiv:1606.04130 arxiv:1606.04130 4zpZxjHR +@tag:Lei2016_rationalizing arxiv:1606.04155 arxiv:1606.04155 ZUCVI5eU @arxiv:1606.05718 arxiv:1606.05718 arxiv:1606.05718 mbEp6jNr +@tag:Wang2016_breast_cancer arxiv:1606.05718 arxiv:1606.05718 mbEp6jNr +@tag:Strobelt2016_visual arxiv:1606.07461 arxiv:1606.07461 1Ad3UOefc +@tag:Kearnes2016_admet arxiv:1606.08793 arxiv:1606.08793 uP7SgBVd @arxiv:1606.08813v3 arxiv:1606.08813v3 arxiv:1606.08813v3 7yE9K08a @arxiv:1607.00133 arxiv:1607.00133 arxiv:1607.00133 ucHUOABT +@tag:Singh2016_deepchrome arxiv:1607.02078 arxiv:1607.02078 G10wkFHt @arxiv:1607.07519 arxiv:1607.07519 arxiv:1607.07519 Ohd1Q9Xw @arxiv:1608.00647 arxiv:1608.00647 arxiv:1608.00647 c6MfDdWP @arxiv:1608.02158 arxiv:1608.02158 arxiv:1608.02158 qXdO2aMm +@tag:Ranganath2016_deep arxiv:1608.02158 arxiv:1608.02158 qXdO2aMm +@tag:Lanchantin2016_motif arxiv:1608.03644 arxiv:1608.03644 Dwi2eAvT +@tag:Choi2016_retain arxiv:1608.05745 arxiv:1608.05745 UcRbawKo +@tag:Jafari2016_skin_lesions arxiv:1609.02374 arxiv:1609.02374 1AJUcl1KV @arxiv:1609.02943 arxiv:1609.02943 arxiv:1609.02943 ULSPV0rh +@tag:Hubara2016_qnn arxiv:1609.07061 arxiv:1609.07061 1GUizyE8e @arxiv:1609.08144 arxiv:1609.08144 arxiv:1609.08144 4TK06zOf +@tag:Chollet2016_xception arxiv:1610.02357 arxiv:1610.02357 VMkPJjVk +@tag:Selvaraju2016_grad arxiv:1610.02391 arxiv:1610.02391 RZsNSRDS @arxiv:1610.02413 arxiv:1610.02413 arxiv:1610.02413 1ENxzq6pT +@tag:Gomezb2016_automatic arxiv:1610.02415 arxiv:1610.02415 2dU8f4XJ +@tag:Shaham2016_batch_effects arxiv:1610.04181 arxiv:1610.04181 T2Md9xLY +@tag:Codella2016_ensemble_melanoma arxiv:1610.04662 arxiv:1610.04662 sLPsrfbl @arxiv:1610.05256 arxiv:1610.05256 arxiv:1610.05256 M2OLWojE @arxiv:1610.05820 arxiv:1610.05820 arxiv:1610.05820 1HbRTExaU @arxiv:1610.09559 arxiv:1610.09559 arxiv:1610.09559 11aqfNfQx @arxiv:1611.02796 arxiv:1611.02796 arxiv:1611.02796 lERqKdZJ @arxiv:1611.03814 arxiv:1611.03814 arxiv:1611.03814 AsLAb71x +@tag:Choi2016_gram arxiv:1611.07012 arxiv:1611.07012 10nDTiETi +@tag:Kindermans2016_investigating arxiv:1611.07270 arxiv:1611.07270 b1sc0cgP +@tag:Lundberg2016_an arxiv:1611.07478 arxiv:1611.07478 DeOI1oGf @arxiv:1611.08373 arxiv:1611.08373 arxiv:1611.08373 dO844vZn +@tag:Ragoza2016_protein arxiv:1612.02751 arxiv:1612.02751 bNBiIiTt +@tag:Segler2017_drug_design arxiv:1701.01329 arxiv:1701.01329 8LWFFeYg @arxiv:1701.06599 arxiv:1701.06599 arxiv:1701.06599 apBChoyF +@tag:Murdoch2017_automatic arxiv:1702.02540 arxiv:1702.02540 10ViHstXn +@tag:Zintgraf2017_visualizing arxiv:1702.04595 arxiv:1702.04595 Kk20paR7 +@tag:Litjens2017_medimage_survey arxiv:1702.05747 arxiv:1702.05747 LL5huVs3 +@tag:Wu2017_molecule_net arxiv:1703.00564 arxiv:1703.00564 16OPHvAij +@tag:Sundararajan2017_axiomatic arxiv:1703.01365 arxiv:1703.01365 WzFOJBiA @arxiv:1703.01925 arxiv:1703.01925 arxiv:1703.01925 AQ3N6Ayw @arxiv:1703.02136 arxiv:1703.02136 arxiv:1703.02136 wKioubsT +@tag:Koh2017_understanding arxiv:1703.04730 arxiv:1703.04730 69wxD9y @arxiv:1703.06490v1 arxiv:1703.06490v1 arxiv:1703.06490v1 xl1ijigK @arxiv:1703.10603 arxiv:1703.10603 arxiv:1703.10603 17YaKNLKk @arxiv:1704.01155 arxiv:1704.01155 arxiv:1704.01155 18lZK7fxH +@tag:Kahng2017_activis arxiv:1704.01942 arxiv:1704.01942 QphVo2P2 +@tag:Shrikumar2017_learning arxiv:1704.02685 arxiv:1704.02685 zhmq9ktJ +@tag:Fong2017_perturb arxiv:1704.03296 arxiv:1704.03296 y4t9EzPn @arxiv:1704.04760 arxiv:1704.04760 arxiv:1704.04760 ULagTifF @arxiv:1704.07207 arxiv:1704.07207 arxiv:1704.07207 39RPiE10 +@tag:Olivecrona2017_drug_design arxiv:1704.07555 arxiv:1704.07555 1EayJRsI +@tag:Johnson2017_integ_cell arxiv:1705.00092 arxiv:1705.00092 71c6rs2z @arxiv:1705.02315 arxiv:1705.02315 arxiv:1705.02315 PGi9g7yV @doi:10.1001/jama.2016.17216 doi:10.1001/jama.2016.17216 doi:10.1001/jama.2016.17216 1mJW6umJ +@tag:Gulshan2016_dt doi:10.1001/jama.2016.17216 doi:10.1001/jama.2016.17216 1mJW6umJ @doi:10.1002/9783527628766 doi:10.1002/9783527628766 doi:10.1002/9783527628766 17eGl2pn9 @doi:10.1002/cyto.a.22909 doi:10.1002/cyto.a.22909 doi:10.1002/cyto.a.22909 m3Ij21U8 @doi:10.1002/jcc.24764 doi:10.1002/jcc.24764 doi:10.1002/jcc.24764 zCt6PUXj @doi:10.1002/minf.201501008 doi:10.1002/minf.201501008 doi:10.1002/minf.201501008 gJE0ExFr +@tag:Yasushi2016_cgbvs_dnn doi:10.1002/minf.201600045 doi:10.1002/minf.201600045 x0M6vals @doi:10.1002/wcms.1104 doi:10.1002/wcms.1104 doi:10.1002/wcms.1104 WeiyYhfy @doi:10.1002/wcms.49 doi:10.1002/wcms.49 doi:10.1002/wcms.49 kJ4hy7E @doi:10.1006/jmbi.1999.3091 doi:10.1006/jmbi.1999.3091 doi:10.1006/jmbi.1999.3091 Aic7UyXM @doi:10.1007/978-3-319-16706-0_20 doi:10.1007/978-3-319-16706-0_20 doi:10.1007/978-3-319-16706-0_20 17B2QAA1k +@tag:Dhungel2015_struct_pred_mamm doi:10.1007/978-3-319-24553-9_74 doi:10.1007/978-3-319-24553-9_74 VFw1VXDP @doi:10.1007/978-3-319-24574-4_28 doi:10.1007/978-3-319-24574-4_28 doi:10.1007/978-3-319-24574-4_28 TutLhFSz +@tag:PerezSianes2016_screening doi:10.1007/978-3-319-40126-3_2 doi:10.1007/978-3-319-40126-3_2 1DTUK3YyI @doi:10.1007/978-3-319-44839-8 doi:10.1007/978-3-319-44839-8 doi:10.1007/978-3-319-44839-8 1CWhXZxos @doi:10.1007/978-3-319-46227-1_1 doi:10.1007/978-3-319-46227-1_1 doi:10.1007/978-3-319-46227-1_1 pNoAbBEu +@tag:Mahendran2016_salient doi:10.1007/978-3-319-46466-4_8 doi:10.1007/978-3-319-46466-4_8 vjXoJqO3 +@tag:Dhungel2016_mamm doi:10.1007/978-3-319-46723-8_13 doi:10.1007/978-3-319-46723-8_13 JK8NuXy3 +@tag:Nie2016_3d_survival doi:10.1007/978-3-319-46723-8_25 doi:10.1007/978-3-319-46723-8_25 18EpaZ7QB +@tag:Yoon2016_cancer_reports doi:10.1007/978-3-319-47898-2_21 doi:10.1007/978-3-319-47898-2_21 yUgE09ve @doi:10.1007/978-3-642-40763-5_51 doi:10.1007/978-3-642-40763-5_51 doi:10.1007/978-3-642-40763-5_51 koEdZRcY +@tag:Ciresan2013_mitosis doi:10.1007/978-3-642-40763-5_51 doi:10.1007/978-3-642-40763-5_51 koEdZRcY @doi:10.1007/BF02478259 doi:10.1007/BF02478259 doi:10.1007/bf02478259 1HVDhhwpK @doi:10.1007/s10278-016-9914-9 doi:10.1007/s10278-016-9914-9 doi:10.1007/s10278-016-9914-9 x6HXFAS4 -@doi:10.1016/S0097-8485(96)80015-5 doi:10.1016/S0097-8485(96)80015-5 doi:10.1016/s0097-8485(96)80015-5 3Ew5V1iC -@doi:10.1016/S0140-6736(13)61752-3 doi:10.1016/S0140-6736(13)61752-3 doi:10.1016/s0140-6736(13)61752-3 N96QKgly -@doi:10.1016/S0167-9473(99)00098-5 doi:10.1016/S0167-9473(99)00098-5 doi:10.1016/s0167-9473(99)00098-5 1921Mctzh -@doi:10.1016/S0168-8510(02)00208-7 doi:10.1016/S0168-8510(02)00208-7 doi:10.1016/s0168-8510(02)00208-7 7BctyA7f +@tag:Rajkomar2017_radiographs doi:10.1007/s10278-016-9914-9 doi:10.1007/s10278-016-9914-9 x6HXFAS4 +@tag:Kearnes2016_graph_conv doi:10.1007/s10822-016-9938-8 doi:10.1007/s10822-016-9938-8 145os4Y6t +@tag:Russakovsky2015_imagenet doi:10.1007/s11263-015-0816-y doi:10.1007/s11263-015-0816-y cBVeXnZx +@tag:Mahendran2016_visualizing doi:10.1007/s11263-016-0911-8 doi:10.1007/s11263-016-0911-8 1FkT6C6oa @doi:10.1016/j.bjid.2015.08.013 doi:10.1016/j.bjid.2015.08.013 doi:10.1016/j.bjid.2015.08.013 oas5tbC7 +@tag:Rosenberg2015_synthetic_seqs doi:10.1016/j.cell.2015.09.054 doi:10.1016/j.cell.2015.09.054 mlqKTlZY +@tag:Gaublomme2015_th17 doi:10.1016/j.cell.2015.11.009 doi:10.1016/j.cell.2015.11.009 1HPu3R2B4 @doi:10.1016/j.cels.2016.04.013 doi:10.1016/j.cels.2016.04.013 doi:10.1016/j.cels.2016.04.013 6XtEfQMC @doi:10.1016/j.copbio.2016.04.003 doi:10.1016/j.copbio.2016.04.003 doi:10.1016/j.copbio.2016.04.003 hkKO4QYl @doi:10.1016/j.drudis.2006.06.016 doi:10.1016/j.drudis.2006.06.016 doi:10.1016/j.drudis.2006.06.016 1D6emOV6q @doi:10.1016/j.ijmedinf.2016.09.014 doi:10.1016/j.ijmedinf.2016.09.014 doi:10.1016/j.ijmedinf.2016.09.014 1Ar4f4vfR +@tag:Ithapu2015_efficient doi:10.1016/j.jalz.2015.01.010 doi:10.1016/j.jalz.2015.01.010 eehGXQlY @doi:10.1016/j.jbi.2014.03.016 doi:10.1016/j.jbi.2014.03.016 doi:10.1016/j.jbi.2014.03.016 1C97CTU9S @doi:10.1016/j.jbi.2014.10.006 doi:10.1016/j.jbi.2014.10.006 doi:10.1016/j.jbi.2014.10.006 13filvWwr @doi:10.1016/j.jbi.2016.10.007 doi:10.1016/j.jbi.2016.10.007 doi:10.1016/j.jbi.2016.10.007 5x3uMSKi +@tag:BeaulieuJones2016_ehr_encode doi:10.1016/j.jbi.2016.10.007 doi:10.1016/j.jbi.2016.10.007 5x3uMSKi @doi:10.1016/j.jhealeco.2016.01.012 doi:10.1016/j.jhealeco.2016.01.012 doi:10.1016/j.jhealeco.2016.01.012 13c9OPizf +@tag:Kooi2016_mamm_lesions doi:10.1016/j.media.2016.07.007 doi:10.1016/j.media.2016.07.007 18cZbigDD +@tag:Dhungel2017_mamm_min_interv doi:10.1016/j.media.2017.01.009 doi:10.1016/j.media.2017.01.009 5kfDbGhA @doi:10.1016/j.neubiorev.2010.07.002 doi:10.1016/j.neubiorev.2010.07.002 doi:10.1016/j.neubiorev.2010.07.002 sOBzMC57 +@tag:Lisboa2006_review doi:10.1016/j.neunet.2005.10.007 doi:10.1016/j.neunet.2005.10.007 qxxwkSAT +@tag:Schmidhuber2014_dnn_overview doi:10.1016/j.neunet.2014.09.003 doi:10.1016/j.neunet.2014.09.003 BQS8ClV0 +@tag:Pratt2016_dr doi:10.1016/j.procs.2016.07.014 doi:10.1016/j.procs.2016.07.014 ayTsooEM @doi:10.1016/j.tig.2015.05.007 doi:10.1016/j.tig.2015.05.007 doi:10.1016/j.tig.2015.05.007 J0PJHcHK +@tag:matis doi:10.1016/S0097-8485(96)80015-5 doi:10.1016/s0097-8485(96)80015-5 3Ew5V1iC +@tag:Mahmood doi:10.1016/S0140-6736(13)61752-3 doi:10.1016/s0140-6736(13)61752-3 N96QKgly +@tag:Xiang doi:10.1016/S0167-9473(99)00098-5 doi:10.1016/s0167-9473(99)00098-5 1921Mctzh +@tag:Serden doi:10.1016/S0168-8510(02)00208-7 doi:10.1016/s0168-8510(02)00208-7 7BctyA7f +@tag:Tu1996_anns doi:10.1016/S0895-4356(96)00002-9 doi:10.1016/s0895-4356(96)00002-9 kL0B4m9d +@tag:Subramanian2016_bace1 doi:10.1021/acs.jcim.6b00290 doi:10.1021/acs.jcim.6b00290 B4cL1o2P +@tag:Pereira2016_docking doi:10.1021/acs.jcim.6b00355 doi:10.1021/acs.jcim.6b00355 Gue0c5Gb @doi:10.1021/acs.jmedchem.5b01849 doi:10.1021/acs.jmedchem.5b01849 doi:10.1021/acs.jmedchem.5b01849 omzv9ryI @doi:10.1021/acs.jproteome.6b00618 doi:10.1021/acs.jproteome.6b00618 doi:10.1021/acs.jproteome.6b00618 oTF8O79C @doi:10.1021/acs.molpharmaceut.5b00982 doi:10.1021/acs.molpharmaceut.5b00982 doi:10.1021/acs.molpharmaceut.5b00982 1VZjheOA @doi:10.1021/acs.molpharmaceut.6b00248 doi:10.1021/acs.molpharmaceut.6b00248 doi:10.1021/acs.molpharmaceut.6b00248 EMDwvRGb +@tag:Hughes2016_macromol_react doi:10.1021/acscentsci.6b00162 doi:10.1021/acscentsci.6b00162 1BARarxfz +@tag:AltaeTran2016_one_shot doi:10.1021/acscentsci.6b00367 doi:10.1021/acscentsci.6b00367 P4ixsM8i +@tag:Rogers2010_fingerprints doi:10.1021/ci100050t doi:10.1021/ci100050t QnZ7V9Rd +@tag:Lusci2013_rnn doi:10.1021/ci400187y doi:10.1021/ci400187y 17Wih4Hd5 @doi:10.1021/ci400518g doi:10.1021/ci400518g doi:10.1021/ci400518g 7QsMcDYy @doi:10.1021/ci500340n doi:10.1021/ci500340n doi:10.1021/ci500340n 16FEYidu2 +@tag:Ma2015_qsar_merck doi:10.1021/ci500747n doi:10.1021/ci500747n xOaTIeBY +@tag:Swamidass2009_irv doi:10.1021/ci8004379 doi:10.1021/ci8004379 cjj5vT3H @doi:10.1021/jm048957q doi:10.1021/jm048957q doi:10.1021/jm048957q YO41GAOP @doi:10.1037/h0037350 doi:10.1037/h0037350 doi:10.1037/h0037350 cpNVdlL7 @doi:10.1038/484155a doi:10.1038/484155a doi:10.1038/484155a 1FjSxrV1k +@tag:Barash2010_splicing_code doi:10.1038/nature09000 doi:10.1038/nature09000 11ETDdRKr +@tag:Consortium2012_encode doi:10.1038/nature11247 doi:10.1038/nature11247 15J98V2qM @doi:10.1038/nature14539 doi:10.1038/nature14539 doi:10.1038/nature14539 BeijBSRE +@tag:Silver2016_alphago doi:10.1038/nature16961 doi:10.1038/nature16961 2gn6PKkv @doi:10.1038/nature21056 doi:10.1038/nature21056 doi:10.1038/nature21056 XnYNYoYB +@tag:Esteva2017_skin_cancer_nature doi:10.1038/nature21056 doi:10.1038/nature21056 XnYNYoYB +@tag:Alipanahi2015_predicting doi:10.1038/nbt.3300 doi:10.1038/nbt.3300 jJHZHWrl @doi:10.1038/nbt.3313 doi:10.1038/nbt.3313 doi:10.1038/nbt.3313 yXqhuueV @doi:10.1038/nbt.3780 doi:10.1038/nbt.3780 doi:10.1038/nbt.3780 Qh7xTLwz +@tag:Schatz2010_dna_cloud doi:10.1038/nbt0710-691 doi:10.1038/nbt0710-691 B6g0qKf4 @doi:10.1038/ncomms13090 doi:10.1038/ncomms13090 doi:10.1038/ncomms13090 8SMDF816 @doi:10.1038/ncomms5022 doi:10.1038/ncomms5022 doi:10.1038/ncomms5022 ogs3PPp7 @doi:10.1038/ncomms5308 doi:10.1038/ncomms5308 doi:10.1038/ncomms5308 TDruxF1s @doi:10.1038/ng.2892 doi:10.1038/ng.2892 doi:10.1038/ng.2892 KxEzGxJ6 @doi:10.1038/ng.3830 doi:10.1038/ng.3830 doi:10.1038/ng.3830 194IoYUs3 +@tag:Metaphlan doi:10.1038/nmeth.2066 doi:10.1038/nmeth.2066 56wEWVIl +@tag:Zhou2015_deep_sea doi:10.1038/nmeth.3547 doi:10.1038/nmeth.3547 2UI1BZuD @doi:10.1038/nmeth.3945 doi:10.1038/nmeth.3945 doi:10.1038/nmeth.3945 JmHFuXEM @doi:10.1038/nmeth.4182 doi:10.1038/nmeth.4182 doi:10.1038/nmeth.4182 On4vW5aU +@tag:Buggenthin2017_imaged_lineage doi:10.1038/nmeth.4182 doi:10.1038/nmeth.4182 On4vW5aU +@tag:McHardy doi:10.1038/nmeth976 doi:10.1038/nmeth976 QV551Nlx @doi:10.1038/npjsba.2016.15 doi:10.1038/npjsba.2016.15 doi:10.1038/npjsba.2016.15 cQAldRdg @doi:10.1038/nrd1086 doi:10.1038/nrd1086 doi:10.1038/nrd1086 RAadmvJN @doi:10.1038/nrd4609 doi:10.1038/nrd4609 doi:10.1038/nrd4609 79Ktl2 +@tag:Gawad2016_singlecell doi:10.1038/nrg.2015.16 doi:10.1038/nrg.2015.16 1AWC7HsO0 +@tag:Scotti2016_missplicing doi:10.1038/nrg.2015.3 doi:10.1038/nrg.2015.3 QFK6GapR +@tag:Bracken2016_mirna doi:10.1038/nrg.2016.134 doi:10.1038/nrg.2016.134 yVKIhIAf @doi:10.1038/nrg2813 doi:10.1038/nrg2813 doi:10.1038/nrg2813 T3GG8iJN +@tag:Berezikov2011_mirna doi:10.1038/nrg3079 doi:10.1038/nrg3079 8lpCCppx @doi:10.1038/nrg3208 doi:10.1038/nrg3208 doi:10.1038/nrg3208 FkSZ1qmz @doi:10.1038/nrg3458 doi:10.1038/nrg3458 doi:10.1038/nrg3458 8yA3foA6 @doi:10.1038/nrg3920 doi:10.1038/nrg3920 doi:10.1038/nrg3920 TNPWa7Br @doi:10.1038/srep11476 doi:10.1038/srep11476 doi:10.1038/srep11476 UpFrhdJf @doi:10.1038/srep18962 doi:10.1038/srep18962 doi:10.1038/srep18962 UO8L6nd @doi:10.1038/srep26094 doi:10.1038/srep26094 doi:10.1038/srep26094 WrNCJ9sO +@tag:Litjens2016_histopath_survey doi:10.1038/srep26286 doi:10.1038/srep26286 dQCjqq0Q @doi:10.1038/srep32404 doi:10.1038/srep32404 doi:10.1038/srep32404 1G2xP5yOM +@tag:Wasson1985_clinical doi:10.1056/NEJM198509263131306 doi:10.1056/nejm198509263131306 xX68eyvs @doi:10.1056/NEJMe1516564 doi:10.1056/NEJMe1516564 doi:10.1056/nejme1516564 o0F1MXBC @doi:10.1056/NEJMp1006304 doi:10.1056/NEJMp1006304 doi:10.1056/nejmp1006304 VOQtVhWs @doi:10.1073/pnas.0805923106 doi:10.1073/pnas.0805923106 doi:10.1073/pnas.0805923106 kboAopkh @doi:10.1073/pnas.2136655100 doi:10.1073/pnas.2136655100 doi:10.1073/pnas.2136655100 9XBPQ8b @doi:10.1080/17460441.2016.1201262 doi:10.1080/17460441.2016.1201262 doi:10.1080/17460441.2016.1201262 xPkT1z7D +@tag:Baskin2015_drug_disc doi:10.1080/17460441.2016.1201262 doi:10.1080/17460441.2016.1201262 xPkT1z7D @doi:10.1083/jcb.201610026 doi:10.1083/jcb.201610026 doi:10.1083/jcb.201610026 McjXFLLq @doi:10.1086/512821 doi:10.1086/512821 doi:10.1086/512821 10shRODux @doi:10.1093/bib/4.1.22 doi:10.1093/bib/4.1.22 doi:10.1093/bib/4.1.22 19jjiGHWc @@ -111,345 +233,223 @@ text citation standard_citation citation_id @doi:10.1093/bib/bbw110 doi:10.1093/bib/bbw110 doi:10.1093/bib/bbw110 ir7ElHha @doi:10.1093/bib/bbw112 doi:10.1093/bib/bbw112 doi:10.1093/bib/bbw112 1BkEtNVsj @doi:10.1093/bib/bbx044 doi:10.1093/bib/bbx044 doi:10.1093/bib/bbx044 11I7bLcP3 +@tag:Stormo2000_dna doi:10.1093/bioinformatics/16.1.16 doi:10.1093/bioinformatics/16.1.16 ywDQIvZJ +@tag:Hochreiter doi:10.1093/bioinformatics/btm247 doi:10.1093/bioinformatics/btm247 G8RKF6sz +@tag:nbc doi:10.1093/bioinformatics/btq619 doi:10.1093/bioinformatics/btq619 1HtJuEkb2 +@tag:Xiong2011_bayesian doi:10.1093/bioinformatics/btr444 doi:10.1093/bioinformatics/btr444 8VPGUHcf @doi:10.1093/bioinformatics/bts475 doi:10.1093/bioinformatics/bts475 doi:10.1093/bioinformatics/bts475 xdoT1yUx @doi:10.1093/bioinformatics/bts598 doi:10.1093/bioinformatics/bts598 doi:10.1093/bioinformatics/bts598 18bNbDNlc @doi:10.1093/bioinformatics/btt234 doi:10.1093/bioinformatics/btt234 doi:10.1093/bioinformatics/btt234 1AU7wzPqa +@tag:lmat doi:10.1093/bioinformatics/btt389 doi:10.1093/bioinformatics/btt389 189TQrQA9 @doi:10.1093/bioinformatics/btu703 doi:10.1093/bioinformatics/btu703 doi:10.1093/bioinformatics/btu703 15E5yG1Ho @doi:10.1093/bioinformatics/btu791 doi:10.1093/bioinformatics/btu791 doi:10.1093/bioinformatics/btu791 7atXz0r +@tag:Chen2015_trans_species doi:10.1093/bioinformatics/btv315 doi:10.1093/bioinformatics/btv315 rmjDc5rm @doi:10.1093/bioinformatics/btv472 doi:10.1093/bioinformatics/btv472 doi:10.1093/bioinformatics/btv472 kqjqFesT +@tag:Vervier doi:10.1093/bioinformatics/btv683 doi:10.1093/bioinformatics/btv683 8DLzxOEt +@tag:Chen2016_gene_expr doi:10.1093/bioinformatics/btw074 doi:10.1093/bioinformatics/btw074 12QQw9p7v +@tag:Zeng2016_convolutional doi:10.1093/bioinformatics/btw255 doi:10.1093/bioinformatics/btw255 182UhQqzp @doi:10.1093/ije/dyn022 doi:10.1093/ije/dyn022 doi:10.1093/ije/dyn022 CVnO5njl @doi:10.1093/ije/dyu188 doi:10.1093/ije/dyu188 doi:10.1093/ije/dyu188 SfxIiPJ1 @doi:10.1093/ije/dyv193 doi:10.1093/ije/dyv193 doi:10.1093/ije/dyv193 1D6b3tMu9 @doi:10.1093/jamia/ocv111 doi:10.1093/jamia/ocv111 doi:10.1093/jamia/ocv111 2cYqPKf1 @doi:10.1093/jamia/ocw011 doi:10.1093/jamia/ocw011 doi:10.1093/jamia/ocw011 A9JeoGV8 +@tag:Horton1992_assessment doi:10.1093/nar/20.16.4331 doi:10.1093/nar/20.16.4331 uZvDdFZo +@tag:Hoff doi:10.1093/nar/gkp327 doi:10.1093/nar/gkp327 q1A2AEtO @doi:10.1093/nar/gkq747 doi:10.1093/nar/gkq747 doi:10.1093/nar/gkq747 QlbXLqH @doi:10.1093/nar/gkr777 doi:10.1093/nar/gkr777 doi:10.1093/nar/gkr777 x1nE5icc @doi:10.1093/nar/gku1058 doi:10.1093/nar/gku1058 doi:10.1093/nar/gku1058 12aqvAgz6 +@tag:Gupta2015_exprs_yeast doi:10.1101/031906 doi:10.1101/031906 AnenJOuU @doi:10.1101/036129 doi:10.1101/036129 doi:10.1101/036129 s5sy4AOi @doi:10.1101/041616 doi:10.1101/041616 doi:10.1101/041616 1HbQQcY2q +@tag:Arvaniti2016_rare_subsets doi:10.1101/046508 doi:10.1101/046508 r3Gbjksq +@tag:Koh2016_denoising doi:10.1101/052118 doi:10.1101/052118 XimuXZlz @doi:10.1101/054775 doi:10.1101/054775 doi:10.1101/054775 1Hk3NTSn2 @doi:10.1101/070441 doi:10.1101/070441 doi:10.1101/070441 11NHbWB1V +@tag:Tan2016_eadage doi:10.1101/078659 doi:10.1101/078659 zuLdSQx3 +@tag:Torracinta2016_sim doi:10.1101/079087 doi:10.1101/079087 ECTm1SuA @doi:10.1101/081364 doi:10.1101/081364 doi:10.1101/081364 gllSeTW +@tag:Eser2016_fiddle doi:10.1101/081380 doi:10.1101/081380 yOz8Ybj2 +@tag:Leibig2016_dr doi:10.1101/084210 doi:10.1101/084210 14Ovc5nPg @doi:10.1101/085118 doi:10.1101/085118 doi:10.1101/085118 BMg062hc +@tag:Pawlowski2016 doi:10.1101/085118 doi:10.1101/085118 BMg062hc @doi:10.1101/085241 doi:10.1101/085241 doi:10.1101/085241 14TqLB9iZ +@tag:Poplin2016_deepvariant doi:10.1101/092890 doi:10.1101/092890 FVfZESYP +@tag:Lee2016_emr_oct_amd doi:10.1101/094276 doi:10.1101/094276 SxsZyrVM +@tag:Artemov2016_clinical doi:10.1101/095653 doi:10.1101/095653 mo3GQwJj @doi:10.1101/095786 doi:10.1101/095786 doi:10.1101/095786 Xxb4t3zO +@tag:Zhu2016_advers_mamm doi:10.1101/095786 doi:10.1101/095786 Xxb4t3zO @doi:10.1101/095794 doi:10.1101/095794 doi:10.1101/095794 9G9Hv1Pp +@tag:Zhu2016_mult_inst_mamm doi:10.1101/095794 doi:10.1101/095794 9G9Hv1Pp +@tag:Torracinta2016_deep_snp doi:10.1101/097469 doi:10.1101/097469 GSLRw2L5 +@tag:Shrikumar2017_reversecomplement doi:10.1101/103663 doi:10.1101/103663 iEmvzeT8 +@tag:Jha2017_integrative_models doi:10.1101/104869 doi:10.1101/104869 N0HBi8MH +@tag:Finnegan2017_maximum doi:10.1101/105957 doi:10.1101/105957 VjsZbMSz +@tag:Qiu2017_graph_embedding doi:10.1101/110668 doi:10.1101/110668 Oljj2W96 +@tag:Chaudhary2017_multiom_liver_cancer doi:10.1101/114892 doi:10.1101/114892 obeRVckH @doi:10.1101/125229 doi:10.1101/125229 doi:10.1101/125229 2M3zXijc @doi:10.1101/138685 doi:10.1101/138685 doi:10.1101/138685 1CAw3FaPI @doi:10.1101/gr.110254.110 doi:10.1101/gr.110254.110 doi:10.1101/gr.110254.110 G0J9P3Ln @doi:10.1101/gr.200535.115 doi:10.1101/gr.200535.115 doi:10.1101/gr.200535.115 2CbHXoFn +@tag:Kelley2016_basset doi:10.1101/gr.200535.115 doi:10.1101/gr.200535.115 2CbHXoFn +@tag:Abe doi:10.1101/gr.634603 doi:10.1101/gr.634603 1HhqhBwrM @doi:10.1103/RevModPhys.34.135 doi:10.1103/RevModPhys.34.135 doi:10.1103/revmodphys.34.135 1G5eCiq4d +@tag:Chryssolouris1996_confidence doi:10.1109/72.478409 doi:10.1109/72.478409 9SnNyc8Y +@tag:Huddar2016_predicting doi:10.1109/ACCESS.2016.2618775 doi:10.1109/access.2016.2618775 fOaBA9Vc @doi:10.1109/BIBM.2014.6999129 doi:10.1109/BIBM.2014.6999129 doi:10.1109/bibm.2014.6999129 TeIxEjqm -@doi:10.1109/TCBB.2014.2343960 doi:10.1109/TCBB.2014.2343960 doi:10.1109/tcbb.2014.2343960 ZzaRyGuJ +@tag:Min2016_deepenhancer doi:10.1109/BIBM.2016.7822593 doi:10.1109/bibm.2016.7822593 jV2YerUS +@tag:Esfahani2016_melanoma doi:10.1109/EMBC.2016.7590963 doi:10.1109/embc.2016.7590963 O39LDkX +@tag:Nemati2016_rl doi:10.1109/EMBC.2016.7591355 doi:10.1109/embc.2016.7591355 16OQvsRqJ +@tag:Seide2014_parallel doi:10.1109/ICASSP.2014.6853593 doi:10.1109/icassp.2014.6853593 IULiPa6L @doi:10.1109/icnn.1994.374637 doi:10.1109/icnn.1994.374637 doi:10.1109/icnn.1994.374637 YUms527e +@tag:Essinger2010_taxonomic doi:10.1109/IJCNN.2010.5596644 doi:10.1109/ijcnn.2010.5596644 11wVLI2Hn +@tag:Burlina2016_amd doi:10.1109/ISBI.2016.7493240 doi:10.1109/isbi.2016.7493240 iBPOt78R +@doi:10.1109/TCBB.2014.2343960 doi:10.1109/TCBB.2014.2343960 doi:10.1109/tcbb.2014.2343960 ZzaRyGuJ @doi:10.1109/tcbb.2014.2343960 doi:10.1109/tcbb.2014.2343960 doi:10.1109/tcbb.2014.2343960 ZzaRyGuJ +@tag:Liang2015_exprs_cancer doi:10.1109/TCBB.2014.2377729 doi:10.1109/tcbb.2014.2377729 1EtavGKI4 +@tag:Roth2015_view_agg_cad doi:10.1109/TMI.2015.2482920 doi:10.1109/tmi.2015.2482920 KseoWN2w +@tag:Shin2016_cad_tl doi:10.1109/TMI.2016.2528162 doi:10.1109/tmi.2016.2528162 1GAyqYBNZ +@tag:Yu2016_melanoma_resnet doi:10.1109/TMI.2016.2642839 doi:10.1109/tmi.2016.2642839 phRCihNB +@tag:Ditzler3 doi:10.1109/TNB.2015.2461219 doi:10.1109/tnb.2015.2461219 i38A0beL +@tag:Ditzler2 doi:10.1109/TNNLS.2014.2320415 doi:10.1109/tnnls.2014.2320415 O9D66oYa +@tag:Knights doi:10.1111/j.1574-6976.2010.00251.x doi:10.1111/j.1574-6976.2010.00251.x aI9g2UOc @doi:10.1111/j.1740-9713.2016.00960.x doi:10.1111/j.1740-9713.2016.00960.x doi:10.1111/j.1740-9713.2016.00960.x 6co0adq +@tag:Ghosh1992_sequence doi:10.1117/12.140112 doi:10.1117/12.140112 oc44yBj0 +@tag:Bar2015_nonmed_tl doi:10.1117/12.2083124 doi:10.1117/12.2083124 1Fy5bcnCI +@tag:Amit2017_breast_mri doi:10.1117/12.2249981 doi:10.1117/12.2249981 SOi9mAC2 @doi:10.1126/science.1132939 doi:10.1126/science.1132939 doi:10.1126/science.1132939 Ot5bUkmI +@tag:Xiong2015_splicing_code doi:10.1126/science.1254806 doi:10.1126/science.1254806 17sgPdcMT @doi:10.1126/science.333.6041.404-a doi:10.1126/science.333.6041.404-a doi:10.1126/science.333.6041.404-a DmQPI43R +@tag:Lodato2015_neurons doi:10.1126/science.aab1785 doi:10.1126/science.aab1785 1GvfSy48x +@tag:Li2016_variation doi:10.1126/science.aad9417 doi:10.1126/science.aad9417 b6p6wxpC @doi:10.1126/science.aah6168 doi:10.1126/science.aah6168 doi:10.1126/science.aah6168 gvyja7v1 +@tag:Karlin doi:10.1128/jb.179.12.3899-3913.1997 doi:10.1128/jb.179.12.3899-3913.1997 N9NzkOjA +@tag:Tan2015_adage doi:10.1128/mSystems.00025-15 doi:10.1128/msystems.00025-15 1CFhfCyWN @doi:10.1136/amiajnl-2011-000150 doi:10.1136/amiajnl-2011-000150 doi:10.1136/amiajnl-2011-000150 sG3iVOTS @doi:10.1136/amiajnl-2011-000501 doi:10.1136/amiajnl-2011-000501 doi:10.1136/amiajnl-2011-000501 uDaRUyh9 @doi:10.1136/amiajnl-2011-000681 doi:10.1136/amiajnl-2011-000681 doi:10.1136/amiajnl-2011-000681 odRiFxnB +@tag:Gultepe2014_sepsis doi:10.1136/amiajnl-2013-001815 doi:10.1136/amiajnl-2013-001815 eCrLGgiX @doi:10.1136/amiajnl-2013-001935 doi:10.1136/amiajnl-2013-001935 doi:10.1136/amiajnl-2013-001935 11OyzMl87 @doi:10.1142/9789813207813_0050 doi:10.1142/9789813207813_0050 doi:10.1142/9789813207813_0050 qe90c1CL @doi:10.1142/9789814644730_0014 doi:10.1142/9789814644730_0014 doi:10.1142/9789814644730_0014 PBiRSdXv +@tag:Tan2014_psb doi:10.1142/9789814644730_0014 doi:10.1142/9789814644730_0014 PBiRSdXv +@tag:TAC-ELM doi:10.1142/S0219720012500151 doi:10.1142/s0219720012500151 Wz7VUS03 @doi:10.1145/1143844.1143874 doi:10.1145/1143844.1143874 doi:10.1145/1143844.1143874 JNnkm5Zt +@tag:Mapreduce doi:10.1145/1327452.1327492 doi:10.1145/1327452.1327492 xE3EYmck +@tag:Raina2009_gpu doi:10.1145/1553374.1553486 doi:10.1145/1553374.1553486 F3e4wfzQ +@tag:RAD2010_view_cc doi:10.1145/1721654.1721672 doi:10.1145/1721654.1721672 ObFN78yp @doi:10.1145/2382577.2382579 doi:10.1145/2382577.2382579 doi:10.1145/2382577.2382579 889LsjDi +@tag:Li2014_minibatch doi:10.1145/2623330.2623612 doi:10.1145/2623330.2623612 fNkl8HFz @doi:10.1145/2661829.2661974 doi:10.1145/2661829.2661974 doi:10.1145/2661829.2661974 XQtuRkTU +@tag:Edwards2015_growing_pains doi:10.1145/2771283 doi:10.1145/2771283 1FocAi7N0 +@tag:Zhang2015_multitask_tl doi:10.1145/2783258.2783304 doi:10.1145/2783258.2783304 HlDY7trA @doi:10.1145/2939672.2939823 doi:10.1145/2939672.2939823 doi:10.1145/2939672.2939823 TwvauiTv +@tag:Shen2017_medimg_review doi:10.1146/annurev-bioeng-071516-044442 doi:10.1146/annurev-bioeng-071516-044442 yEstnIOT +@tag:Vera2016_sc_analysis doi:10.1146/annurev-genet-120215-034854 doi:10.1146/annurev-genet-120215-034854 v97iPXDw @doi:10.1148/radiol.2017162326 doi:10.1148/radiol.2017162326 doi:10.1148/radiol.2017162326 Qve94Jra +@tag:Lakhani2017_radiography doi:10.1148/radiol.2017162326 doi:10.1148/radiol.2017162326 Qve94Jra @doi:10.1155/2013/769639 doi:10.1155/2013/769639 doi:10.1155/2013/769639 3JyJ3DTh @doi:10.1158/1078-0432.CCR-13-0583 doi:10.1158/1078-0432.CCR-13-0583 doi:10.1158/1078-0432.ccr-13-0583 pEIw87Mp +@tag:Abramoff2016_dr doi:10.1167/iovs.16-19964 doi:10.1167/iovs.16-19964 e3vyHBV2 +@tag:Salzberg doi:10.1186/1471-2105-11-544 doi:10.1186/1471-2105-11-544 c4rnN1wo +@tag:yok doi:10.1186/1471-2105-12-20 doi:10.1186/1471-2105-12-20 qUGH5CX8 @doi:10.1186/1758-2946-5-30 doi:10.1186/1758-2946-5-30 doi:10.1186/1758-2946-5-30 M1EW8Rfl +@tag:Stratnikov doi:10.1186/2049-2618-1-11 doi:10.1186/2049-2618-1-11 c5P9jHCg +@tag:Stein2010_cloud doi:10.1186/gb-2010-11-5-207 doi:10.1186/gb-2010-11-5-207 q0SsFrZd +@tag:Zeng2015 doi:10.1186/s12859-015-0553-9 doi:10.1186/s12859-015-0553-9 z3I2IudI +@tag:Ding doi:10.1186/s12859-015-0753-3 doi:10.1186/s12859-015-0753-3 5W4KMSdT +@tag:Ditzler doi:10.1186/s12859-015-0793-8 doi:10.1186/s12859-015-0793-8 1AN5UPfb1 @doi:10.1186/s12859-015-0845-0 doi:10.1186/s12859-015-0845-0 doi:10.1186/s12859-015-0845-0 18lqFDKRR +@tag:Chen2016_exprs_yeast doi:10.1186/s12859-015-0852-1 doi:10.1186/s12859-015-0852-1 yVBx9Qx4 @doi:10.1186/s12859-017-1561-8 doi:10.1186/s12859-017-1561-8 doi:10.1186/s12859-017-1561-8 qnKdqG0P +@tag:Gerstein2016_scaling doi:10.1186/s13059-016-0917-0 doi:10.1186/s13059-016-0917-0 1E7bFCRV4 +@tag:Angermueller2016_single_methyl doi:10.1186/s13059-017-1189-z doi:10.1186/s13059-017-1189-z 19EJTHByG +@tag:Lusci2015_irv doi:10.1186/s13321-015-0110-6 doi:10.1186/s13321-015-0110-6 1E0x7QgLP @doi:10.1200/JCO.2008.18.1370 doi:10.1200/JCO.2008.18.1370 doi:10.1200/jco.2008.18.1370 lnK82Ey6 @doi:10.1208/s12248-012-9322-0 doi:10.1208/s12248-012-9322-0 doi:10.1208/s12248-012-9322-0 13iyYvEcB +@tag:Liu2016_sc_transcriptome doi:10.12688/f1000research.7223.1 doi:10.12688/f1000research.7223.1 QafUwNKn @doi:10.1371/journal.pbio.1001661 doi:10.1371/journal.pbio.1001661 doi:10.1371/journal.pbio.1001661 dKwyEWWF @doi:10.1371/journal.pbio.1002195 doi:10.1371/journal.pbio.1002195 doi:10.1371/journal.pbio.1002195 13bxiY1vo @doi:10.1371/journal.pcbi.1002957 doi:10.1371/journal.pcbi.1002957 doi:10.1371/journal.pcbi.1002957 5tvnB4uW @doi:10.1371/journal.pcbi.1003711 doi:10.1371/journal.pcbi.1003711 doi:10.1371/journal.pcbi.1003711 JxuQvvyk +@tag:Ghandi2014_enhanced doi:10.1371/journal.pcbi.1003711 doi:10.1371/journal.pcbi.1003711 JxuQvvyk @doi:10.1371/journal.pcbi.1003889 doi:10.1371/journal.pcbi.1003889 doi:10.1371/journal.pcbi.1003889 F13xtRbV +@tag:Setty2015_seqgl doi:10.1371/journal.pcbi.1004271 doi:10.1371/journal.pcbi.1004271 138dgb9Ca +@tag:Segata doi:10.1371/journal.pcbi.1004977 doi:10.1371/journal.pcbi.1004977 y9s5irW @doi:10.1371/journal.pcbi.1005135 doi:10.1371/journal.pcbi.1005135 doi:10.1371/journal.pcbi.1005135 1SIuofeg @doi:10.1371/journal.pcbi.1005177 doi:10.1371/journal.pcbi.1005177 doi:10.1371/journal.pcbi.1005177 40EG4ZEU @doi:10.1371/journal.pcbi.1005219 doi:10.1371/journal.pcbi.1005219 doi:10.1371/journal.pcbi.1005219 tOpadZQw @doi:10.1371/journal.pcbi.1005324 doi:10.1371/journal.pcbi.1005324 doi:10.1371/journal.pcbi.1005324 BhfjKSY3 +@tag:Wang2016_protein_contact doi:10.1371/journal.pcbi.1005324 doi:10.1371/journal.pcbi.1005324 BhfjKSY3 +@tag:Qin2017_onehot doi:10.1371/journal.pcbi.1005403 doi:10.1371/journal.pcbi.1005403 Qbtqlmhf @doi:10.1371/journal.pone.0028766 doi:10.1371/journal.pone.0028766 doi:10.1371/journal.pone.0028766 10dNuD89l @doi:10.1371/journal.pone.0032235 doi:10.1371/journal.pone.0032235 doi:10.1371/journal.pone.0032235 1AlhRKQbe @doi:10.1371/journal.pone.0061318 doi:10.1371/journal.pone.0061318 doi:10.1371/journal.pone.0061318 QcwZC8wG @doi:10.1371/journal.pone.0066341 doi:10.1371/journal.pone.0066341 doi:10.1371/journal.pone.0066341 FLX0o7bL +@tag:wgsquikr doi:10.1371/journal.pone.0091784 doi:10.1371/journal.pone.0091784 RqhGD9c7 +@tag:Liu doi:10.1371/journal.pone.0097958 doi:10.1371/journal.pone.0097958 Kt9NojjR +@tag:Bach2015_on doi:10.1371/journal.pone.0130140 doi:10.1371/journal.pone.0130140 au5CLIOH +@tag:Asgari doi:10.1371/journal.pone.0141287 doi:10.1371/journal.pone.0141287 1E1PWjqTm @doi:10.1371/journal.pone.0171410 doi:10.1371/journal.pone.0171410 doi:10.1371/journal.pone.0171410 as2HfoSh +@tag:Graphlab doi:10.14778/2212351.2212354 doi:10.14778/2212351.2212354 1XcexUAV +@tag:Shapely doi:10.1515/9781400881970-018 doi:10.1515/9781400881970-018 YBJdA6LJ +@tag:Soueidan doi:10.1515/metgen-2016-0001 doi:10.1515/metgen-2016-0001 yFOAeemA @doi:10.15252/msb.20156651 doi:10.15252/msb.20156651 doi:10.15252/msb.20156651 irSe12Sm +@tag:Angermueller2016_dl_review doi:10.15252/msb.20156651 doi:10.15252/msb.20156651 irSe12Sm +@tag:Kraus2017_deeploc doi:10.15252/msb.20177551 doi:10.15252/msb.20177551 DcnNfASG +@tag:JuanMateu2016_t1d doi:10.1530/EJE-15-0916 doi:10.1530/eje-15-0916 CNz9HwZ3 +@tag:Parnamaa2017 doi:10.1534/g3.116.033654 doi:10.1534/g3.116.033654 2a7MHtAx @doi:10.1561/0400000042 doi:10.1561/0400000042 doi:10.1561/0400000042 v8Lp4ibI @doi:10.1590/2176-9451.19.4.027-029.ebo doi:10.1590/2176-9451.19.4.027-029.ebo doi:10.1590/2176-9451.19.4.027-029.ebo Fx5qVQlk @doi:10.2307/2281868 doi:10.2307/2281868 doi:10.2307/2281868 6RHepB1T @doi:10.2337/diab.35.2.237 doi:10.2337/diab.35.2.237 doi:10.2337/diab.35.2.237 ws1zvGoZ @doi:10.2337/diabetes.54.suppl_2.S68 doi:10.2337/diabetes.54.suppl_2.S68 doi:10.2337/diabetes.54.suppl_2.s68 WuOqsORY @doi:10.3109/10409238.2015.1135868 doi:10.3109/10409238.2015.1135868 doi:10.3109/10409238.2015.1135868 MmRGFVUu +@tag:Mayr2016_deep_tox doi:10.3389/fenvs.2015.00080 doi:10.3389/fenvs.2015.00080 Y1D0SZrO @doi:10.3389/fgene.2014.00342 doi:10.3389/fgene.2014.00342 doi:10.3389/fgene.2014.00342 ppGS5h4v +@tag:Baxt1991_myocardial doi:10.7326/0003-4819-115-11-843 doi:10.7326/0003-4819-115-11-843 jdg2u7bX +@tag:Agarwal2015_targetscan doi:10.7554/eLife.05005 doi:10.7554/elife.05005 12vPQi3gp @pmid:21347133 pmid:21347133 pmid:21347133 y9ONtSZ9 @pmid:24159271 pmid:24159271 pmid:24159271 11sli93ov @pmid:26958203 pmid:26958203 pmid:26958203 FUIfIdE @pmid:27134610 pmid:27134610 pmid:27134610 4rTluXLs -@tag:Abe doi:10.1101/gr.634603 doi:10.1101/gr.634603 1HhqhBwrM -@tag:Abramoff2016_dr doi:10.1167/iovs.16-19964 doi:10.1167/iovs.16-19964 e3vyHBV2 -@tag:Agarwal2015_targetscan doi:10.7554/eLife.05005 doi:10.7554/elife.05005 12vPQi3gp -@tag:Alipanahi2015_predicting doi:10.1038/nbt.3300 doi:10.1038/nbt.3300 jJHZHWrl -@tag:AltaeTran2016_one_shot doi:10.1021/acscentsci.6b00367 doi:10.1021/acscentsci.6b00367 P4ixsM8i -@tag:Amit2017_breast_mri doi:10.1117/12.2249981 doi:10.1117/12.2249981 SOi9mAC2 -@tag:Angermueller2016_dl_review doi:10.15252/msb.20156651 doi:10.15252/msb.20156651 irSe12Sm -@tag:Angermueller2016_single_methyl doi:10.1186/s13059-017-1189-z doi:10.1186/s13059-017-1189-z 19EJTHByG -@tag:Artemov2016_clinical doi:10.1101/095653 doi:10.1101/095653 mo3GQwJj -@tag:Arvaniti2016_rare_subsets doi:10.1101/046508 doi:10.1101/046508 r3Gbjksq -@tag:Asgari doi:10.1371/journal.pone.0141287 doi:10.1371/journal.pone.0141287 1E1PWjqTm -@tag:Bach2015_on doi:10.1371/journal.pone.0130140 doi:10.1371/journal.pone.0130140 au5CLIOH -@tag:Bahdanu2014_neural arxiv:1409.0473 arxiv:1409.0473 haHzVaaz -@tag:Bar2015_nonmed_tl doi:10.1117/12.2083124 doi:10.1117/12.2083124 1Fy5bcnCI -@tag:Barash2010_splicing_code doi:10.1038/nature09000 doi:10.1038/nature09000 11ETDdRKr -@tag:Baskin2015_drug_disc doi:10.1080/17460441.2016.1201262 doi:10.1080/17460441.2016.1201262 xPkT1z7D -@tag:Baxt1991_myocardial doi:10.7326/0003-4819-115-11-843 doi:10.7326/0003-4819-115-11-843 jdg2u7bX -@tag:BeaulieuJones2016_ehr_encode doi:10.1016/j.jbi.2016.10.007 doi:10.1016/j.jbi.2016.10.007 5x3uMSKi -@tag:Bengio2015_prec arxiv:1412.7024 arxiv:1412.7024 1G3owNNps -@tag:Berezikov2011_mirna doi:10.1038/nrg3079 doi:10.1038/nrg3079 8lpCCppx -@tag:Bergstra2011_hyper url:https://papers.nips.cc/paper/4443-algorithms-for-hyper-parameter-optimization.pdf url:https://papers.nips.cc/paper/4443-algorithms-for-hyper-parameter-optimization.pdf wz83yfHF -@tag:Bergstra2012_random url:http://www.jmlr.org/papers/v13/bergstra12a.html url:http://www.jmlr.org/papers/v13/bergstra12a.html 1FSwIjR9s -@tag:Boza arxiv:1603.09195 arxiv:1603.09195 1BTJ1KqRa -@tag:Bracken2016_mirna doi:10.1038/nrg.2016.134 doi:10.1038/nrg.2016.134 yVKIhIAf -@tag:Buggenthin2017_imaged_lineage doi:10.1038/nmeth.4182 doi:10.1038/nmeth.4182 On4vW5aU -@tag:Burlina2016_amd doi:10.1109/ISBI.2016.7493240 doi:10.1109/isbi.2016.7493240 iBPOt78R -@tag:Caruana2014_need arxiv:1312.6184 arxiv:1312.6184 1AhGoHZP9 -@tag:Chaudhary2017_multiom_liver_cancer doi:10.1101/114892 doi:10.1101/114892 obeRVckH -@tag:Che2015_distill arxiv:1512.03542 arxiv:1512.03542 14DAmZTDg -@tag:Che2016_rnn arxiv:1606.01865 arxiv:1606.01865 O7Vbecm2 -@tag:Chen2015_hashing arxiv:1504.04788 arxiv:1504.04788 15lYGmZpY -@tag:Chen2015_trans_species doi:10.1093/bioinformatics/btv315 doi:10.1093/bioinformatics/btv315 rmjDc5rm -@tag:Chen2016_exprs_yeast doi:10.1186/s12859-015-0852-1 doi:10.1186/s12859-015-0852-1 yVBx9Qx4 -@tag:Chen2016_gene_expr doi:10.1093/bioinformatics/btw074 doi:10.1093/bioinformatics/btw074 12QQw9p7v -@tag:Choi2016_gram arxiv:1611.07012 arxiv:1611.07012 10nDTiETi -@tag:Choi2016_retain arxiv:1608.05745 arxiv:1608.05745 UcRbawKo -@tag:Chollet2016_xception arxiv:1610.02357 arxiv:1610.02357 VMkPJjVk -@tag:Chryssolouris1996_confidence doi:10.1109/72.478409 doi:10.1109/72.478409 9SnNyc8Y -@tag:Ciresan2013_mitosis doi:10.1007/978-3-642-40763-5_51 doi:10.1007/978-3-642-40763-5_51 koEdZRcY -@tag:Coates2013_cots_hpc url:http://www.jmlr.org/proceedings/papers/v28/coates13.html url:http://www.jmlr.org/proceedings/papers/v28/coates13.html 4MZ2tmZ8 -@tag:Codella2016_ensemble_melanoma arxiv:1610.04662 arxiv:1610.04662 sLPsrfbl -@tag:Consortium2012_encode doi:10.1038/nature11247 doi:10.1038/nature11247 15J98V2qM -@tag:CudNN arxiv:1410.0759 arxiv:1410.0759 YwdqeYZi -@tag:Dahl2014_multi_qsar arxiv:1406.1231 arxiv:1406.1231 1Dzz0P0qr -@tag:Dean2012_nips_downpour url:http://research.google.com/archive/large_deep_networks_nips2012.html url:http://research.google.com/archive/large_deep_networks_nips2012.html 17cBimWgp -@tag:DeepChem url:https://github.com/deepchem/deepchem url:https://github.com/deepchem/deepchem Ytvk62dX -@tag:Deming2016_genetic arxiv:1605.07156 arxiv:1605.07156 SAvEOARL -@tag:Dhungel2015_struct_pred_mamm doi:10.1007/978-3-319-24553-9_74 doi:10.1007/978-3-319-24553-9_74 VFw1VXDP -@tag:Dhungel2016_mamm doi:10.1007/978-3-319-46723-8_13 doi:10.1007/978-3-319-46723-8_13 JK8NuXy3 -@tag:Dhungel2017_mamm_min_interv doi:10.1016/j.media.2017.01.009 doi:10.1016/j.media.2017.01.009 5kfDbGhA -@tag:Ding doi:10.1186/s12859-015-0753-3 doi:10.1186/s12859-015-0753-3 5W4KMSdT -@tag:Ditzler doi:10.1186/s12859-015-0793-8 doi:10.1186/s12859-015-0793-8 1AN5UPfb1 -@tag:Ditzler2 doi:10.1109/TNNLS.2014.2320415 doi:10.1109/tnnls.2014.2320415 O9D66oYa -@tag:Ditzler3 doi:10.1109/TNB.2015.2461219 doi:10.1109/tnb.2015.2461219 i38A0beL @tag:Dogwild pmid:27330264 pmid:27330264 188FA7whS -@tag:Dragonn url:http://kundajelab.github.io/dragonn/ url:http://kundajelab.github.io/dragonn/ 117PEpTMe -@tag:Dream_tf_binding url:https://www.synapse.org/#!Synapse:syn6131484/wiki/402026 url:https://www.synapse.org/#!Synapse:syn6131484/wiki/402026 wW6QbBXz -@tag:Duvenaud2015_graph_conv url:http://papers.nips.cc/paper/5954-convolutional-networks-on-graphs-for-learning-molecular-fingerprints url:http://papers.nips.cc/paper/5954-convolutional-networks-on-graphs-for-learning-molecular-fingerprints Oe573FaL -@tag:Edwards2015_growing_pains doi:10.1145/2771283 doi:10.1145/2771283 1FocAi7N0 -@tag:Ehran2009_visualizing url:http://www.iro.umontreal.ca/~lisa/publications2/index.php/publications/show/247 url:http://www.iro.umontreal.ca/~lisa/publications2/index.php/publications/show/247 UAAd9Uez -@tag:Elephas url:https://github.com/maxpumperla/elephas url:https://github.com/maxpumperla/elephas y9IoEy4r -@tag:Eser2016_fiddle doi:10.1101/081380 doi:10.1101/081380 yOz8Ybj2 -@tag:Esfahani2016_melanoma doi:10.1109/EMBC.2016.7590963 doi:10.1109/embc.2016.7590963 O39LDkX -@tag:Essinger2010_taxonomic doi:10.1109/IJCNN.2010.5596644 doi:10.1109/ijcnn.2010.5596644 11wVLI2Hn -@tag:Esteva2017_skin_cancer_nature doi:10.1038/nature21056 doi:10.1038/nature21056 XnYNYoYB @tag:Faruqi url:http://alifar76.github.io/sklearn-metrics/ url:http://alifar76.github.io/sklearn-metrics/ NQ5jiN7B -@tag:Finnegan2017_maximum doi:10.1101/105957 doi:10.1101/105957 VjsZbMSz -@tag:Fong2017_perturb arxiv:1704.03296 arxiv:1704.03296 y4t9EzPn -@tag:Gal2015_dropout arxiv:1506.02142 arxiv:1506.02142 1FDihfnM -@tag:Gaublomme2015_th17 doi:10.1016/j.cell.2015.11.009 doi:10.1016/j.cell.2015.11.009 1HPu3R2B4 -@tag:Gawad2016_singlecell doi:10.1038/nrg.2015.16 doi:10.1038/nrg.2015.16 1AWC7HsO0 -@tag:Gerstein2016_scaling doi:10.1186/s13059-016-0917-0 doi:10.1186/s13059-016-0917-0 1E7bFCRV4 -@tag:Ghandi2014_enhanced doi:10.1371/journal.pcbi.1003711 doi:10.1371/journal.pcbi.1003711 JxuQvvyk -@tag:Ghosh1992_sequence doi:10.1117/12.140112 doi:10.1117/12.140112 oc44yBj0 -@tag:Gomezb2016_automatic arxiv:1610.02415 arxiv:1610.02415 2dU8f4XJ -@tag:Graphlab doi:10.14778/2212351.2212354 doi:10.14778/2212351.2212354 1XcexUAV -@tag:Guetterman url:http://www.fasebj.org/content/30/1_Supplement/406.3 url:http://www.fasebj.org/content/30/1_Supplement/406.3 W0cYSf89 -@tag:Gulshan2016_dt doi:10.1001/jama.2016.17216 doi:10.1001/jama.2016.17216 1mJW6umJ -@tag:Gultepe2014_sepsis doi:10.1136/amiajnl-2013-001815 doi:10.1136/amiajnl-2013-001815 eCrLGgiX -@tag:Gupta2015_exprs_yeast doi:10.1101/031906 doi:10.1101/031906 AnenJOuU -@tag:Gupta2015_prec arxiv:1502.02551 arxiv:1502.02551 CKcJuj03 -@tag:Hadjas2015_cc arxiv:1504.04343 arxiv:1504.04343 13KjSCKB2 -@tag:He2015_images arxiv:1512.03385 arxiv:1512.03385 j7KrVyi8 -@tag:Hinton2015_dark_knowledge arxiv:1503.02531 arxiv:1503.02531 1CRF3gAV -@tag:Hochreiter doi:10.1093/bioinformatics/btm247 doi:10.1093/bioinformatics/btm247 G8RKF6sz -@tag:Hoff doi:10.1093/nar/gkp327 doi:10.1093/nar/gkp327 q1A2AEtO -@tag:Horton1992_assessment doi:10.1093/nar/20.16.4331 doi:10.1093/nar/20.16.4331 uZvDdFZo -@tag:Hubara2016_qnn arxiv:1609.07061 arxiv:1609.07061 1GUizyE8e -@tag:Huddar2016_predicting doi:10.1109/ACCESS.2016.2618775 doi:10.1109/access.2016.2618775 fOaBA9Vc -@tag:Hughes2016_macromol_react doi:10.1021/acscentsci.6b00162 doi:10.1021/acscentsci.6b00162 1BARarxfz -@tag:Ithapu2015_efficient doi:10.1016/j.jalz.2015.01.010 doi:10.1016/j.jalz.2015.01.010 eehGXQlY -@tag:Jafari2016_skin_lesions arxiv:1609.02374 arxiv:1609.02374 1AJUcl1KV -@tag:Jha2017_integrative_models doi:10.1101/104869 doi:10.1101/104869 N0HBi8MH -@tag:Johnson2017_integ_cell arxiv:1705.00092 arxiv:1705.00092 71c6rs2z -@tag:JuanMateu2016_t1d doi:10.1530/EJE-15-0916 doi:10.1530/eje-15-0916 CNz9HwZ3 -@tag:Kahng2017_activis arxiv:1704.01942 arxiv:1704.01942 QphVo2P2 -@tag:Karlin doi:10.1128/jb.179.12.3899-3913.1997 doi:10.1128/jb.179.12.3899-3913.1997 N9NzkOjA -@tag:Karpathy2015_visualizing arxiv:1506.02078 arxiv:1506.02078 2cpYveR4 -@tag:Katzman2016_deepsurv arxiv:1606.00931 arxiv:1606.00931 1FE0F2pQ -@tag:Kearnes2016_admet arxiv:1606.08793 arxiv:1606.08793 uP7SgBVd -@tag:Kearnes2016_graph_conv doi:10.1007/s10822-016-9938-8 doi:10.1007/s10822-016-9938-8 145os4Y6t -@tag:Kelley2016_basset doi:10.1101/gr.200535.115 doi:10.1101/gr.200535.115 2CbHXoFn -@tag:Keras url:https://github.com/fchollet/keras url:https://github.com/fchollet/keras FwEK0msb -@tag:Kindermans2016_investigating arxiv:1611.07270 arxiv:1611.07270 b1sc0cgP -@tag:Knights doi:10.1111/j.1574-6976.2010.00251.x doi:10.1111/j.1574-6976.2010.00251.x aI9g2UOc -@tag:Koh2016_denoising doi:10.1101/052118 doi:10.1101/052118 XimuXZlz -@tag:Koh2017_understanding arxiv:1703.04730 arxiv:1703.04730 69wxD9y -@tag:Kooi2016_mamm_lesions doi:10.1016/j.media.2016.07.007 doi:10.1016/j.media.2016.07.007 18cZbigDD -@tag:Kraus2017_deeploc doi:10.15252/msb.20177551 doi:10.15252/msb.20177551 DcnNfASG -@tag:Krizhevsky2013_nips_cnn url:https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf url:https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf CCS5KSIM -@tag:Krizhevsky2014_weird_trick arxiv:1404.5997 arxiv:1404.5997 ZSVsnPVO -@tag:Lacey2016_dl_fpga arxiv:1602.04283 arxiv:1602.04283 9NKsJjSw -@tag:Lakhani2017_radiography doi:10.1148/radiol.2017162326 doi:10.1148/radiol.2017162326 Qve94Jra -@tag:Lanchantin2016_motif arxiv:1608.03644 arxiv:1608.03644 Dwi2eAvT -@tag:Lee2016_deeptarget arxiv:1603.09123v2 arxiv:1603.09123v2 1GwC1ll6h -@tag:Lee2016_emr_oct_amd doi:10.1101/094276 doi:10.1101/094276 SxsZyrVM -@tag:Lei2016_rationalizing arxiv:1606.04155 arxiv:1606.04155 ZUCVI5eU -@tag:Leibig2016_dr doi:10.1101/084210 doi:10.1101/084210 14Ovc5nPg -@tag:Li2014_minibatch doi:10.1145/2623330.2623612 doi:10.1145/2623330.2623612 fNkl8HFz -@tag:Li2016_variation doi:10.1126/science.aad9417 doi:10.1126/science.aad9417 b6p6wxpC -@tag:Liang2015_exprs_cancer doi:10.1109/TCBB.2014.2377729 doi:10.1109/tcbb.2014.2377729 1EtavGKI4 -@tag:Lipton2015_lstm arxiv:1510.07641 arxiv:1510.07641 glyI7H6F -@tag:Lipton2016_missing arxiv:1606.04130 arxiv:1606.04130 4zpZxjHR -@tag:Lisboa2006_review doi:10.1016/j.neunet.2005.10.007 doi:10.1016/j.neunet.2005.10.007 qxxwkSAT -@tag:Litjens2016_histopath_survey doi:10.1038/srep26286 doi:10.1038/srep26286 dQCjqq0Q -@tag:Litjens2017_medimage_survey arxiv:1702.05747 arxiv:1702.05747 LL5huVs3 -@tag:Liu doi:10.1371/journal.pone.0097958 doi:10.1371/journal.pone.0097958 Kt9NojjR -@tag:Liu2016_sc_transcriptome doi:10.12688/f1000research.7223.1 doi:10.12688/f1000research.7223.1 QafUwNKn -@tag:Liu2016_towards arxiv:1604.07043 arxiv:1604.07043 AEc66xxR -@tag:Lodato2015_neurons doi:10.1126/science.aab1785 doi:10.1126/science.aab1785 1GvfSy48x -@tag:Lundberg2016_an arxiv:1611.07478 arxiv:1611.07478 DeOI1oGf -@tag:Lusci2013_rnn doi:10.1021/ci400187y doi:10.1021/ci400187y 17Wih4Hd5 -@tag:Lusci2015_irv doi:10.1186/s13321-015-0110-6 doi:10.1186/s13321-015-0110-6 1E0x7QgLP -@tag:Ma2015_qsar_merck doi:10.1021/ci500747n doi:10.1021/ci500747n xOaTIeBY -@tag:Mahendran2014_understanding arxiv:1412.0035 arxiv:1412.0035 19mGl6pfy -@tag:Mahendran2016_salient doi:10.1007/978-3-319-46466-4_8 doi:10.1007/978-3-319-46466-4_8 vjXoJqO3 -@tag:Mahendran2016_visualizing doi:10.1007/s11263-016-0911-8 doi:10.1007/s11263-016-0911-8 1FkT6C6oa -@tag:Mapreduce doi:10.1145/1327452.1327492 doi:10.1145/1327452.1327492 xE3EYmck -@tag:Mayr2016_deep_tox doi:10.3389/fenvs.2015.00080 doi:10.3389/fenvs.2015.00080 Y1D0SZrO -@tag:McHardy doi:10.1038/nmeth976 doi:10.1038/nmeth976 QV551Nlx -@tag:Meng2016_mllib arxiv:1505.06807 arxiv:1505.06807 rZnxDitd -@tag:Metaphlan doi:10.1038/nmeth.2066 doi:10.1038/nmeth.2066 56wEWVIl -@tag:Min2016_deepenhancer doi:10.1109/BIBM.2016.7822593 doi:10.1109/bibm.2016.7822593 jV2YerUS -@tag:Mordvintsev2015_inceptionism url:http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html url:http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html XLHInhc1 -@tag:Moritz2015_sparknet arxiv:1511.06051 arxiv:1511.06051 rmJZ2Aui -@tag:Mrzelj url:https://repozitorij.uni-lj.si/IzpisGradiva.php?id=85515 url:https://repozitorij.uni-lj.si/IzpisGradiva.php?id=85515 iPIJrVVs -@tag:Murdoch2017_automatic arxiv:1702.02540 arxiv:1702.02540 10ViHstXn -@tag:Nemati2016_rl doi:10.1109/EMBC.2016.7591355 doi:10.1109/embc.2016.7591355 16OQvsRqJ -@tag:Ngiam2011 url:https://ai.stanford.edu/~ang/papers/icml11-MultimodalDeepLearning.pdf url:https://ai.stanford.edu/~ang/papers/icml11-MultimodalDeepLearning.pdf 1eN66lwn -@tag:Nguyen2014_adversarial arxiv:1412.1897v4 arxiv:1412.1897v4 1AkF8Wsv7 -@tag:Nie2016_3d_survival doi:10.1007/978-3-319-46723-8_25 doi:10.1007/978-3-319-46723-8_25 18EpaZ7QB -@tag:Nih_curiosity url:https://www.nigms.nih.gov/Education/Documents/curiosity.pdf url:https://www.nigms.nih.gov/Education/Documents/curiosity.pdf ru0hjGeQ -@tag:Olivecrona2017_drug_design arxiv:1704.07555 arxiv:1704.07555 1EayJRsI -@tag:Papernot2017_pate url:https://openreview.net/forum?id=HkwoSDPgg url:https://openreview.net/forum?id=HkwoSDPgg b8DJ1u6W -@tag:Park2016_deepmirgene arxiv:1605.00017 arxiv:1605.00017 1TeyWffV -@tag:Parnamaa2017 doi:10.1534/g3.116.033654 doi:10.1534/g3.116.033654 2a7MHtAx -@tag:Pawlowski2016 doi:10.1101/085118 doi:10.1101/085118 BMg062hc -@tag:Pereira2016_docking doi:10.1021/acs.jcim.6b00355 doi:10.1021/acs.jcim.6b00355 Gue0c5Gb -@tag:PerezSianes2016_screening doi:10.1007/978-3-319-40126-3_2 doi:10.1007/978-3-319-40126-3_2 1DTUK3YyI -@tag:Poplin2016_deepvariant doi:10.1101/092890 doi:10.1101/092890 FVfZESYP -@tag:Pratt2016_dr doi:10.1016/j.procs.2016.07.014 doi:10.1016/j.procs.2016.07.014 ayTsooEM -@tag:Qin2017_onehot doi:10.1371/journal.pcbi.1005403 doi:10.1371/journal.pcbi.1005403 Qbtqlmhf -@tag:Qiu2017_graph_embedding doi:10.1101/110668 doi:10.1101/110668 Oljj2W96 -@tag:RAD2010_view_cc doi:10.1145/1721654.1721672 doi:10.1145/1721654.1721672 ObFN78yp -@tag:Ragoza2016_protein arxiv:1612.02751 arxiv:1612.02751 bNBiIiTt -@tag:Raina2009_gpu doi:10.1145/1553374.1553486 doi:10.1145/1553374.1553486 F3e4wfzQ -@tag:Rajkomar2017_radiographs doi:10.1007/s10278-016-9914-9 doi:10.1007/s10278-016-9914-9 x6HXFAS4 -@tag:Ramsundar2015_multitask_drug arxiv:1502.02072 arxiv:1502.02072 yAoN5gTU -@tag:Ranganath2016_deep arxiv:1608.02158 arxiv:1608.02158 qXdO2aMm -@tag:Ribeiro2016_lime arxiv:1602.04938 arxiv:1602.04938 QwXSJhr0 -@tag:Rogers2010_fingerprints doi:10.1021/ci100050t doi:10.1021/ci100050t QnZ7V9Rd -@tag:Romero2017_diet url:https://openreview.net/pdf?id=Sk-oDY9ge url:https://openreview.net/pdf?id=Sk-oDY9ge 15JUKBg9y -@tag:Rosenberg2015_synthetic_seqs doi:10.1016/j.cell.2015.09.054 doi:10.1016/j.cell.2015.09.054 mlqKTlZY -@tag:Roth2015_view_agg_cad doi:10.1109/TMI.2015.2482920 doi:10.1109/tmi.2015.2482920 KseoWN2w -@tag:Russakovsky2015_imagenet doi:10.1007/s11263-015-0816-y doi:10.1007/s11263-015-0816-y cBVeXnZx -@tag:Sa2015_buckwild arxiv:1506.06438 arxiv:1506.06438 w6CoVmFK -@tag:Salzberg doi:10.1186/1471-2105-11-544 doi:10.1186/1471-2105-11-544 c4rnN1wo -@tag:Schatz2010_dna_cloud doi:10.1038/nbt0710-691 doi:10.1038/nbt0710-691 B6g0qKf4 -@tag:Schmidhuber2014_dnn_overview doi:10.1016/j.neunet.2014.09.003 doi:10.1016/j.neunet.2014.09.003 BQS8ClV0 -@tag:Scotti2016_missplicing doi:10.1038/nrg.2015.3 doi:10.1038/nrg.2015.3 QFK6GapR -@tag:Segata doi:10.1371/journal.pcbi.1004977 doi:10.1371/journal.pcbi.1004977 y9s5irW -@tag:Segler2017_drug_design arxiv:1701.01329 arxiv:1701.01329 8LWFFeYg -@tag:Seide2014_parallel doi:10.1109/ICASSP.2014.6853593 doi:10.1109/icassp.2014.6853593 IULiPa6L -@tag:Selvaraju2016_grad arxiv:1610.02391 arxiv:1610.02391 RZsNSRDS -@tag:Setty2015_seqgl doi:10.1371/journal.pcbi.1004271 doi:10.1371/journal.pcbi.1004271 138dgb9Ca -@tag:Shaham2016_batch_effects arxiv:1610.04181 arxiv:1610.04181 T2Md9xLY -@tag:Shapely doi:10.1515/9781400881970-018 doi:10.1515/9781400881970-018 YBJdA6LJ -@tag:Shen2017_medimg_review doi:10.1146/annurev-bioeng-071516-044442 doi:10.1146/annurev-bioeng-071516-044442 yEstnIOT -@tag:Shin2016_cad_tl doi:10.1109/TMI.2016.2528162 doi:10.1109/tmi.2016.2528162 1GAyqYBNZ -@tag:Shrikumar2017_learning arxiv:1704.02685 arxiv:1704.02685 zhmq9ktJ -@tag:Shrikumar2017_reversecomplement doi:10.1101/103663 doi:10.1101/103663 iEmvzeT8 -@tag:Silver2016_alphago doi:10.1038/nature16961 doi:10.1038/nature16961 2gn6PKkv -@tag:Simonyan2013_deep arxiv:1312.6034 arxiv:1312.6034 1YcKYTvO -@tag:Singh2016_deepchrome arxiv:1607.02078 arxiv:1607.02078 G10wkFHt -@tag:Sonderby arxiv:1503.01919 arxiv:1503.01919 81Cl5QSM -@tag:Soueidan doi:10.1515/metgen-2016-0001 doi:10.1515/metgen-2016-0001 yFOAeemA -@tag:Speech_recognition url:http://www.businessinsider.com/ibm-edges-closer-to-human-speech-recognition-2017-3 url:http://www.businessinsider.com/ibm-edges-closer-to-human-speech-recognition-2017-3 nyjAIan4 -@tag:Springenberg2014_striving arxiv:1412.6806 arxiv:1412.6806 f2L6isRj -@tag:Stein2010_cloud doi:10.1186/gb-2010-11-5-207 doi:10.1186/gb-2010-11-5-207 q0SsFrZd -@tag:Stormo2000_dna doi:10.1093/bioinformatics/16.1.16 doi:10.1093/bioinformatics/16.1.16 ywDQIvZJ -@tag:Stratnikov doi:10.1186/2049-2618-1-11 doi:10.1186/2049-2618-1-11 c5P9jHCg -@tag:Strobelt2016_visual arxiv:1606.07461 arxiv:1606.07461 1Ad3UOefc -@tag:Su2015_gpu arxiv:1507.01239 arxiv:1507.01239 aClNvbyM -@tag:Subramanian2016_bace1 doi:10.1021/acs.jcim.6b00290 doi:10.1021/acs.jcim.6b00290 B4cL1o2P -@tag:Sun2016_ensemble arxiv:1606.00575 arxiv:1606.00575 JUF9VoRD -@tag:Sundararajan2017_axiomatic arxiv:1703.01365 arxiv:1703.01365 WzFOJBiA -@tag:Sutskever arxiv:1409.3215 arxiv:1409.3215 2cMhMv5A -@tag:Swamidass2009_irv doi:10.1021/ci8004379 doi:10.1021/ci8004379 cjj5vT3H -@tag:TAC-ELM doi:10.1142/S0219720012500151 doi:10.1142/s0219720012500151 Wz7VUS03 -@tag:Tan2014_psb doi:10.1142/9789814644730_0014 doi:10.1142/9789814644730_0014 PBiRSdXv -@tag:Tan2015_adage doi:10.1128/mSystems.00025-15 doi:10.1128/msystems.00025-15 1CFhfCyWN -@tag:Tan2016_eadage doi:10.1101/078659 doi:10.1101/078659 zuLdSQx3 -@tag:TensorFlow arxiv:1603.04467 arxiv:1603.04467 hOeUlCvS -@tag:Torracinta2016_deep_snp doi:10.1101/097469 doi:10.1101/097469 GSLRw2L5 -@tag:Torracinta2016_sim doi:10.1101/079087 doi:10.1101/079087 ECTm1SuA -@tag:Tu1996_anns doi:10.1016/S0895-4356(96)00002-9 doi:10.1016/s0895-4356(96)00002-9 kL0B4m9d -@tag:Unterthiner2014_screening url:http://www.bioinf.at/publications/2014/NIPS2014a.pdf url:http://www.bioinf.at/publications/2014/NIPS2014a.pdf F8fP2vAg -@tag:Vanhoucke2011_cpu url:https://research.google.com/pubs/pub37631.html url:https://research.google.com/pubs/pub37631.html NSgduYNT -@tag:Vera2016_sc_analysis doi:10.1146/annurev-genet-120215-034854 doi:10.1146/annurev-genet-120215-034854 v97iPXDw -@tag:Vervier doi:10.1093/bioinformatics/btv683 doi:10.1093/bioinformatics/btv683 8DLzxOEt -@tag:Wallach2015_atom_net arxiv:1510.02855 arxiv:1510.02855 Z7fd0BYf -@tag:Wang2016_breast_cancer arxiv:1606.05718 arxiv:1606.05718 mbEp6jNr -@tag:Wang2016_protein_contact doi:10.1371/journal.pcbi.1005324 doi:10.1371/journal.pcbi.1005324 BhfjKSY3 -@tag:Wasson1985_clinical doi:10.1056/NEJM198509263131306 doi:10.1056/nejm198509263131306 xX68eyvs -@tag:Word2Vec arxiv:1301.3781 arxiv:1301.3781 1GhHIDxuW -@tag:Wu2017_molecule_net arxiv:1703.00564 arxiv:1703.00564 16OPHvAij -@tag:Xiong2011_bayesian doi:10.1093/bioinformatics/btr444 doi:10.1093/bioinformatics/btr444 8VPGUHcf -@tag:Xiong2015_splicing_code doi:10.1126/science.1254806 doi:10.1126/science.1254806 17sgPdcMT -@tag:Xu2015_show arxiv:1502.03044 arxiv:1502.03044 yHn4SDRI -@tag:Yasushi2016_cgbvs_dnn doi:10.1002/minf.201600045 doi:10.1002/minf.201600045 x0M6vals -@tag:Yoon2016_cancer_reports doi:10.1007/978-3-319-47898-2_21 doi:10.1007/978-3-319-47898-2_21 yUgE09ve -@tag:Yosinksi2015_understanding arxiv:1506.06579 arxiv:1506.06579 17i18PMkR -@tag:Yosinski2014 url:https://papers.nips.cc/paper/5347-how-transferable-are-features-in-deep-neural-networks url:https://papers.nips.cc/paper/5347-how-transferable-are-features-in-deep-neural-networks enhj7VT6 -@tag:Yu2016_melanoma_resnet doi:10.1109/TMI.2016.2642839 doi:10.1109/tmi.2016.2642839 phRCihNB -@tag:Zeiler2013_visualizing arxiv:1311.2901 arxiv:1311.2901 voh0OiT2 -@tag:Zeng2015 doi:10.1186/s12859-015-0553-9 doi:10.1186/s12859-015-0553-9 z3I2IudI -@tag:Zeng2016_convolutional doi:10.1093/bioinformatics/btw255 doi:10.1093/bioinformatics/btw255 182UhQqzp -@tag:Zhang2015_multitask_tl doi:10.1145/2783258.2783304 doi:10.1145/2783258.2783304 HlDY7trA -@tag:Zhou2015_deep_sea doi:10.1038/nmeth.3547 doi:10.1038/nmeth.3547 2UI1BZuD -@tag:Zhu2016_advers_mamm doi:10.1101/095786 doi:10.1101/095786 Xxb4t3zO -@tag:Zhu2016_mult_inst_mamm doi:10.1101/095794 doi:10.1101/095794 9G9Hv1Pp -@tag:Zintgraf2017_visualizing arxiv:1702.04595 arxiv:1702.04595 Kk20paR7 -@tag:lmat doi:10.1093/bioinformatics/btt389 doi:10.1093/bioinformatics/btt389 189TQrQA9 -@tag:nbc doi:10.1093/bioinformatics/btq619 doi:10.1093/bioinformatics/btq619 1HtJuEkb2 -@tag:wgsquikr doi:10.1371/journal.pone.0091784 doi:10.1371/journal.pone.0091784 RqhGD9c7 -@tag:yok doi:10.1186/1471-2105-12-20 doi:10.1186/1471-2105-12-20 qUGH5CX8 @url:http://ana.blogs.com/maestros/2006/11/data_is_the_new.html url:http://ana.blogs.com/maestros/2006/11/data_is_the_new.html url:http://ana.blogs.com/maestros/2006/11/data_is_the_new.html 6fE0Vrba @url:http://blog.dhimmel.com/irreproducible-timestamps/ url:http://blog.dhimmel.com/irreproducible-timestamps/ url:http://blog.dhimmel.com/irreproducible-timestamps/ QBWMEuxW +@tag:Mordvintsev2015_inceptionism url:http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html url:http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html XLHInhc1 @url:http://hazyresearch.github.io/snorkel/blog/weak_supervision.html url:http://hazyresearch.github.io/snorkel/blog/weak_supervision.html url:http://hazyresearch.github.io/snorkel/blog/weak_supervision.html hfcf5Hmi +@tag:Dragonn url:http://kundajelab.github.io/dragonn/ url:http://kundajelab.github.io/dragonn/ 117PEpTMe +@tag:Duvenaud2015_graph_conv url:http://papers.nips.cc/paper/5954-convolutional-networks-on-graphs-for-learning-molecular-fingerprints url:http://papers.nips.cc/paper/5954-convolutional-networks-on-graphs-for-learning-molecular-fingerprints Oe573FaL @url:http://proceedings.mlr.press/v54/mcmahan17a.html url:http://proceedings.mlr.press/v54/mcmahan17a.html url:http://proceedings.mlr.press/v54/mcmahan17a.html U0ySdznJ +@tag:Dean2012_nips_downpour url:http://research.google.com/archive/large_deep_networks_nips2012.html url:http://research.google.com/archive/large_deep_networks_nips2012.html 17cBimWgp @url:http://research.google.com/archive/unsupervised_icml2012.html url:http://research.google.com/archive/unsupervised_icml2012.html url:http://research.google.com/archive/unsupervised_icml2012.html IiNJE32f +@tag:Unterthiner2014_screening url:http://www.bioinf.at/publications/2014/NIPS2014a.pdf url:http://www.bioinf.at/publications/2014/NIPS2014a.pdf F8fP2vAg +@tag:Speech_recognition url:http://www.businessinsider.com/ibm-edges-closer-to-human-speech-recognition-2017-3 url:http://www.businessinsider.com/ibm-edges-closer-to-human-speech-recognition-2017-3 nyjAIan4 @url:http://www.cameo3d.org/ url:http://www.cameo3d.org/ url:http://www.cameo3d.org/ u9uApoaB @url:http://www.deeplearningbook.org/ url:http://www.deeplearningbook.org/ url:http://www.deeplearningbook.org/ yg8NW0K7 +@tag:Guetterman url:http://www.fasebj.org/content/30/1_Supplement/406.3 url:http://www.fasebj.org/content/30/1_Supplement/406.3 W0cYSf89 @url:http://www.intel.com/pressroom/archive/speeches/ag080998.htm url:http://www.intel.com/pressroom/archive/speeches/ag080998.htm url:http://www.intel.com/pressroom/archive/speeches/ag080998.htm mAXsmd43 +@tag:Ehran2009_visualizing url:http://www.iro.umontreal.ca/~lisa/publications2/index.php/publications/show/247 url:http://www.iro.umontreal.ca/~lisa/publications2/index.php/publications/show/247 UAAd9Uez +@tag:Bergstra2012_random url:http://www.jmlr.org/papers/v13/bergstra12a.html url:http://www.jmlr.org/papers/v13/bergstra12a.html 1FSwIjR9s +@tag:Coates2013_cots_hpc url:http://www.jmlr.org/proceedings/papers/v28/coates13.html url:http://www.jmlr.org/proceedings/papers/v28/coates13.html 4MZ2tmZ8 @url:http://www.predictioncenter.org/casp12/rrc_avrg_results.cgi url:http://www.predictioncenter.org/casp12/rrc_avrg_results.cgi url:http://www.predictioncenter.org/casp12/rrc_avrg_results.cgi zScWGveU +@tag:Ngiam2011 url:https://ai.stanford.edu/~ang/papers/icml11-MultimodalDeepLearning.pdf url:https://ai.stanford.edu/~ang/papers/icml11-MultimodalDeepLearning.pdf 1eN66lwn @url:https://eprint.iacr.org/2017/281.pdf url:https://eprint.iacr.org/2017/281.pdf url:https://eprint.iacr.org/2017/281.pdf 1GprsH3DV +@tag:DeepChem url:https://github.com/deepchem/deepchem url:https://github.com/deepchem/deepchem Ytvk62dX +@tag:Keras url:https://github.com/fchollet/keras url:https://github.com/fchollet/keras FwEK0msb @url:https://github.com/greenelab/deep-review url:https://github.com/greenelab/deep-review url:https://github.com/greenelab/deep-review yLr4pV4G +@tag:Elephas url:https://github.com/maxpumperla/elephas url:https://github.com/maxpumperla/elephas y9IoEy4r @url:https://medium.com/twenty-one-hundred/data-is-the-new-oil-a-ludicrous-proposition-1d91bba4f294 url:https://medium.com/twenty-one-hundred/data-is-the-new-oil-a-ludicrous-proposition-1d91bba4f294 url:https://medium.com/twenty-one-hundred/data-is-the-new-oil-a-ludicrous-proposition-1d91bba4f294 o8mib4CN @url:https://nihcc.app.box.com/v/ChestXray-NIHCC url:https://nihcc.app.box.com/v/ChestXray-NIHCC url:https://nihcc.app.box.com/v/ChestXray-NIHCC odFR7ptt +@tag:Papernot2017_pate url:https://openreview.net/forum?id=HkwoSDPgg url:https://openreview.net/forum?id=HkwoSDPgg b8DJ1u6W +@tag:Romero2017_diet url:https://openreview.net/pdf?id=Sk-oDY9ge url:https://openreview.net/pdf?id=Sk-oDY9ge 15JUKBg9y @url:https://opentimestamps.org/ url:https://opentimestamps.org/ url:https://opentimestamps.org/ igA8jklc +@tag:Bergstra2011_hyper url:https://papers.nips.cc/paper/4443-algorithms-for-hyper-parameter-optimization.pdf url:https://papers.nips.cc/paper/4443-algorithms-for-hyper-parameter-optimization.pdf wz83yfHF +@tag:Krizhevsky2013_nips_cnn url:https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf url:https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf CCS5KSIM +@tag:Yosinski2014 url:https://papers.nips.cc/paper/5347-how-transferable-are-features-in-deep-neural-networks url:https://papers.nips.cc/paper/5347-how-transferable-are-features-in-deep-neural-networks enhj7VT6 @url:https://phekb.org/implementations url:https://phekb.org/implementations url:https://phekb.org/implementations ziudr6hx +@tag:Mrzelj url:https://repozitorij.uni-lj.si/IzpisGradiva.php?id=85515 url:https://repozitorij.uni-lj.si/IzpisGradiva.php?id=85515 iPIJrVVs +@tag:Vanhoucke2011_cpu url:https://research.google.com/pubs/pub37631.html url:https://research.google.com/pubs/pub37631.html NSgduYNT @url:https://www.bgcarlisle.com/blog/2014/08/25/proof-of-prespecified-endpoints-in-medical-research-with-the-bitcoin-blockchain/ url:https://www.bgcarlisle.com/blog/2014/08/25/proof-of-prespecified-endpoints-in-medical-research-with-the-bitcoin-blockchain/ url:https://www.bgcarlisle.com/blog/2014/08/25/proof-of-prespecified-endpoints-in-medical-research-with-the-bitcoin-blockchain/ 6MR50hyY +@tag:Nih_curiosity url:https://www.nigms.nih.gov/Education/Documents/curiosity.pdf url:https://www.nigms.nih.gov/Education/Documents/curiosity.pdf ru0hjGeQ +@tag:Dream_tf_binding url:https://www.synapse.org/#!Synapse:syn6131484/wiki/402026 url:https://www.synapse.org/#!Synapse:syn6131484/wiki/402026 wW6QbBXz diff --git a/manuscript.html b/manuscript.html new file mode 100644 index 00000000..cb6a5e13 --- /dev/null +++ b/manuscript.html @@ -0,0 +1,2510 @@ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + Opportunities and obstacles for deep learning in biology and medicine + + + + + + + + +
+

Opportunities and obstacles for deep learning in biology and medicine

+
+

A DOI-citable preprint of this manuscript is available at https://doi.org/10.1101/142760.

+

This manuscript was automatically generated from greenelab/deep-review@8eb858a on November 3, 2017.

+

Authors

+ +

Abstract

+

Deep learning, which describes a class of machine learning algorithms, has recently showed impressive results across a variety of domains. Biology and medicine are data rich, but the data are complex and often ill-understood. Problems of this nature may be particularly well-suited to deep learning techniques. We examine applications of deep learning to a variety of biomedical problems – patient classification, fundamental biological processes, and treatment of patients – and discuss whether deep learning will transform these tasks or if the biomedical sphere poses unique challenges. We find that deep learning has yet to revolutionize or definitively resolve any of these problems, but promising advances have been made on the prior state of the art. Even when improvement over a previous baseline has been modest, we have seen signs that deep learning methods may speed or aid human investigation. More work is needed to address concerns related to interpretability and how to best model each problem. Furthermore, the limited amount of labeled data for training presents problems in some domains, as do legal and privacy constraints on work with sensitive health records. Nonetheless, we foresee deep learning powering changes at both bench and bedside with the potential to transform several areas of biology and medicine.

+

Introduction to deep learning

+

Biology and medicine are rapidly becoming data-intensive. A recent comparison of genomics with social media, online videos, and other data-intensive disciplines suggests that genomics alone will equal or surpass other fields in data generation and analysis within the next decade [1]. The volume and complexity of these data present new opportunities, but also pose new challenges. Automated algorithms that extract meaningful patterns could lead to actionable knowledge and change how we develop treatments, categorize patients, or study diseases, all within privacy-critical environments.

+

The term deep learning has come to refer to a collection of new techniques that, together, have demonstrated breakthrough gains over existing best-in-class machine learning algorithms across several fields. Over the past five years these methods have revolutionized image classification and speech recognition due to their flexibility and high accuracy [2]. More recently, deep learning algorithms have shown promise in fields as diverse as high-energy physics [3], dermatology [4], and translation among written languages [5]. Across fields, “off-the-shelf” implementations of these algorithms have produced comparable or higher accuracy than previous best-in-class methods that required years of extensive customization, and specialized implementations are now being used at industrial scales.

+

Neural networks were first proposed in 1943 [6] as a model for how our brains process information. The history of neural networks is interesting in its own right [7]. In neural networks, inputs are fed into a hidden layer, which feeds into one or more subsequent hidden layers, which eventually produce an output layer. The neural networks used for deep learning have multiple hidden layers. Each layer essentially performs feature construction for the layers before it. The training process used often allows layers deeper in the network to contribute to the refinement of earlier layers. For this reason, these algorithms can automatically engineer features that are suitable for many tasks and customize those features for one or more specific tasks. Deep learning does many of the same things as more familiar machine learning approaches [8]. Like a clustering algorithm, it can build features that describe recurrent patterns in data. Like a regression approach, deep learning methods can predict some output. However, deep learning methods combine both of these steps. When sufficient data are available, these methods construct features tuned to a specific problem and combine those features into a predictor. Recently, hardware improvements and very large training datasets have allowed these deep learning techniques to surpass other machine learning algorithms for many problems.

+

Neural networks are most widely associated with supervised machine learning, where the goal is to accurately predict one or more labels associated with each data point. However, deep learning algorithms can also be used in an exploratory, “unsupervised” mode, where the goal is to summarize, explain, or identify interesting patterns in a data set. In a famous early example, scientists from Google demonstrated that a neural network “discovered” that cats, faces, and pedestrians were important components of online videos [9] without being told to look for them. What if, more generally, deep learning could solve the challenges presented by the growth of data in biomedicine? Could these algorithms identify the “cats” hidden in our data – the patterns unknown to the researcher – and suggest ways to act on them? In this review, we examine deep learning’s application to biomedical science and discuss the unique challenges that biomedical data pose for deep learning methods.

+

Several important advances make the current surge of work done in this area possible. Easy-to-use software packages have brought the techniques of the field out of the specialist’s toolkit to a broad community of computational scientists. Additionally, new techniques for fast training have enabled their application to larger datasets [10]. Dropout of nodes, edges, and layers makes networks more robust, even when the number of parameters is very large. New neural network approaches are also well-suited for addressing distinct challenges. For example, neural networks structured as autoencoders or as adversarial networks require no labels and are now regularly used for unsupervised tasks. In this review, we do not exhaustively discuss the different types of deep neural network architectures. A recent book from Goodfellow et al. [11] covers these in detail. Finally, the larger datasets now available are also sufficient for fitting the many parameters that exist for deep neural networks. The convergence of these factors currently makes deep learning extremely adaptable and capable of addressing the nuanced differences of each domain to which it is applied.

+

Will deep learning transform the study of human disease?

+

With this review, we ask the question: what is needed for deep learning to transform how we categorize, study, and treat individuals to maintain or restore health? We choose a high bar for “transform.” Andrew Grove, the former CEO of Intel, coined the term Strategic Inflection Point to refer to a change in technologies or environment that requires a business to be fundamentally reshaped [12]. Here, we seek to identify whether deep learning is an innovation that can induce a Strategic Inflection Point in the practice of biology or medicine.

+

There are already a number of reviews focused on applications of deep learning in biology [13–17], healthcare [18,19], and drug discovery [20–23]. Under our guiding question, we sought to highlight cases where deep learning enabled researchers to solve challenges that were previously considered infeasible or simplified tedious analyses. We also identified approaches that researchers are using to address challenges posed by biomedical data. We find that domain-specific considerations have greatly influenced how to best harness the power and flexibility of deep learning. Model interpretability is often critical. Understanding the patterns in data may be just as important as fitting the data. In addition, there are important and pressing questions about how to build networks that efficiently represent the underlying structure and logic of the data. Domain experts can play important roles in designing networks to represent data appropriately, encoding the most salient prior knowledge and assessing success or failure. There is also great potential to create deep learning systems that augment biologists and clinicians by prioritizing experiments or streamlining tasks that do not require expert judgment. We have divided the large range of topics into three broad classes: (1) disease and patient categorization, (2) fundamental biological study, and (3) treatment of patients. Below, we briefly introduce the types of questions, approaches and data that are typical for each class in the application of deep learning.

+

Disease and patient categorization

+

A key challenge in biomedicine is the accurate classification of diseases and disease subtypes. In oncology, current “gold standard” approaches include histology, which requires interpretation by experts, or assessment of molecular markers such as cell surface receptors or gene expression. One example is the PAM50 approach to classifying breast cancer where the expression of 50 marker genes divides breast cancer patients into four subtypes. Substantial heterogeneity still remains within these four subtypes [24,25]. Given the increasing wealth of molecular data available, a more comprehensive subtyping seems possible. Several studies have used deep learning methods to better categorize breast cancer patients. Denoising autoencoders, an unsupervised approach, can be used to cluster breast cancer patients [26], and convolutional neural networks (CNNs) can help count mitotic divisions, a feature that is highly correlated with disease outcome in histological images [27]. Despite these recent advances, a number of challenges exist in this area of research, most notably the integration of molecular and imaging data with other disparate types of data such as electronic health records (EHRs).

+

Fundamental biological study

+

Deep learning can be applied to answer more fundamental biological questions; it is especially suited to leveraging large amounts of data from high-throughput “omics” studies. One classic biological problem where machine learning, and now deep learning, has been extensively applied is molecular target prediction. For example, deep recurrent neural networks (RNNs) have been used to predict gene targets of microRNAs [28], and CNNs have been applied to predict protein residue-residue contacts and secondary structure [29–31]. Other recent exciting applications of deep learning include recognition of functional genomic elements such as enhancers and promoters [32–34] and prediction of the deleterious effects of nucleotide polymorphisms [35].

+

Treatment of patients

+

Although the application of deep learning to patient treatment is just beginning, we expect new methods to recommend patient treatments, predict treatment outcomes, and guide the development of new therapies. One type of effort in this area aims to identify drug targets and interactions or predict drug response. Another uses deep learning on protein structures to predict drug interactions and drug bioactivity [36]. Drug repositioning using deep learning on transcriptomic data is another exciting area of research [37]. Restricted Boltzmann machines (RBMs) can be combined into deep belief networks (DBNs) to predict novel drug-target interactions and formulate drug repositioning hypotheses [38,39]. Finally, deep learning is also prioritizing chemicals in the early stages of drug discovery for new targets [23].

+

Deep learning and patient categorization

+

In healthcare, individuals are diagnosed with a disease or condition based on symptoms, the results of certain diagnostic tests, or other factors. Once diagnosed with a disease, an individual might be assigned a stage based on another set of human-defined rules. While these rules are refined over time, the process is evolutionary and ad hoc, potentially impeding the identification of underlying biological mechanisms and their corresponding treatment interventions.

+

Deep learning methods coupled with a large corpus of patient phenotypes may provide a more data-driven approach to patient categorization. A deep neural network has the potential to identify entirely new categories of health or disease that are only present when data from multiple lab tests are integrated.

+

As an example, consider the condition Latent Autoimmune Diabetes in Adults (LADA; reviewed in [40]). In the absence of a pre-specified disease definition, a deep neural network might have identified a subgroup of individuals with blood glucose levels that indicated diabetes as well as auto-antibodies, even though the individuals had never been diagnosed with type 1 diabetes – the autoimmune form of the disease that arises in young people. Such a neural network would be identifying patients with LADA. As no such computational approach existed, LADA was actually identified by Groop et al. [41].

+

One should not regard recapitulation of existing disease categories as a gold-standard for deep learning results. Instead, a meaningful contribution to patient categorization would be to identify new shared mechanisms that would otherwise be obscured due to ad hoc historical definitions of disease. Perhaps deep neural networks, by reevaluating data without the context of our assumptions, can reveal novel classes of treatable conditions.

+

In spite of such optimism, the ability of deep learning models to indiscriminately extract predictive signals must also be assessed and operationalized with care. Imagine a deep neural network is provided with clinical test results gleaned from electronic health records. Because physicians may order certain tests based on their suspected diagnosis, a deep neural network may learn to “diagnose” patients simply based on the tests that are ordered. For some objective functions, such as predicting an International Classification of Diseases (ICD) code, this may offer good performance even though it does not provide insight into the underlying disease beyond physician activity. This challenge is not unique to deep learning approaches; however, it is important for practitioners to be aware of these challenges and the possibility in this domain of constructing highly predictive classifiers of questionable actual utility.

+

Our goal in this section is to assess the extent to which deep learning is already contributing to the discovery of novel categories. Where it is not, we focus on barriers to achieving these goals. We also highlight approaches that researchers are taking to address challenges within the field, particularly with regards to data availability and labeling.

+

Imaging applications in healthcare

+

Deep learning methods have transformed the analysis of natural images and video, and similar examples are beginning to emerge with medical images. Deep learning has been used to classify lesions and nodules; localize organs, regions, landmarks and lesions; segment organs, organ substructures and lesions; retrieve images based on content; generate and enhance images; and combine images with clinical reports [19,42].

+

Though there are many commonalities with the analysis of natural images, there are also key differences. In all cases that we examined, fewer than one million images were available for training, and datasets are often many orders of magnitude smaller than collections of natural images. Researchers have developed subtask-specific strategies to address this challenge.

+

The first strategy repurposes features extracted from natural images by deep learning models, such as ImageNet [43], for new purposes. Diagnosing diabetic retinopathy through color fundus images became an area of focus for deep learning researchers after a large labeled image set was made publicly available during a 2015 Kaggle competition [44]. Most participants trained neural networks from scratch [44–46], but Gulshan et al. [47] repurposed a 48-layer Inception-v3 deep architecture pre-trained on natural images and surpassed the state-of-the-art specificity and sensitivity. Such features were also repurposed to detect melanoma, the deadliest form of skin cancer, from dermoscopic [48,49] and non-dermoscopic images of skin lesions [4,50,51] as well as age-related macular degeneration [52]. Pre-training on natural images can enable very deep networks to succeed without overfitting. For the melanoma task, reported performance was competitive with or better than a board of certified dermatologists [4,48].

+

Reusing features from natural images is also an emerging approach for radiographic images, where datasets are often too small to train large deep neural networks without these techniques [53–56]. A deep CNN trained on natural images boosts performance in radiographic images [55]. However, the target task required either re-training the initial model from scratch with special pre-processing or fine-tuning of the whole network on radiographs with heavy data augmentation to avoid overfitting.

+

The technique of reusing features from a different task falls into the broader area of transfer learning (see Discussion). Though we’ve mentioned numerous successes for the transfer of natural image features to new tasks, we expect that a lower proportion of negative results have been published. The analysis of magnetic resonance images (MRIs) is also faced with the challenge of small training sets. In this domain, Amit et al. [57] investigated the tradeoff between pre-trained models from a different domain and a small CNN trained only with MRI images. In contrast with the other selected literature, they found a smaller network trained with data augmentation on few hundred images from a few dozen patients can outperform a pre-trained out-of-domain classifier. Data augmentation is a different strategy to deal with small training sets. The practice is exemplified by a series of papers that analyze images from mammographies [58–62]. To expand the number and diversity of images, researchers constructed adversarial examples [61]. Adversarial examples are constructed by applying a transformation that changes training images but not their content – for example by rotating an image by a random amount. An alternative in the domain is to train towards human-created features before subsequent fine-tuning [59], which can help to sidestep this challenge though it does give up deep learning techniques’ strength as feature constructors.

+

Another way of dealing with limited training data is to divide rich data – e.g. 3D images – into numerous reduced projections. Shin et al. [54] compared various deep network architectures, dataset characteristics, and training procedures for computer tomography-based (CT) abnormality detection. They concluded that networks as deep as 22 layers could be useful for 3D data, despite the limited size of training datasets. However, they noted that choice of architecture, parameter setting, and model fine-tuning needed is very problem- and dataset-specific. Moreover, this type of task often depends on both lesion localization and appearance, which poses challenges for CNN-based approaches. Straightforward attempts to capture useful information from full-size images in all three dimensions simultaneously via standard neural network architectures were computationally unfeasible. Instead, two-dimensional models were used to either process image slices individually (2D), or aggregate information from a number of 2D projections in the native space (2.5D).

+

Roth et al. compared 2D, 2.5D, and 3D CNNs on a number of tasks for computer-aided detection from CT scans and showed that 2.5D CNNs performed comparably well to 3D analogs, while requiring much less training time, especially on augmented training sets [63]. Another advantage of 2D and 2.5D networks is the wider availability of pre-trained models. But reducing the dimensionality is not always helpful. Nie et al. [64] showed that multimodal, multi-channel 3D deep architecture was successful at learning high-level brain tumor appearance features jointly from MRI, functional MRI, and diffusion MRI images, outperforming single-modality or 2D models. Overall, the variety of modalities, properties and sizes of training sets, the dimensionality of input, and the importance of end goals in medical image analysis are provoking a development of specialized deep neural network architectures, training and validation protocols, and input representations that are not characteristic of widely-studied natural images.

+

Predictions from deep neural networks can be evaluated for use in workflows that also incorporate human experts. In a large dataset of mammography images, Kooi et al. [65] demonstrated that deep neural networks outperform the traditional computer-aided diagnosis system at low sensitivity and perform comparably at high sensitivity. They also compared network performance to certified screening radiologists on a patch level and found no significant difference between the network and the readers. However, using deep methods for clinical practice is challenged by the difficulty of assigning a level of confidence to each prediction. Leibig et al. [46] estimated the uncertainty of deep networks for diabetic retinopathy diagnosis by linking dropout networks with approximate Bayesian inference. Techniques that assign confidences to each prediction should aid pathologist-computer interactions and improve uptake by physicians.

+

Systems to aid in the analysis of histology slides are also promising use cases for deep learning [66]. Ciresan et al. [27] developed one of the earliest approaches for histology slides, winning the 2012 International Conference on Pattern Recognition’s Contest on Mitosis Detection while achieving human-competitive accuracy. In more recent work, Wang et al. [67] analyzed stained slides of lymph node slices to identify cancers. On this task a pathologist has about a 3% error rate. The pathologist did not produce any false positives, but did have a number of false negatives. The algorithm had about twice the error rate of a pathologist, but the errors were not strongly correlated. In this area, these algorithms may be ready to be incorporated into existing tools to aid pathologists and reduce the false negative rate. Ensembles of deep learning and human experts may help overcome some of the challenges presented by data limitations.

+

One source of training examples with rich clinical annotations is electronic health records. Recently, Lee et al. [68] developed an approach to distinguish individuals with age-related macular degeneration from control individuals. They trained a deep neural network on approximately 100,000 images extracted from structured electronic health records, reaching greater than 93% accuracy. The authors used their test set to evaluate when to stop training. In other domains, this has resulted in a minimal change in the estimated accuracy [69], but we recommend the use of an independent test set whenever feasible.

+

Chest X-rays are a common radiological examination for screening and diagnosis of lung diseases. Although hospitals have accumulated a large number of raw radiology images and reports in Picture Archiving and Communication Systems and their related reports in Radiology Information Systems, it is not yet known how to effectively use them to learn the correlation between pathology categories and X-rays. In the last few years, deep learning methods showed remarkable results in chest X-ray image analysis [70,71]. However, it is both costly and time-consuming to annotate a large-scale fully-labeled corpus to facilitate data-intensive deep learning models. As an alternative, Wang et al. [71] proposed to use weakly labeled images. To generate weak labels for X-ray images, they applied a series of natural language processing (NLP) techniques to the associated chest X-ray radiological reports. Specifically, they first extracted all diseases mentioned in the reports using a state-of-the-art NLP tool, then applied a newly-developed negation and uncertainty detection tool (NegBio) to filter negative and equivocal findings in the reports. Evaluation on three independent datasets demonstrated that NegBio is highly accurate for detecting negative and equivocal findings (~90% in F-measure, which balances precision and recall [72]). These highly-accurate results meet the need to generate a corpus with weak labels, which serves as a solid foundation for the later process of image classification. The resulting dataset [73] consists of 112,120 frontal-view chest X-ray images from 30,805 patients, and each image is associated with one or more weakly-labeled pathology category (e.g. pneumonia and cardiomegaly) or “normal” otherwise. Further, Wang et al. [71] used this dataset with a unified weakly-supervised multi-label image classification framework, to detect common thoracic diseases. It showed superior performance over a benchmark using fully-labeled data.

+

With the exception of natural image-like problems (e.g. melanoma detection), biomedical imaging poses a number of challenges for deep learning. Datasets are typically small, annotations can be sparse, and images are often high-dimensional, multimodal, and multi-channel. Techniques like transfer learning, heavy dataset augmentation, and the use of multi-view and multi-stream architectures are more common than in the natural image domain. Furthermore, high model sensitivity and specificity can translate directly into clinical value. Thus, prediction evaluation, uncertainty estimation, and model interpretation methods are also of great importance in this domain (see Discussion). Finally, there is a need for better pathologist-computer interaction techniques that will allow combining the power of deep learning methods with human expertise and lead to better-informed decisions for patient treatment and care.

+

Electronic health records

+

EHR data include substantial amounts of free text, which remains challenging to approach [74]. Often, researchers developing algorithms that perform well on specific tasks must design and implement domain-specific features [75]. These features capture unique aspects of the literature being processed. Deep learning methods are natural feature constructors. In recent work, the authors evaluated the extent to which deep learning methods could be applied on top of generic features for domain-specific concept extraction [76]. They found that performance was in line with, but lower than the best domain-specific method [76]. This raises the possibility that deep learning may impact the field by reducing the researcher time and cost required to develop specific solutions, but it may not always lead to performance increases.

+

In recent work, Yoon et al.[77] analyzed simple features using deep neural networks and found that the patterns recognized by the algorithms could be re-used across tasks. Their aim was to analyze the free text portions of pathology reports to identify the primary site and laterality of tumors. The only features the authors supplied to the algorithms were unigrams (counts for single words) and bigrams (counts for two-word combinations) in a free text document. They subset the full set of words and word combinations to the 400 most common. The machine learning algorithms that they employed (naïve Bayes, logistic regression, and deep neural networks) all performed relatively similarly on the task of identifying the primary site. However, when the authors evaluated the more challenging task, evaluating the laterality of each tumor, the deep neural network outperformed the other methods. Of particular interest, when the authors first trained a neural network to predict primary site and then repurposed those features as a component of a secondary neural network trained to predict laterality, the performance was higher than a laterality-trained neural network. This demonstrates how deep learning methods can repurpose features across tasks, improving overall predictions as the field tackles new challenges. The Discussion further reviews this type of transfer learning.

+

Several authors have created reusable feature sets for medical terminologies using natural language processing and neural embedding models, as popularized by Word2vec [78]. A goal of learning terminologies for different entities in the same vector space is to find relationships between different domains (e.g. drugs and the diseases they treat). It is difficult for us to provide a strong statement on the broad utility of these methods. Manuscripts in this area tend to compare algorithms applied to the same data but lack a comparison against overall best-practices for one or more tasks addressed by these methods. Techniques have been developed for free text medical notes [79], ICD and National Drug Codes, and claims data [80]. Methods for neural embeddings learned from electronic health records have at least some ability to predict disease-disease associations and implicate genes with a statistical association with a disease [81]. However, the evaluations performed did not differentiate between simple predictions (i.e. the same disease in different sites of the body) and non-intuitive ones. While promising, a lack of rigorous evaluations of the real-world utility of these kinds of features makes current contributions in this area difficult to evaluate. To examine the true utility, comparisons need to be performed against leading approaches (i.e. algorithms and data) as opposed to simply evaluating multiple algorithms on the same potentially limited dataset.

+

Identifying consistent subgroups of individuals and individual health trajectories from clinical tests is also an active area of research. Approaches inspired by deep learning have been used for both unsupervised feature construction and supervised prediction. Early work by Lasko et al. [82], combined sparse autoencoders and Gaussian processes to distinguish gout from leukemia from uric acid sequences. Later work showed that unsupervised feature construction of many features via denoising autoencoder neural networks could dramatically reduce the number of labeled examples required for subsequent supervised analyses [83]. In addition, it pointed towards learned features being useful for subtyping within a single disease. In a concurrent large-scale analysis of EHR data from 700,000 patients, Miotto et al. [84] used a deep denoising autoencoder architecture applied to the number and co-occurrence of clinical events (DeepPatient) to learn a representation of patients. The model was able to predict disease trajectories within one year with over 90% accuracy and patient-level predictions were improved by up to 15% when compared to other methods. Razavian et al. [85] used a set of 18 common lab tests to predict disease onset using both CNN and long short-term memory (LSTM) architectures and demonstrated an improvement over baseline regression models. However, numerous challenges including data integration (patient demographics, family history, laboratory tests, text-based patient records, image analysis, genomic data) and better handling of streaming temporal data with many features, will need to be overcome before we can fully assess the potential of deep learning for this application area.

+

Still, recent work has also revealed domains in which deep networks have proven superior to traditional methods. Survival analysis models the time leading to an event of interest from a shared starting point, and in the context of EHR data, often associates these events to subject covariates. Exploring this relationship is difficult, however, given that EHR data types are often heterogeneous, covariates are often missing, and conventional approaches require the covariate-event relationship be linear and aligned to a specific starting point [86]. Early approaches, such as the Faraggi-Simon feed-forward network, aimed to relax the linearity assumption, but performance gains were lacking [87]. Katzman et al. in turn developed a deep implementation of the Faraggi-Simon network that, in addition to outperforming Cox regression, was capable of comparing the risk between a given pair of treatments, thus potentially acting as recommender system [88]. To overcome the remaining difficulties, researchers have turned to deep exponential families, a class of latent generative models that are constructed from any type of exponential family distributions [89]. The result was a deep survival analysis model capable of overcoming challenges posed by missing data and heterogeneous data types, while uncovering nonlinear relationships between covariates and failure time. They showed their model more accurately stratified patients as a function of disease risk score compared to the current clinical implementation.

+

There is a computational cost for these methods, however, when compared to traditional, non-neural network approaches. For the exponential family models, despite their scalability [90], an important question for the investigator is whether he or she is interested in estimates of posterior uncertainty. Given that these models are effectively Bayesian neural networks, much of their utility simplifies to whether a Bayesian approach is warranted for a given increase in computational cost. Moreover, as with all variational methods, future work must continue to explore just how well the posterior distributions are approximated, especially as model complexity increases [91].

+

Challenges and opportunities in patient categorization

+

Generating ground-truth labels can be expensive or impossible

+

A dearth of true labels is perhaps among the biggest obstacles for EHR-based analyses that employ machine learning. Popular deep learning (and other machine learning) methods are often used to tackle classification tasks and thus require ground-truth labels for training. For EHRs this can mean that researchers must hire multiple clinicians to manually read and annotate individual patients’ records through a process called chart review. This allows researchers to assign “true” labels, i.e. those that match our best available knowledge. Depending on the application, sometimes the features constructed by algorithms also need to be manually validated and interpreted by clinicians. This can be time consuming and expensive [92]. Because of these costs, much of this research, including the work cited in this review, skips the process of expert review. Clinicians’ skepticism for research without expert review may greatly dampen their enthusiasm for the work and consequently reduce its impact. To date, even well-resourced large national consortia have been challenged by the task of acquiring enough expert-validated labeled data. For instance, in the eMERGE consortia and PheKB database [93], most samples with expert validation contain only 100 to 300 patients. These datasets are quite small even for simple machine learning algorithms. The challenge is greater for deep learning models with many parameters. While unsupervised and semi-supervised approaches can help with small sample sizes, the field would benefit greatly from large collections of anonymized records in which a substantial number of records have undergone expert review. This challenge is not unique to EHR-based studies. Work on medical images, omics data in applications for which detailed metadata are required, and other applications for which labels are costly to obtain will be hampered as long as abundant curated data are unavailable.

+

Successful approaches to date in this domain have sidestepped this challenge by making methodological choices that either reduce the need for labeled examples or that use transformations to training data to increase the number of times it can be used before overfitting occurs. For example, the unsupervised and semi-supervised methods that we have discussed reduce the need for labeled examples [83]. The anchor and learn framework [94] uses expert knowledge to identify high-confidence observations from which labels can be inferred. The adversarial training example strategies mentioned above can reduce overfitting, if transformations are available that preserve the meaningful content of the data while transforming irrelevant features [61]. While adversarial training examples can be easily imagined for certain methods that operate on images, it is more challenging to figure out what an equivalent transformation would be for a patient’s clinical test results. Consequently, it may be hard to employ adversarial training examples, not to be confused with generative adversarial neural networks, with other applications. Finally, approaches that transfer features can also help use valuable training data most efficiently. Rajkomar et al. trained a deep neural network using generic images before tuning using only radiology images [55]. Datasets that require many of the same types of features might be used for initial training, before fine tuning takes place with the more sparse biomedical examples. Though the analysis has not yet been attempted, it is possible that analogous strategies may be possible with electronic health records. For example, features learned from the electronic health record for one type of clinical test (e.g. a decrease over time in a lab value) may transfer across phenotypes.

+

Methods to accomplish more with little high-quality labeled data are also being applied in other domains and may also be adapted to this challenge, e.g. data programming [95]. In data programming, noisy automated labeling functions are integrated. Numerous writers have described data as the new oil [96,97]. The idea behind this metaphor is that data are available in large quantities, valuable once refined, and the underlying resource that will enable a data-driven revolution in how work is done. Contrasting with this perspective, Ratner, Bach, and Ré described labeled training data as “The New New Oil” [98]. In this framing, data are abundant and not a scarce resource. Instead, new approaches to solving problems arise when labeled training data become sufficient to enable them. Based on our review of research on deep learning methods to categorize disease, the latter framing rings true.

+

We expect improved methods for domains with limited data to play an important role if deep learning is going to transform how we categorize states of human health. We don’t expect that deep learning methods will replace expert review. We expect them to complement expert review by allowing more efficient use of the costly practice of manual annotation.

+

Data sharing is hampered by standardization and privacy considerations

+

To construct the types of very large datasets that deep learning methods thrive on, we need robust sharing of large collections of data. This is in part a cultural challenge. We touch on this challenge in Discussion. Beyond the cultural hurdles around data sharing, there are also technological and legal hurdles related to sharing individual health records or deep models built from such records. This subsection deals primarily with these challenges.

+

EHRs are designed chiefly for clinical, administrative and financial purposes, such as patient care, insurance and billing [99]. Science is at best a tertiary priority, presenting challenges to EHR-based research in general and to deep learning research in particular. Although there is significant work in the literature around EHR data quality and the impact on research [100], we focus on three types of challenges: local bias, wider standards, and legal issues. Note these problems are not restricted to EHRs but can also apply to any large biomedical dataset, e.g. clinical trial data.

+

Even within the same healthcare system, EHRs can be used differently [101,102]. Individual users have unique documentation and ordering patterns, with different departments and different hospitals having different priorities that code patients and introduce missing data in a non-random fashion [103]. Patient data may be kept across several “silos” within a single health system (e.g. separate nursing documentation, registries, etc.). Even the most basic task of matching patients across systems can be challenging due to data entry issues [104]. The situation is further exacerbated by the ongoing introduction, evolution, and migration of EHR systems, especially where reorganized and acquired healthcare facilities have to merge. Further, even the ostensibly least-biased data type, laboratory measurements, can be biased based by both the healthcare process and patient health state [105]. As a result, EHR data can be less complete and less objective than expected.

+

In the wider picture, standards for EHRs are numerous and evolving. Proprietary systems, indifferent and scattered use of health information standards, and controlled terminologies makes combining and comparison of data across systems challenging [106]. Further diversity arises from variation in languages, healthcare practices, and demographics. Merging EHRs gathered in different systems (and even under different assumptions) is challenging [107].

+

Combining or replicating studies across systems thus requires controlling for both the above biases and dealing with mismatching standards. This has the practical effect of reducing cohort size, limiting statistical significance, preventing the detection of weak effects [108], and restricting the number of parameters that can be trained in a model. Further, rules-based algorithms have been popular in EHR-based research, but because these are developed at a single institution and trained with a specific patient population, they do not transfer easily to other healthcare systems [109]. Genetic studies using EHR data are subject to even more bias, as the differences in population ancestry across health centers (e.g. proportion of patients with African or Asian ancestry) can affect algorithm performance. For example, Wiley et al. [110] showed that warfarin dosing algorithms often under-perform in African Americans, illustrating that some of these issues are unresolved even at a treatment best practices level. Lack of standardization also makes it challenging for investigators skilled in deep learning to enter the field, as numerous data processing steps must be performed before algorithms are applied.

+

Finally, even if data were perfectly consistent and compatible across systems, attempts to share and combine EHR data face considerable legal and ethical barriers. Patient privacy can severely restrict the sharing and use of EHR data [111]. Here again, standards are heterogeneous and evolving, but often EHR data can often not be exported or even accessed directly for research purposes without appropriate consent. In the United States, research use of EHR data is subject both to the Common Rule and the Health Insurance Portability and Accountability Act (HIPAA). Ambiguity in the regulatory language and individual interpretation of these rules can hamper use of EHR data [112]. Once again, this has the effect of making data gathering more laborious and expensive, reducing sample size and study power.

+

Several technological solutions have been proposed in this direction, allowing access to sensitive data satisfying privacy and legal concerns. Software like DataShield [113] and ViPAR [114], although not EHR-specific, allows querying and combining of datasets and calculation of summary statistics across remote sites by “taking the analysis to the data”. The computation is carried out at the remote site. Conversely, the EH4CR project [106] allows analysis of private data by use of an inter-mediation layer that interprets remote queries across internal formats and datastores and returns the results in a de-identified standard form, thus giving real-time consistent but secure access. Continuous Analysis [115] can allow reproducible computing on private data. Using such techniques, intermediate results can be automatically tracked and shared without sharing the original data. While none of these have been used in deep learning, the potential is there.

+

Even without sharing data, algorithms trained on confidential patient data may present security risks or accidentally allow for the exposure of individual level patient data. Tramer et al. [116] showed the ability to steal trained models via public application programming interfaces (APIs). Dwork and Roth [117] demonstrate the ability to expose individual level information from accurate answers in a machine learning model. Attackers can use similar attacks to find out if a particular data instance was present in the original training set for the machine learning model [118], in this case, whether a person’s record was present. This presents a potential hazard for approaches that aim to generate data. Choi et al. propose generative adversarial neural networks as a tool to make sharable EHR data [119]; however, the authors did not take steps to protect the model from such attacks.

+

There are approaches to protect models, but they pose their own challenges. Training in a differentially private manner provides a limited guarantee that an algorithm’s output will be equally likely to occur regardless of the participation of any one individual. The limit is determined by a single parameter which provides a quantification of privacy. Simmons et al. [120] present the ability to perform genome-wide association studies (GWASs) in a differentially private manner, and Abadi et al. [121] show the ability to train deep learning classifiers under the differential privacy framework. Federated learning [122] and secure aggregations [123,124] are complementary approaches that reinforce differential privacy. Both aim to maintain privacy by training deep learning models from decentralized data sources such as personal mobile devices without transferring actual training instances. This is becoming of increasing importance with the rapid growth of mobile health applications. However, the training process in these approaches places constraints on the algorithms used and can make fitting a model substantially more challenging. In our own experience, it can be trivial to train a model without differential privacy, but quite difficult to train one within the differential privacy framework. The problem can be particularly pronounced with small sample sizes.

+

While none of these problems are insurmountable or restricted to deep learning, they present challenges that cannot be ignored. Technical evolution in EHRs and data standards will doubtless ease – although not solve – the problems of data sharing and merging. More problematic are the privacy issues. Those applying deep learning to the domain should consider the potential of inadvertently disclosing the participants’ identities. Techniques that enable training on data without sharing the raw data may have a part to play. Training within a differential privacy framework may often be warranted.

+

Discrimination and “right to an explanation” laws

+

In April 2016, the European Union adopted new rules regarding the use of personal information, the General Data Protection Regulation [125]. A component of these rules can be summed up by the phrase “right to an explanation”. Those who use machine learning algorithms must be able to explain how a decision was reached. For example, a clinician treating a patient who is aided by a machine learning algorithm may be expected to explain decisions that use the patient’s data. The new rules were designed to target categorization or recommendation systems, which inherently profile individuals. Such systems can do so in ways that are discriminatory and unlawful.

+

As datasets become larger and more complex, we may begin to identify relationships in data that are important for human health but difficult to understand. The algorithms described in this review and others like them may become highly accurate and useful for various purposes, including within medical practice. However, to discover and avoid discriminatory applications it will be important to consider interpretability alongside accuracy. A number of properties of genomic and healthcare data will make this difficult.

+

First, research samples are frequently non-representative of the general population of interest; they tend to be disproportionately sick [126], male [127], and European in ancestry [128]. One well-known consequence of these biases in genomics is that penetrance is consistently lower in the general population than would be implied by case-control data, as reviewed in [126]. Moreover, real genetic associations found in one population may not hold in other populations with different patterns of linkage disequilibrium (even when population stratification is explicitly controlled for [129]). As a result, many genomic findings are of limited value for people of non-European ancestry [128] and may even lead to worse treatment outcomes for them. Methods have been developed for mitigating some of these problems in genomic studies [126,129], but it is not clear how easily they can be adapted for deep models that are designed specifically to extract subtle effects from high-dimensional data. For example, differences in the equipment that tended to be used for cases versus controls have led to spurious genetic findings ( e.g. Sebastiani et al.’s retraction [130]). In some contexts, it may not be possible to correct for all of these differences to the degree that a deep network is unable to use them. Moreover, the complexity of deep networks makes it difficult to determine when their predictions are likely to be based on such nominally-irrelevant features of the data (called “leakage” in other fields [131]). When we are not careful with our data and models, we may inadvertently say more about the way the data was collected (which may involve a history of unequal access and discrimination) than about anything of scientific or predictive value. This fact can undermine the privacy of patient data [131] or lead to severe discriminatory consequences [132].

+

There is a small but growing literature on the prevention and mitigation of data leakage [131], as well as a closely-related literature on discriminatory model behavior [133], but it remains difficult to predict when these problems will arise, how to diagnose them, and how to resolve them in practice. There is even disagreement about which kinds of algorithmic outcomes should be considered discriminatory [134]. Despite the difficulties and uncertainties, machine learning practitioners (and particularly those who use deep neural networks, which are challenging to interpret) must remain cognizant of these dangers and make every effort to prevent harm from discriminatory predictions. To reach their potential in this domain, deep learning methods will need to be interpretable. Researchers need to consider the extent to which biases may be learned by the model and whether or not a model is sufficiently interpretable to identify bias. We discuss the challenge of model interpretability more thoroughly in Discussion.

+

Applications of deep learning to longitudinal analysis

+

Longitudinal analysis follows a population across time, for example, prospectively from birth or from the onset of particular conditions. In large patient populations, longitudinal analyses such as the Framingham Heart Study [135] and the Avon Longitudinal Study of Parents and Children [136] have yielded important discoveries about the development of disease and the factors contributing to health status. Yet, a common practice in EHR-based research is to take a snapshot at a point in time and convert patient data to a traditional vector for machine learning and statistical analysis. This results in loss of information as timing and order of events can provide insight into a patient’s disease and treatment [137]. Efforts to model sequences of events have shown promise [138] but require exceedingly large patient sizes due to discrete combinatorial bucketing. Lasko et al. [82] used autoencoders on longitudinal sequences of serum urine acid measurements to identify population subtypes. More recently, deep learning has shown promise working with both sequences (CNNs) [139] and the incorporation of past and current state (RNNs, LSTMs) [140]. This may be a particular area of opportunity for deep neural networks. The ability to recognize relevant sequences of events from a large number of trajectories requires powerful and flexible feature construction methods – an area in which deep neural networks excel.

+

Deep learning to study the fundamental biological processes underlying human disease

+

The study of cellular structure and core biological processes – transcription, translation, signaling, metabolism, etc. – in humans and model organisms will greatly impact our understanding of human disease over the long horizon [141]. Predicting how cellular systems respond to environmental perturbations and are altered by genetic variation remain daunting tasks. Deep learning offers new approaches for modeling biological processes and integrating multiple types of omic data [142], which could eventually help predict how these processes are disrupted in disease. Recent work has already advanced our ability to identify and interpret genetic variants, study microbial communities, and predict protein structures, which also relates to the problems discussed in the drug development section. In addition, unsupervised deep learning has enormous potential for discovering novel cellular states from gene expression, fluorescence microscopy, and other types of data that may ultimately prove to be clinically relevant.

+

Progress has been rapid in genomics and imaging, fields where important tasks are readily adapted to well-established deep learning paradigms. One-dimensional convolutional and recurrent neural networks are well-suited for tasks related to DNA- and RNA-binding proteins, epigenomics, and RNA splicing. Two dimensional CNNs are ideal for segmentation, feature extraction, and classification in fluorescence microscopy images [17]. Other areas, such as cellular signaling, are biologically important but studied less-frequently to date, with some exceptions [143]. This may be a consequence of data limitations or greater challenges in adapting neural network architectures to the available data. Here, we highlight several areas of investigation and assess how deep learning might move these fields forward.

+

Gene expression

+

Gene expression technologies characterize the abundance of many thousands of RNA transcripts within a given organism, tissue, or cell. This characterization can represent the underlying state of the given system and can be used to study heterogeneity across samples as well as how the system reacts to perturbation. While gene expression measurements were traditionally made by quantitative polymerase chain reaction (qPCR), low-throughput fluorescence-based methods, and microarray technologies, the field has shifted in recent years to primarily performing RNA sequencing (RNA-seq) to catalog whole transcriptomes. As RNA-seq continues to fall in price and rise in throughput, sample sizes will increase and training deep models to study gene expression will become even more useful.

+

Already several deep learning approaches have been applied to gene expression data with varying aims. For instance, many researchers have applied unsupervised deep learning models to extract meaningful representations of gene modules or sample clusters. Denoising autoencoders have been used to cluster yeast expression microarrays into known modules representing cell cycle processes [144] and to stratify yeast strains based on chemical and mutational perturbations [145]. Shallow (one hidden layer) denoising autoencoders have also been fruitful in extracting biological insight from thousands of Pseudomonas aeruginosa experiments [146,147] and in aggregating features relevant to specific breast cancer subtypes [26]. These unsupervised approaches applied to gene expression data are powerful methods for identifying gene signatures that may otherwise be overlooked. An additional benefit of unsupervised approaches is that ground truth labels, which are often difficult to acquire or are incorrect, are nonessential. However, the genes that have been aggregated into features must be interpreted carefully. Attributing each node to a single specific biological function risks over-interpreting models. Batch effects could cause models to discover non-biological features, and downstream analyses should take this into consideration.

+

Deep learning approaches are also being applied to gene expression prediction tasks. For example, a deep neural network with three hidden layers outperformed linear regression in inferring the expression of over 20,000 target genes based on a representative, well-connected set of about 1,000 landmark genes [148]. However, while the deep learning model outperformed existing algorithms in nearly every scenario, the model still displayed poor performance. The paper was also limited by computational bottlenecks that required data to be split randomly into two distinct models and trained separately. It is unclear how much performance would have increased if not for computational restrictions.

+

Epigenetic data, combined with deep learning, may have sufficient explanatory power to infer gene expression. For instance, the DeepChrome CNN [149] improved prediction accuracy of high or low gene expression from histone modifications over existing methods. Deep learning can also integrate different data types. For example, Liang et al. combined RBMs to integrate gene expression, DNA methylation, and miRNA data to define ovarian cancer subtypes [150]. While these approaches are promising, many convert gene expression measurements to categorical or binary variables, thus ablating many complex gene expression signatures present in intermediate and relative numbers.

+

Deep learning applied to gene expression data is still in its infancy, but the future is bright. Many previously untestable hypotheses can now be interrogated as deep learning enables analysis of increasing amounts of data generated by new technologies. For example, the effects of cellular heterogeneity on basic biology and disease etiology can now be explored by single-cell RNA-seq and high-throughput fluorescence-based imaging, techniques we discuss below that will benefit immensely from deep learning approaches.

+

Splicing

+

Pre-mRNA transcripts can be spliced into different isoforms by retaining or skipping subsets of exons or including parts of introns, creating enormous spatiotemporal flexibility to generate multiple distinct proteins from a single gene. This remarkable complexity can lend itself to defects that underlie many diseases. For instance, splicing mutations in the lamin A (LMNA) gene can lead to specific variants of dilated cardiomyopathy and limb girdle muscular dystrophy [151]. A recent study found that quantitative trait loci that affect splicing in lymphoblastoid cell lines are enriched within risk loci for schizophrenia, multiple sclerosis, and other immune diseases, implicating mis-splicing as a more widespread feature of human pathologies than previously thought [152]. Therapeutic strategies that aim to modulate splicing are also currently being considered for disorders such as Duchenne muscular dystrophy and spinal muscular atrophy [151].

+

Sequencing studies routinely return thousands of unannotated variants, but which cause functional changes in splicing and how are those changes manifested? Prediction of a “splicing code” has been a goal of the field for the past decade. Initial machine learning approaches used a naïve Bayes model and a 2-layer Bayesian neural network with thousands of hand-derived sequence-based features to predict the probability of exon skipping [153,154]. With the advent of deep learning, more complex models provided better predictive accuracy [155,156]. Importantly, these new approaches can take in multiple kinds of epigenomic measurements as well as tissue identity and RNA binding partners of splicing factors. Deep learning is critical in furthering these kinds of integrative studies where different data types and inputs interact in unpredictable (often nonlinear) ways to create higher-order features. Moreover, as in gene expression network analysis, interrogating the hidden nodes within neural networks could potentially illuminate important aspects of splicing behavior. For instance, tissue-specific splicing mechanisms could be inferred by training networks on splicing data from different tissues, then searching for common versus distinctive hidden nodes, a technique employed by Qin et al. for tissue-specific transcription factor (TF) binding predictions [157].

+

A parallel effort has been to use more data with simpler models. An exhaustive study using readouts of splicing for millions of synthetic intronic sequences uncovered motifs that influence the strength of alternative splice sites [158]. The authors built a simple linear model using hexamer motif frequencies that successfully generalized to exon skipping. In a limited analysis using single nucleotide polymorphisms (SNPs) from three genes, it predicted exon skipping with three times the accuracy of an existing deep learning-based framework [155]. This case is instructive in that clever sources of data, not just more descriptive models, are still critical.

+

We already understand how mis-splicing of a single gene can cause diseases such as limb girdle muscular dystrophy. The challenge now is to uncover how genome-wide alternative splicing underlies complex, non-Mendelian diseases such as autism, schizophrenia, Type 1 diabetes, and multiple sclerosis [159]. As a proof of concept, Xiong et al. [155] sequenced five autism spectrum disorder and 12 control samples, each with an average of 42,000 rare variants, and identified mis-splicing in 19 genes with neural functions. Such methods may one day enable scientists and clinicians to rapidly profile thousands of unannotated variants for functional effects on splicing and nominate candidates for further investigation. Moreover, these nonlinear algorithms can deconvolve the effects of multiple variants on a single splice event without the need to perform combinatorial in vitro experiments. The ultimate goal is to predict an individual’s tissue-specific, exon-specific splicing patterns from their genome sequence and other measurements to enable a new branch of precision diagnostics that also stratifies patients and suggests targeted therapies to correct splicing defects. However, to achieve this we expect that methods to interpret the “black box” of deep neural networks and integrate diverse data sources will be required.

+

Transcription factors and RNA-binding proteins

+

Transcription factors and RNA-binding proteins are key components in gene regulation and higher-level biological processes. TFs are regulatory proteins that bind to certain genomic loci and control the rate of mRNA production. While high-throughput sequencing techniques such as chromatin immunoprecipitation and massively parallel DNA sequencing (ChIP-seq) have been able to accurately identify targets for TFs, these experiments are both time consuming and expensive. Thus, there is a need to computationally predict binding sites and understand binding specificities de novo from sequence data. In this section we focus on TFs, with the understanding that deep learning methods for TFs are similar to those for RNA-binding proteins, though RNA-specific models do exist [160].

+

ChIP-seq and related technologies are able to identify highly likely binding sites for a certain TF, and databases such as ENCODE [161] have made freely available ChIP-seq data for hundreds of different TFs across many laboratories. In order to computationally predict transcription factor binding sites (TFBSs) on a DNA sequence, researchers initially used consensus sequences and position weight matrices to match against a test sequence [162]. Simple neural network classifiers were then proposed to differentiate positive and negative binding sites but did not show meaningful improvements over the weight matrix matching methods [163]. Later, support vector machines (SVMs) outperformed the generative methods by using k-mer features [164,165], but string kernel-based SVM systems are limited by their expensive computational cost, which is proportional to the number of training and testing sequences.

+

With the advent of deep learning, Alipanahi et al. [166] showed that convolutional neural network models could achieve state of the art results on the TFBS task and are scalable to a large number of genomic sequences. Lanchantin et al. [167] introduced several new convolutional and recurrent neural network models that further improved TFBS predictive accuracy. Due to the motif-driven nature of the TFBS task, most architectures have been convolution-based [168]. While many models for TFBS prediction resemble computer vision and NLP tasks, it is important to note that DNA sequence tasks are fundamentally different. Thus the models should be adapted from traditional deep learning models in order to account for such differences. For example, motifs may appear in either strand of a DNA sequence, resulting in two different forms of the motif (forward and reverse complement) due to complementary base pairing. To handle this issue, specialized reverse complement convolutional models share parameters to find motifs in both directions [169].

+

Despite these advances, several challenges remain. First, because the inputs (ChIP-seq measurements) are continuous and most current algorithms are designed to produce binary outputs (whether or not there is TF binding at a particular site), false positives or false negatives can result depending on the threshold chosen by the algorithm. Second, most methods predict binding of TFs at sites in isolation, whereas in reality multiple TFs may compete for binding at a single site or act synergistically to co-occupy it. Fortunately, multi-task models are rapidly improving at simultaneous prediction of many TFs’ binding at any given site [170]. Third, it is unclear exactly how to define a non-binding or “negative” site in the training data because the number of positive binding sites of a particular TF is relatively small with respect to the total number of base-pairs in a genome (see Discussion).

+

While deep learning-based models can automatically extract features for TFBS prediction at the sequence level, they often cannot predict binding patterns for cell types or conditions that have not been previously studied. One solution could be to introduce a multimodal model that, in addition to sequence data, incorporates cell-line specific features such as chromatin accessibility, DNA methylation, or gene expression. Without cell-specific features, another solution could be to use domain adaptation methods where the model trains on a source cell type and uses unsupervised feature extraction methods to predict on a target cell type. TFImpute [157] predicts binding in new cell type-TF pairs, but the cell types must be in the training set for other TFs. This is a step in the right direction, but a more general domain transfer model across cell types would be more useful.

+

Deep learning can also illustrate TF binding preferences. Lanchantin et al. [167] and Shrikumar et al. [171] developed tools to visualize TF motifs learned from TFBS classification tasks. Alipanahi et al. [166] also introduced mutation maps, where they could easily mutate, add, or delete base pairs in a sequence and see how the model changed its prediction. Though time consuming to assay in a lab, this was easy to simulate with a computational model. As we learn to better visualize and analyze the hidden nodes within deep learning models, our understanding of TF binding motifs and dynamics will likely improve.

+

Promoters and enhancers

+

From TF binding to promoters and enhancers

+

Multiple TFs act in concert to coordinate changes in gene regulation at the genomic regions known as promoters and enhancers. Each gene has an upstream promoter, essential for initiating that gene’s transcription. The gene may also interact with multiple enhancers, which can amplify transcription in particular cellular contexts. These contexts include different cell types in development or environmental stresses.

+

Promoters and enhancers provide a nexus where clusters of TFs and binding sites mediate downstream gene regulation, starting with transcription. The gold standard to identify an active promoter or enhancer requires demonstrating its ability to affect transcription or other downstream gene products. Even extensive biochemical TF binding data has thus far proven insufficient on its own to accurately and comprehensively locate promoters and enhancers. We lack sufficient understanding of these elements to derive a mechanistic “promoter code” or “enhancer code”. But extensive labeled data on promoters and enhancers lends itself to probabilistic classification. The complex interplay of TFs and chromatin leading to the emergent properties of promoter and enhancer activity seems particularly apt for representation by deep neural networks.

+

Promoters

+

Despite decades of work, computational identification of promoters remains a stubborn problem [172]. Researchers have used neural networks for promoter recognition as early as 1996 [173]. Recently, a CNN recognized promoter sequences with sensitivity and specificity exceeding 90% [174]. Most activity in computational prediction of regulatory regions, however, has moved to enhancer identification. Because one can identify promoters with straightforward biochemical assays [175,176], the direct rewards of promoter prediction alone have decreased. But the reliable ground truth provided by these assays makes promoter identification an appealing test bed for deep learning approaches that can also identify enhancers.

+

Enhancers

+

Recognizing enhancers presents additional challenges. Enhancers may be up to 1,000,000 bp away from the affected promoter, and even within introns of other genes [177]. Enhancers do not necessarily operate on the nearest gene and may affect multiple genes. Their activity is frequently tissue- or context-specific. No biochemical assay can reliably identify all enhancers. Distinguishing them from other regulatory elements remains difficult, and some believe the distinction somewhat artificial [178]. While these factors make the enhancer identification problem more difficult, they also make a solution more valuable.

+

Several neural network approaches yielded promising results in enhancer prediction. Both Basset [179] and DeepEnhancer [180] used CNNs to predict enhancers. DECRES used a feed-forward neural network [181] to distinguish between different kinds of regulatory elements, such as active enhancers, and promoters. DECRES had difficulty distinguishing between inactive enhancers and promoters. They also investigated the power of sequence features to drive classification, finding that beyond CpG islands, few were useful.

+

Comparing the performance of enhancer prediction methods illustrates the problems in using metrics created with different benchmarking procedures. Both the Basset and DeepEnhancer studies include comparisons to a baseline SVM approach, gkm-SVM [164]. The Basset study reports gkm-SVM attains a mean auPRC of 0.322 over 164 cell types [179]. The DeepEnhancer study reports for gkm-SVM a dramatically different auPRC of 0.899 on nine cell types [180]. This large difference means it’s impossible to directly compare the performance of Basset and DeepEnhancer based solely on their reported metrics. DECRES used a different set of metrics altogether. To drive further progress in enhancer identification, we must develop a common and comparable benchmarking procedure (see Discussion).

+

Promoter-enhancer interactions

+

In addition to the location of enhancers, identifying enhancer-promoter interactions in three-dimensional space will provide critical knowledge for understanding transcriptional regulation. SPEID used a CNN to predicted these interactions with only sequence and the location of putative enhancers and promoters along a one-dimensional chromosome [182]. It compared well to other methods using a full complement of biochemical data from ChIP-seq and other epigenomic methods. Of course, the putative enhancers and promoters used were themselves derived from epigenomic methods. But one could easily replace them with the output of one of the enhancer or promoter prediction methods above.

+

Micro-RNA binding

+

Prediction of microRNAs (miRNAs) and miRNA targets is of great interest, as they are critical components of gene regulatory networks and are often conserved across great evolutionary distance [183,184]. While many machine learning algorithms have been applied to these tasks, they currently require extensive feature selection and optimization. For instance, one of the most widely adopted tools for miRNA target prediction, TargetScan, trained multiple linear regression models on 14 hand-curated features including structural accessibility of the target site on the mRNA, the degree of site conservation, and predicted thermodynamic stability of the miRNA-mRNA complex [185]. Some of these features, including structural accessibility, are imperfect or empirically derived. In addition, current algorithms suffer from low specificity [186].

+

As in other applications, deep learning promises to achieve equal or better performance in predictive tasks by automatically engineering complex features to minimize an objective function. Two recently published tools use different recurrent neural network-based architectures to perform miRNA and target prediction with solely sequence data as input [186,187]. Though the results are preliminary and still based on a validation set rather than a completely independent test set, they were able to predict microRNA target sites with higher specificity and sensitivity than TargetScan. Excitingly, these tools seem to show that RNNs can accurately align sequences and predict bulges, mismatches, and wobble base pairing without requiring the user to input secondary structure predictions or thermodynamic calculations. Further incremental advances in deep learning for miRNA and target prediction will likely be sufficient to meet the current needs of systems biologists and other researchers who use prediction tools mainly to nominate candidates that are then tested experimentally.

+

Protein secondary and tertiary structure

+

Proteins play fundamental roles in almost all biological processes, and understanding their structure is critical for basic biology and drug development. UniProt currently has about 94 million protein sequences, yet fewer than 100,000 proteins across all species have experimentally-solved structures in Protein Data Bank (PDB). As a result, computational structure prediction is essential for a majority of proteins. However, this is very challenging, especially when similar solved structures, called templates, are not available in PDB. Over the past several decades, many computational methods have been developed to predict aspects of protein structure such as secondary structure, torsion angles, solvent accessibility, inter-residue contact maps, disorder regions, and side-chain packing. In recent years, multiple deep learning architectures have been applied, including deep belief networks, LSTMs, CNNs, and deep convolutional neural fields (DeepCNFs) [31,188].

+

Here we focus on deep learning methods for two representative sub-problems: secondary structure prediction and contact map prediction. Secondary structure refers to local conformation of a sequence segment, while a contact map contains information on all residue-residue contacts. Secondary structure prediction is a basic problem and an almost essential module of any protein structure prediction package. Contact prediction is much more challenging than secondary structure prediction, but it has a much larger impact on tertiary structure prediction. In recent years, the accuracy of contact prediction has greatly improved [29,189–191].

+

One can represent protein secondary structure with three different states (alpha helix, beta strand, and loop regions) or eight finer-grained states. Accuracy of a three-state prediction is called Q3, and accuracy of an 8-state prediction is called Q8. Several groups [30,192,193] applied deep learning to protein secondary structure prediction but were unable to achieve significant improvement over the de facto standard method PSIPRED [194], which uses two shallow feedforward neural networks. In 2014, Zhou and Troyanskaya demonstrated that they could improve Q8 accuracy by using a deep supervised and convolutional generative stochastic network [195]. In 2016 Wang et al. developed a DeepCNF model that improved Q3 and Q8 accuracy as well as prediction of solvent accessibility and disorder regions [31,188]. DeepCNF achieved a higher Q3 accuracy than the standard maintained by PSIPRED for more than 10 years. This improvement may be mainly due to the ability of convolutional neural fields to capture long-range sequential information, which is important for beta strand prediction. Nevertheless, the improvements in secondary structure prediction from DeepCNF are unlikely to result in a commensurate improvement in tertiary structure prediction since secondary structure mainly reflects coarse-grained local conformation of a protein structure.

+

Protein contact prediction and contact-assisted folding (i.e. folding proteins using predicted contacts as restraints) represents a promising new direction for ab initio folding of proteins without good templates in PDB. Co-evolution analysis is effective for proteins with a very large number (>1000) of sequence homologs [191], but fares poorly for proteins without many sequence homologs. By combining co-evolution information with a few other protein features, shallow neural network methods such as MetaPSICOV [189] and CoinDCA-NN [196] have shown some advantage over pure co-evolution analysis for proteins with few sequence homologs, but their accuracy is still far from satisfactory. In recent years, deeper architectures have been explored for contact prediction, such as CMAPpro [197], DNCON [198] and PConsC [199]. However, blindly tested in the well-known CASP competitions, these methods did not show any advantage over MetaPSICOV [189].

+

Recently, Wang et al. proposed the deep learning method RaptorX-Contact [29], which significantly improves contact prediction over MetaPSICOV and pure co-evolution methods, especially for proteins without many sequence homologs. It employs a network architecture formed by one 1D residual neural network and one 2D residual neural network. Blindly tested in the latest CASP competition (i.e. CASP12 [200]), RaptorX-Contact ranked first in F1 score on free-modeling targets as well as the whole set of targets. In CAMEO (which can be interpreted as a fully-automated CASP) [201], its predicted contacts were also able to fold proteins with a novel fold and only 65-330 sequence homologs. This technique also worked well on membrane proteins even when trained on non-membrane proteins [202]. RaptorX-Contact performed better mainly due to introduction of residual neural networks and exploitation of contact occurrence patterns by simultaneously predicting all the contacts in a single protein.

+

Taken together, ab initio folding is becoming much easier with the advent of direct evolutionary coupling analysis and deep learning techniques. We expect further improvements in contact prediction for proteins with fewer than 1000 homologs by studying new deep network architectures. However, it is unclear if there is an effective way to use deep learning to improve prediction for proteins with few or no sequence homologs. Finally, the deep learning methods summarized above also apply to interfacial contact prediction for protein complexes but may be less effective since on average protein complexes have fewer sequence homologs.

+

Morphological phenotypes

+

A field poised for dramatic revolution by deep learning is bioimage analysis. Thus far, the primary use of deep learning for biological images has been for segmentation – that is, for the identification of biologically relevant structures in images such as nuclei, infected cells, or vasculature – in fluorescence or even brightfield channels [203]. Once so-called regions of interest have been identified, it is often straightforward to measure biological properties of interest, such as fluorescence intensities, textures, and sizes. Given the dramatic successes of deep learning in biological imaging, we simply refer to articles that review recent advancements [17,203,204]. For deep learning to become commonplace for biological image segmentation, we need user-friendly tools.

+

We anticipate an additional paradigm shift in bioimaging that will be brought about by deep learning: what if images of biological samples, from simple cell cultures to three-dimensional organoids and tissue samples, could be mined for much more extensive biologically meaningful information than is currently standard? For example, a recent study demonstrated the ability to predict lineage fate in hematopoietic cells up to three generations in advance of differentiation [205]. In biomedical research, most often biologists decide in advance what feature to measure in images from their assay system. Although classical methods of segmentation and feature extraction can produce hundreds of metrics per cell in an image, deep learning is unconstrained by human intuition and can in theory extract more subtle features through its hidden nodes. Already, there is evidence deep learning can surpass the efficacy of classical methods [206], even using generic deep convolutional networks trained on natural images [207], known as transfer learning. Recent work by Johnson et al. [208] demonstrated how the use of a conditional adversarial autoencoder allows for a probabilistic interpretation of cell and nuclear morphology and structure localization from fluorescence images. The proposed model is able to generalize well to a wide range of subcellular localizations. The generative nature of the model allows it to produce high-quality synthetic images predicting localization of subcellular structures by directly modeling the localization of fluorescent labels. Notably, this approach reduces the modeling time by omitting the subcellular structure segmentation step.

+

The impact of further improvements on biomedicine could be enormous. Comparing cell population morphologies using conventional methods of segmentation and feature extraction has already proven useful for functionally annotating genes and alleles, identifying the cellular target of small molecules, and identifying disease-specific phenotypes suitable for drug screening [209–211]. Deep learning would bring to these new kinds of experiments – known as image-based profiling or morphological profiling – a higher degree of accuracy, stemming from the freedom from human-tuned feature extraction strategies.

+

Single-cell data

+

Single-cell methods are generating excitement as biologists characterize the vast heterogeneity within unicellular species and between cells of the same tissue type in the same organism [212]. For instance, tumor cells and neurons can both harbor extensive somatic variation [213]. Understanding single-cell diversity in all its dimensions – genetic, epigenetic, transcriptomic, proteomic, morphologic, and metabolic – is key if treatments are to be targeted not only to a specific individual, but also to specific pathological subsets of cells. Single-cell methods also promise to uncover a wealth of new biological knowledge. A sufficiently large population of single cells will have enough representative “snapshots” to recreate timelines of dynamic biological processes. If tracking processes over time is not the limiting factor, single-cell techniques can provide maximal resolution compared to averaging across all cells in bulk tissue, enabling the study of transcriptional bursting with single-cell fluorescence in situ hybridization or the heterogeneity of epigenetic patterns with single-cell Hi-C or ATAC-seq [214,215]. Joint profiling of single-cell epigenetic and transcriptional states provides unprecedented views of regulatory processes [216].

+

However, large challenges exist in studying single cells. Relatively few cells can be assayed at once using current droplet, imaging, or microwell technologies, and low-abundance molecules or modifications may not be detected by chance due to a phenomenon known as dropout. To solve this problem, Angermueller et al. [217] trained a neural network to predict the presence or absence of methylation of a specific CpG site in single cells based on surrounding methylation signal and underlying DNA sequence, achieving several percentage points of improvement compared to random forests or deep networks trained only on CpG or sequence information. Similar deep learning methods have been applied to impute low-resolution ChIP-seq signal from bulk tissue with great success, and they could easily be adapted to single-cell data [157,218]. Deep learning has also been useful for dealing with batch effects [219].

+

Examining populations of single cells can reveal biologically meaningful subsets of cells as well as their underlying gene regulatory networks [220]. Unfortunately, machine learning methods generally struggle with imbalanced data – when there are many more examples of class 1 than class 2 – because prediction accuracy is usually evaluated over the entire dataset. To tackle this challenge, Arvaniti et al. [221] classified healthy and cancer cells expressing 25 markers by using the most discriminative filters from a CNN trained on the data as a linear classifier. They achieved impressive performance, even for cell types where the subset percentage ranged from 0.1 to 1%, significantly outperforming logistic regression and distance-based outlier detection methods. However, they did not benchmark against random forests, which tend to work better for imbalanced data, and their data was relatively low dimensional. Future work is needed to establish the utility of deep learning in cell subset identification, but the stunning improvements in image classification over the past 5 years [222] suggest transformative potential.

+

The sheer quantity of omic information that can be obtained from each cell, as well as the number of cells in each dataset, uniquely position single-cell data to benefit from deep learning. In the future, lineage tracing could be revolutionized by using autoencoders to reduce the feature space of transcriptomic or variant data followed by algorithms to learn optimal cell differentiation trajectories [223] or by feeding cell morphology and movement into neural networks [205]. Reinforcement learning algorithms [224] could be trained on the evolutionary dynamics of cancer cells or bacterial cells undergoing selection pressure and reveal whether patterns of adaptation are random or deterministic, allowing us to develop therapeutic strategies that forestall resistance. We are excited to see the creative applications of deep learning to single-cell biology that emerge over the next few years.

+

Metagenomics

+

Metagenomics, which refers to the study of genetic material – 16S rRNA or whole-genome shotgun DNA – from microbial communities, has revolutionized the study of micro-scale ecosystems within and around us. In recent years, machine learning has proved to be a powerful tool for metagenomic analysis. 16S rRNA has long been used to deconvolve mixtures of microbial genomes, yet this ignores more than 99% of the genomic content. Subsequent tools aimed to classify 300 bp-3000 bp reads from complex mixtures of microbial genomes based on tetranucleotide frequencies, which differ across organisms [225], using supervised [226,227] or unsupervised methods [228]. Then, researchers began to use techniques that could estimate relative abundances from an entire sample faster than classifying individual reads [229–232]. There is also great interest in identifying and annotating sequence reads [233,234]. However, the focus on taxonomic and functional annotation is just the first step. Several groups have proposed methods to determine host or environment phenotypes from the organisms that are identified [235–238] or overall sequence composition [239]. Also, researchers have looked into how feature selection can improve classification [238,240], and techniques have been proposed that are classifier-independent [241,242].

+

Most neural networks are used for phylogenetic classification or functional annotation from sequence data where there is ample data for training. Neural networks have been applied successfully to gene annotation (e.g. Orphelia [243] and FragGeneScan [244]). Representations (similar to Word2Vec [78] in natural language processing) for protein family classification have been introduced and classified with a skip-gram neural network [245]. Recurrent neural networks show good performance for homology and protein family identification [246,247].

+

One of the first techniques of de novo genome binning used self-organizing maps, a type of neural network [228]. Essinger et al. [248] used Adaptive Resonance Theory to cluster similar genomic fragments and showed that it had better performance than k-means. However, other methods based on interpolated Markov models [249] have performed better than these early genome binners. Neural networks can be slow and therefore have had limited use for reference-based taxonomic classification, with TAC-ELM [250] being the only neural network-based algorithm to taxonomically classify massive amounts of metagenomic data. An initial study successfully applied neural networks to taxonomic classification of 16S rRNA genes, with convolutional networks providing about 10% accuracy genus-level improvement over RNNs and random forests [251]. However, this study evaluated only 3000 sequences.

+

Neural network uses for classifying phenotype from microbial composition are just beginning. A simple multi-layer perceptron (MLP) was able to classify wound severity from microbial species present in the wound [252]. Recently, Ditzler et al. associated soil samples with pH level using MLPs, DBNs, and RNNs [253]. Besides classifying samples appropriately, internal phylogenetic tree nodes inferred by the networks represented features for low and high pH. Thus, hidden nodes might provide biological insight as well as new features for future metagenomic sample comparison. Also, an initial study has shown promise of these networks for diagnosing disease [254].

+

Challenges remain in applying deep neural networks to metagenomics problems. They are not yet ideal for phenotype classification because most studies contain tens of samples and hundreds or thousands of features (species). Such underdetermined, or ill-conditioned, problems are still a challenge for deep neural networks that require many training examples. Also, due to convergence issues [255], taxonomic classification of reads from whole genome sequencing seems out of reach at the moment for deep neural networks. There are only thousands of full-sequenced genomes as compared to hundreds of thousands of 16S rRNA sequences available for training.

+

However, because RNNs have been applied to base calls for the Oxford Nanopore long-read sequencer with some success [256] (discussed below), one day the entire pipeline, from denoising to functional classification, may be combined into one step using powerful LSTMs [257]. For example, metagenomic assembly usually requires binning then assembly, but could deep neural nets accomplish both tasks in one network? We believe the greatest potential in deep learning is to learn the complete characteristics of a metagenomic sample in one complex network.

+

Sequencing and variant calling

+

While we have so far primarily discussed the role of deep learning in analyzing genomic data, deep learning can also substantially improve our ability to obtain the genomic data itself. We discuss two specific challenges: calling SNPs and indels (insertions and deletions) with high specificity and sensitivity and improving the accuracy of new types of data such as nanopore sequencing. These two tasks are critical for studying rare variation, allele-specific transcription and translation, and splice site mutations. In the clinical realm, sequencing of rare tumor clones and other genetic diseases will require accurate calling of SNPs and indels.

+

Current methods achieve relatively high (>99%) precision at 90% recall for SNPs and indel calls from Illumina short-read data [258], yet this leaves a large number of potentially clinically-important remaining false positives and false negatives. These methods have so far relied on experts to build probabilistic models that reliably separate signal from noise. However, this process is time consuming and fundamentally limited by how well we understand and can model the factors that contribute to noise. Recently, two groups have applied deep learning to construct data-driven unbiased noise models. One of these models, DeepVariant, leverages Inception, a neural network trained for image classification by Google Brain, by encoding reads around a candidate SNP as a 221x100 bitmap image, where each column is a nucleotide and each row is a read from the sample library [258]. The top 5 rows represent the reference, and the bottom 95 rows represent randomly sampled reads that overlap the candidate variant. Each RGBA (red/green/blue/alpha) image pixel encodes the base (A, C, G, T) as a different red value, quality score as a green value, strand as a blue value, and variation from the reference as the alpha value. The neural network outputs genotype probabilities for each candidate variant. They were able to achieve better performance than GATK, a leading genotype caller, even when GATK was given information about population variation for each candidate variant. Another method, still in its infancy, hand-developed 62 features for each candidate variant and fed these vectors into a fully connected deep neural network [259]. Unfortunately, this feature set required at least 15 iterations of software development to fine-tune, which suggests that these models may not generalize.

+

Variant calling will benefit more from optimizing neural network architectures than from developing features by hand. An interesting and informative next step would be to rigorously test if encoding raw sequence and quality data as an image, tensor, or some other mixed format produces the best variant calls. Because many of the latest neural network architectures (ResNet, Inception, Xception, and others) are already optimized for and pre-trained on generic, large-scale image datasets [260], encoding genomic data as images could prove to be a generally effective and efficient strategy.

+

In limited experiments, DeepVariant was robust to sequencing depth, read length, and even species [258]. However, a model built on Illumina data, for instance, may not be optimal for Pacific Biosciences long-read data or MinION nanopore data, which have vastly different specificity and sensitivity profiles and signal-to-noise characteristics. Recently, Boza et al. used bidirectional recurrent neural networks to infer the E. coli sequence from MinION nanopore electric current data with higher per-base accuracy than the proprietary hidden Markov model-based algorithm Metrichor [256]. Unfortunately, training any neural network requires a large amount of data, which is often not available for new sequencing technologies. To circumvent this, one very preliminary study simulated mutations and spiked them into somatic and germline RNA-seq data, then trained and tested a neural network on simulated paired RNA-seq and exome sequencing data [261]. However, because this model was not subsequently tested on ground-truth datasets, it is unclear whether simulation can produce sufficiently realistic data to produce reliable models.

+

Method development for interpreting new types of sequencing data has historically taken two steps: first, easily implemented hard cutoffs that prioritize specificity over sensitivity, then expert development of probabilistic models with hand-developed inputs [261]. We anticipate that these steps will be replaced by deep learning, which will infer features simply by its ability to optimize a complex model against data.

+

The impact of deep learning in treating disease and developing new treatments

+

Given the need to make better, faster interventions at the point of care – incorporating the complex calculus of a patients symptoms, diagnostics, and life history – there have been many attempts to apply deep learning to patient treatment. Success in this area could help to enable personalized healthcare or precision medicine [262,263]. Earlier, we reviewed approaches for patient categorization. Here, we examine the potential for better treatment, which broadly, may divided into methods for improved choices of interventions for patients and those for development of new interventions.

+

Clinical decision making

+

In 1996, Tu [264] compared the effectiveness of artificial neural networks and logistic regression, questioning whether these techniques would replace traditional statistical methods for predicting medical outcomes such as myocardial infarction [265] or mortality [266]. He posited that while neural networks have several advantages in representational power, the difficulties in interpretation may limit clinical applications, a limitation that still remains today. In addition, the challenges faced by physicians parallel those encountered by deep learning. For a given patient, the number of possible diseases is very large, with a long tail of rare diseases and patients are highly heterogeneous and may present with very different signs and symptoms for the same disease. Still, in 2006 Lisboa and Taktak [267] examined the use of artificial neural networks in medical journals, concluding that they improved healthcare relative to traditional screening methods in 21 of 27 studies.

+

While further progress has been made in using deep learning for clinical decision making, it is hindered by a challenge common to many deep learning applications: it is much easier to predict an outcome than to suggest an action to change the outcome. Several attempts [86,88] at recasting the clinical decision-making problem into a prediction problem (i.e. prediction of which treatment will most improve the patient’s health) have accurately predicted survival patterns, but technical and medical challenges remain for clinical adoption (similar to those for categorization). In particular, remaining barriers include actionable interpretability of deep learning models, fitting deep models to limited and heterogeneous data, and integrating complex predictive models into a dynamic clinical environment.

+

A critical challenge in providing treatment recommendations is identifying a causal relationship for each recommendation. Causal inference is often framed in terms of counterfactual question [268]. Johansson et al. [269] use deep neural networks to create representation models for covariates that capture nonlinear effects and show significant performance improvements over existing models. In a less formal approach, Kale et al. [270] first create a deep neural network to model clinical time series and then analyze the relationship of the hidden features to the output using a causal approach.

+

A common challenge for deep learning is the interpretability of the models and their predictions. The task of clinical decision making is necessarily risk-averse, so model interpretability is key. Without clear reasoning, it is difficult to establish trust in a model. As described above, there has been some work to directly assign treatment plans without interpretability; however, the removal of human experts from the decision-making loop make the models difficult to integrate with clinical practice. To alleviate this challenge, several studies have attempted to create more interpretable deep models, either specifically for healthcare or as a general procedure for deep learning (see Discussion).

+

Predicting patient trajectories

+

A common application for deep learning in this domain is the temporal structure of healthcare records. Many studies [271–274] have used RNNs to categorize patients, but most stop short of suggesting clinical decisions. Nemati et al. [275] used deep reinforcement learning to optimize a heparin dosing policy for intensive care patients. However, because the ideal dosing policy is unknown, the model’s predictions must be evaluated on counter-factual data. This represents a common challenge when bridging the gap between research and clinical practice. Because the ground-truth is unknown, researchers struggle to evaluate model predictions in the absence of interventional data, but clinical application is unlikely until the model has been shown to be effective. The impressive applications of deep reinforcement learning to other domains [224] have relied on knowledge of the underlying processes (e.g. the rules of the game). Some models have been developed for targeted medical problems [276], but a generalized engine is beyond current capabilities.

+

Clinical trial efficiency

+

A clinical deep learning task that has been more successful is the assignment of patients to clinical trials. Ithapu et al. [277] used a randomized denoising autoencoder to learn a multimodal imaging marker that predicts future cognitive and neural decline from positron emission tomography (PET), amyloid florbetapir PET, and structural magnetic resonance imaging. By accurately predicting which cases will progress to dementia, they were able to efficiently assign patients to a clinical trial and reduced the required sample sizes by a factor of five. Similarly, Artemov et al. [278] applied deep learning to predict which clinical trials were likely to fail and which were likely to succeed. By predicting the side effects and pathway activations of each drug and translating these activations to a success probability, their deep learning-based approach was able to significantly outperform a random forest classifier trained on gene expression changes. These approaches suggest promising directions to improve the efficiency of clinical trials and accelerate drug development.

+

Drug repositioning

+

Drug repositioning (or repurposing) is an attractive option for delivering new drugs to the market because of the high costs and failure rates associated with more traditional drug discovery approaches [279,280]. A decade ago, the Connectivity Map [281] had a sizeable impact. Reverse matching disease gene expression signatures with a large set of reference compound profiles allowed researchers to formulate repurposing hypotheses at scale using a simple non-parametric test. Since then, several advanced computational methods have been applied to formulate and validate drug repositioning hypotheses [282–284]. Using supervised learning and collaborative filtering to tackle this type of problem is proving successful, especially when coupling disease or compound omic data with topological information from protein-protein or protein-compound interaction networks [285–287].

+

For example, Menden et al. [288] used a shallow neural network to predict sensitivity of cancer cell lines to drug treatment using both cell line and drug features, opening the door to precision medicine and drug repositioning opportunities in cancer. More recently, Aliper et al. [37] used gene- and pathway-level drug perturbation transcriptional profiles from the Library of Network-Based Cellular Signatures [289] to train a fully connected deep neural network to predict drug therapeutic uses and indications. By using confusion matrices and leveraging misclassification, the authors formulated a number of interesting hypotheses, including repurposing cardiovascular drugs such as otenzepad and pinacidil for neurological disorders.

+

Drug repositioning can also be approached by attempting to predict novel drug-target interactions and then repurposing the drug for the associated indication [290,291]. Wang et al. [292] devised a pairwise input neural network with two hidden layers that takes two inputs, a drug and a target binding site, and predicts whether they interact. Wang et al. [38] trained individual RBMs for each target in a drug-target interaction network and used these models to predict novel interactions pointing to new indications for existing drugs. Wen et al. [39] extended this concept to deep learning by creating a DBN called DeepDTIs, which predicts interactions using chemical structure and protein sequence features.

+

Drug repositioning appears an obvious candidate for deep learning both because of the large amount of high-dimensional data available and the complexity of the question being asked. However, perhaps the most promising piece of work in this space [37] is more of a proof of concept than a real-world hypothesis-generation tool; notably, deep learning was used to predict drug indications but not for the actual repositioning. At present, some of the most popular state-of-the-art methods for signature-based drug repurposing [293] do not use predictive modeling. A mature and production-ready framework for drug repositioning via deep learning is currently missing.

+

Drug development

+

Ligand-based prediction of bioactivity

+

High-throughput chemical screening in biomedical research aims to improve therapeutic options over a long term horizon [22]. The objective is to discover which small molecules (also referred to as chemical compounds or ligands) specifically affect the activity of a target, such as a kinase, protein-protein interaction, or broader cellular phenotype. This screening is often one of the first steps in a long drug discovery pipeline, where novel molecules are pursued for their ability to inhibit or enhance disease-relevant biological mechanisms [294]. Initial hits are confirmed to eliminate false positives and proceed to the lead generation stage [295], where they are evaluated for absorption, distribution, metabolism, excretion, and toxicity (ADMET) and other properties. It is desirable to advance multiple lead series, clusters of structurally-similar active chemicals, for further optimization by medicinal chemists to protect against unexpected failures in the later stages of drug discovery [294].

+

Computational work in this domain aims to identify sufficient candidate active compounds without exhaustively screening libraries of hundreds of thousands or millions of chemicals. Predicting chemical activity computationally is known as virtual screening. This task has been treated variously as a classification, regression, or ranking problem. In reality, it does not fit neatly into any of those categories. An ideal algorithm will rank a sufficient number of active compounds before the inactives, but the rankings of actives relative to other actives and inactives are less important [296]. Computational modeling also has the potential to predict ADMET traits for lead generation [297] and how drugs are metabolized [298].

+

Ligand-based approaches train on chemicals’ features without modeling target features (e.g. protein structure). Chemical features may be represented as a list of molecular descriptors such as molecular weight, atom counts, functional groups, charge representations, summaries of atom-atom relationships in the molecular graph, and more sophisticated derived properties [299]. Alternatively, chemicals can be characterized with the fingerprint bit vectors, textual strings, or novel learned representations described below. Neural networks have a long history in this domain [20,23], and the 2012 Merck Molecular Activity Challenge on Kaggle generated substantial excitement about the potential for high-parameter deep learning approaches. The winning submission was an ensemble that included a multi-task multi-layer perceptron network [300]. The sponsors noted drastic improvements over a random forest baseline, remarking “we have seldom seen any method in the past 10 years that could consistently outperform [random forest] by such a margin” [301]. Subsequent work (reviewed in more detail by Goh et al. [21]) explored the effects of jointly modeling far more targets than the Merck challenge [302,303], with Ramsundar et al. [303] showing that the benefits of multi-task networks had not yet saturated even with 259 targets. Although DeepTox [304], a deep learning approach, won another competition, the Toxicology in the 21st Century (Tox21) Data Challenge, it did not dominate alternative methods as thoroughly as in other domains. DeepTox was the top performer on 9 of 15 targets and highly competitive with the top performer on the others. However, for many targets there was little separation between the top two or three methods.

+

The nuanced Tox21 performance may be more reflective of the practical challenges encountered in ligand-based chemical screening than the extreme enthusiasm generated by the Merck competition. A study of 22 ADMET tasks demonstrated that there are limitations to multi-task transfer learning that are in part a consequence of the degree to which tasks are related [297]. Some of the ADMET datasets showed superior performance in multi-task models with only 22 ADMET tasks compared to multi-task models with over 500 less-similar tasks. In addition, the training datasets encountered in practical applications may be tiny relative to what is available in public datasets and organized competitions. A study of BACE-1 inhibitors included only 1547 compounds [305]. Machine learning models were able to train on this limited dataset, but overfitting was a challenge and the differences between random forests and a deep neural network were negligible, especially in the classification setting. Overfitting is still a problem in larger chemical screening datasets with tens or hundreds of thousands of compounds because the number of active compounds can be very small, on the order of 0.1% of all tested chemicals for a typical target [306]. This is consistent with the strong performance of low-parameter neural networks that emphasize compound-compound similarity, such as influence-relevance voter [296,307], instead of predicting compound activity directly from chemical features.

+

Much of the recent excitement in this domain has come from what could be considered a creative experimentation phase, in which deep learning has offered novel possibilities for feature representation and modeling of chemical compounds. A molecular graph, where atoms are labeled nodes and bonds are labeled edges, is a natural way to represent a chemical structure. Traditional machine learning approaches relied on preprocessing the graph into a feature vector, such as a fixed-width bit vector fingerprint [308]. The same fingerprints have been used by some drug-target interaction methods discussed above [39]. An overly simplistic but approximately correct view of chemical fingerprints is that each bit represents the presence or absence of a particular chemical substructure in the molecular graph. Modern neural networks can operate directly on the molecular graph as input. Duvenaud et al. [309] generalized standard circular fingerprints by substituting discrete operations in the fingerprinting algorithm with operations in a neural network, producing a real-valued feature vector instead of a bit vector. Other approaches offer trainable networks that can learn chemical feature representations that are optimized for a particular prediction task. Lusci et al. [310] applied recursive neural networks for directed acyclic graphs to undirected molecular graphs by creating an ensemble of directed graphs in which one atom is selected as the root node. Graph convolutions on undirected molecular graphs have eliminated the need to enumerate artificially directed graphs, learning feature vectors for atoms that are a function of the properties of neighboring atoms and local regions on the molecular graph [311,312].

+

Advances in chemical representation learning have also enabled new strategies for learning chemical-chemical similarity functions. Altae-Tran et al. developed a one-shot learning network [312] to address the reality that most practical chemical screening studies are unable to provide the thousands or millions of training compounds that are needed to train larger multi-task networks. Using graph convolutions to featurize chemicals, the network learns an embedding from compounds into a continuous feature space such that compounds with similar activities in a set of training tasks have similar embeddings. The approach is evaluated in an extremely challenging setting. The embedding is learned from a subset of prediction tasks (e.g. activity assays for individual proteins), and only one to ten labeled examples are provided as training data on a new task. On Tox21 targets, even when trained with one task-specific active compound and one inactive compound, the model is able to generalize reasonably well because it has learned an informative embedding function from the related tasks. Random forests, which cannot take advantage of the related training tasks, trained in the same setting are only slightly better than a random classifier. Despite the success on Tox21, performance on MUV datasets, which contains assays designed to be challenging for chemical informatics algorithms, is considerably worse. The authors also demonstrate the limitations of transfer learning as embeddings learned from the Tox21 assays have little utility for a drug adverse reaction dataset.

+

These novel, learned chemical feature representations may prove to be essential for accurately predicting why some compounds with similar structures yield similar target effects and others produce drastically different results. Currently, these methods are enticing but do not necessarily outperform classic approaches by a large margin. The neural fingerprints [309] were narrowly beaten by regression using traditional circular fingerprints on a drug efficacy prediction task but were superior for predicting solubility or photovoltaic efficiency. In the original study, graph convolutions [311] performed comparably to a multi-task network using standard fingerprints and slightly better than the neural fingerprints [309] on the drug efficacy task but were slightly worse than the influence-relevance voter method on an HIV dataset. [296]. Broader recent benchmarking has shown that relative merits of these methods depends on the dataset and cross validation strategy [313], though evaluation often uses auROC (area under the receiver operating characteristic curve), which has limited utility due to the large class imbalance (see Discussion).

+

We remain optimistic for the potential of deep learning and specifically representation learning in drug discovery. Rigorous benchmarking on broad and diverse prediction tasks will be as important as novel neural network architectures to advance the state of the art and convincingly demonstrate superiority over traditional cheminformatics techniques. Fortunately, there has recently been much progress in this direction. The DeepChem software [312,314] and MoleculeNet benchmarking suite [313] built upon it contain chemical bioactivity and toxicity prediction datasets, multiple compound featurization approaches including graph convolutions, and various machine learning algorithms ranging from standard baselines like logistic regression and random forests to recent neural network architectures. Independent research groups have already contributed additional datasets and prediction algorithms to DeepChem. Adoption of common benchmarking evaluation metrics, datasets, and baseline algorithms has the potential to establish the practical utility of deep learning in chemical bioactivity prediction and lower the barrier to entry for machine learning researchers without biochemistry expertise.

+

One open question in ligand-based screening pertains to the benefits and limitations of transfer learning. Multi-task neural networks have shown the advantages of jointly modeling many targets [302,303]. Other studies have shown the limitations of transfer learning when the prediction tasks are insufficiently related [297,312]. This has important implications for representation learning. The typical approach to improve deep learning models by expanding the dataset size may not be applicable if only “related” tasks are beneficial, especially because task-task relatedness is ill-defined. The massive chemical state space will also influence the development of unsupervised representation learning methods [315]. Future work will establish whether it is better to train on massive collections of diverse compounds, drug-like small molecules, or specialized subsets.

+

Structure-based prediction of bioactivity

+

When protein structure is available, virtual screening has traditionally relied on docking programs to predict how a compound best fits in the target’s binding site and score the predicted ligand-target complex [316]. Recently, deep learning approaches have been developed to model protein structure, which is expected to improve upon the simpler drug-target interaction algorithms described above that represent proteins with feature vectors derived from amino acid sequences [39,292].

+

Structure-based deep learning methods differ in whether they use experimentally-derived or predicted ligand-target complexes and how they represent the 3D structure. The Atomic CNN [317] takes 3D crystal structures from PDBBind [318] as input, ensuring it uses a reliable ligand-target complex. AtomNet [36] samples multiple ligand poses within the target binding site, and DeepVS [319] and Ragoza et al. [320] use a docking program to generate protein-compound complexes. If they are sufficiently accurate, these latter approaches would have wider applicability to a much larger set of compounds and proteins. However, incorrect ligand poses will be misleading during training, and the predictive performance is sensitive to the docking quality [319].

+

There are two established options for representing a protein-compound complex. One option, a 3D grid, can featurize the input complex [36,320]. Each entry in the grid tracks the types of protein and ligand atoms in that region of the 3D space or descriptors derived from those atoms. Alternatively, DeepVS [319] and atomic convolutions [317] offer greater flexibility in their convolutions by eschewing the 3D grid. Instead, they each implement techniques for executing convolutions over atoms’ neighboring atoms in the 3D space. Gomes et al. demonstrate that currently random forest on a 1D feature vector that describes the 3D ligand-target structure generally outperforms neural networks on the same feature vector as well as atomic convolutions and ligand-based neural networks when predicting the continuous-valued inhibition constant on the PDBBind refined dataset [317]. However, in the long term, atomic convolutions may ultimately overtake grid-based methods, as they provide greater freedom to model atom-atom interactions and the forces that govern binding affinity.

+

De novo drug design

+

De novo drug design attempts to model the typical design-synthesize-test cycle of drug discovery [321,322]. It explores an estimated 1060 synthesizable organic molecules with drug-like properties without explicit enumeration [306]. To test or score structures, algorithms like those discussed earlier are used. To “design” and “synthesize”, traditional de novo design software relied on classical optimizers such as genetic algorithms. Unfortunately, this often leads to overfit, “weird” molecules, which are difficult to synthesize in the lab. Current programs have settled on rule-based virtual chemical reactions to generate molecular structures [322]. Deep learning models that generate realistic, synthesizable molecules have been proposed as an alternative. In contrast to the classical, symbolic approaches, generative models learned from data would not depend on laboriously encoded expert knowledge. The challenge of generating molecules has parallels to the generation of syntactically and semantically correct text [323].

+

As deep learning models that directly output (molecular) graphs remain under-explored, generative neural networks for drug design typically represent chemicals with the simplified molecular-input line-entry system (SMILES), a standard string-based representation with characters that represent atoms, bonds, and rings [324]. This allows treating molecules as sequences and leveraging recent progress in recurrent neural networks. Gómez-Bombarelli et al. designed a SMILES-to-SMILES autoencoder to learn a continuous latent feature space for chemicals [315]. In this learned continuous space it was possible to interpolate between continuous representations of chemicals in a manner that is not possible with discrete (e.g. bit vector or string) features or in symbolic, molecular graph space. Even more interesting is the prospect of performing gradient-based or Bayesian optimization of molecules within this latent space. The strategy of constructing simple, continuous features before applying supervised learning techniques is reminiscent of autoencoders trained on high-dimensional EHR data [83]. A drawback of the SMILES-to-SMILES autoencoder is that not all SMILES strings produced by the autoencoder’s decoder correspond to valid chemical structures. Recently, the Grammar Variational Autoencoder, which takes the SMILES grammar into account and is guaranteed to produce syntactically valid SMILES, has been proposed to alleviate this issue [325].

+

Another approach to de novo design is to train character-based RNNs on large collections of molecules, for example, ChEMBL [326], to first obtain a generic generative model for drug-like compounds [324]. These generative models successfully learn the grammar of compound representations, with 94% [327] or nearly 98% [324] of generated SMILES corresponding to valid molecular structures. The initial RNN is then fine-tuned to generate molecules that are likely to be active against a specific target by either continuing training on a small set of positive examples [324] or adopting reinforcement learning strategies [327,328]. Both the fine-tuning and reinforcement learning approaches can rediscover known, held-out active molecules. The great flexibility of neural networks, and progress in generative models offers many opportunities for deep architectures in de novo design (e.g. the adaptation of Generative Adversarial Networks (GANs) for molecules).

+

Discussion

+

Despite the disparate types of data and scientific goals in the learning tasks covered above, several challenges are broadly important for deep learning in the biomedical domain. Here we examine these factors that may impede further progress, ask what steps have already been taken to overcome them, and suggest future research directions.

+

Evaluation

+

There are unique challenges to evaluating deep learning predictions in the biomedical domain. We focus on TF binding prediction as a representative task to illustrate some of these issues. The human genome has 3 billion base pairs, and only a small fraction of them are implicated in specific biochemical activities. As a result, classification of genomic regions based on their biochemical activity results in highly imbalanced classification. Class imbalance also arises in other problems we review, such as virtual screening for drug discovery. What are appropriate evaluation metrics that account for the label imbalance? The classification labels are formulated based on continuous value experimental signals. Practitioners must determine an appropriate procedure for formulating binary classification labels based on these signals. In addition, the experimental signals are only partially reproducible across experimental replicates. An appropriate upper bound for classification performance must account for the experimental reproducibility.

+

Evaluation metrics for imbalanced classification

+

Less than 1% of the genome can be confidently labeled as bound for most transcription factors. Therefore, it is important to evaluate the genome-wide recall and false discovery rate (FDR) of classification models of biochemical activities. Targeted validation experiments of specific biochemical activities usually necessitate an FDR of 5%-25%. When predicted biochemical activities are used as features in other models, such as gene expression models, a low FDR may not be as critical if the downstream models can satisfy their evaluation criteria. An FDR of 50% in this context may suffice.

+

What is the correspondence between these metrics and commonly used classification metrics such as auPRC (area under the precision-recall curve) and auROC? auPRC evaluates the average precision, or equivalently, the average FDR across all recall thresholds. This metric provides an overall estimate of performance across all possible use cases, which can be misleading for targeted validation experiments. For example, classification of TF binding sites can exhibit a recall of 0% at 10% FDR and auPRC greater than 0.6. In this case, the auPRC may be competitive, but the predictions are ill-suited for targeted validation that can only examine a few of the highest-confidence predictions. Likewise, auROC evaluates the average recall across all false positive rate (FPR) thresholds, which is often a highly misleading metric in class-imbalanced domains [72,329]. For example, consider a classification model with recall of 0% at FDR less than 25% and 100% recall at FDR greater than 25%. In the context of TF binding predictions where only 1% of genomic regions are bound by the TF, this is equivalent to a recall of 100% for FPR greater than 0.33%. In other words, the auROC would be 0.9967, but the classifier would be useless for targeted validation. It is not unusual to obtain a chromosome-wide auROC greater than 0.99 for TF binding predictions but a recall of 0% at 10% FDR.

+

Formulation of classification labels

+

Genome-wide continuous signals are commonly formulated into classification labels through signal peak detection. ChIP-seq peaks are used to identify locations of TF binding and histone modifications. Such procedures rely on thresholding criteria to define what constitutes a peak in the signal. This inevitably results in a set of signal peaks that are close to the threshold, not sufficient to constitute a positive label but too similar to positively labeled examples to constitute a negative label. To avoid an arbitrary label for these example they may be labeled as “ambiguous”. Ambiguously labeled examples can then be ignored during model training and evaluation of recall and FDR. The correlation between model predictions on these examples and their signal values can be used to evaluate if the model correctly ranks these examples between positive and negative examples.

+

Formulation of a performance upper bound

+

Genome-wide signals across experiments can lead to different sets of positive examples. When experimental replicates do not completely agree, perfect recall at a low FDR is not possible. The upper bound on the recall is the fraction of positive examples that are in agreement across experiments. This fraction will vary depending on the available experimental data. Reproducibility for experimental replicates from the same lab is typically higher than experimental replicates across multiple labs. One way to handle the range of reproducibility is the use of multiple reproducibility criteria such as reproducibility across technical replicates, biological replicates from the same lab, and biological replicates from multiple labs.

+

Interpretation

+

As deep learning models achieve state-of-the-art performance in a variety of domains, there is a growing need to make the models more interpretable. Interpretability matters for two main reasons. First, a model that achieves breakthrough performance may have identified patterns in the data that practitioners in the field would like to understand. However, this would not be possible if the model is a black box. Second, interpretability is important for trust. If a model is making medical diagnoses, it is important to ensure the model is making decisions for reliable reasons and is not focusing on an artifact of the data. A motivating example of this can be found in Ba and Caruana [330], where a model trained to predict the likelihood of death from pneumonia assigned lower risk to patients with asthma, but only because such patients were treated as higher priority by the hospital. In the context of deep learning, understanding the basis of a model’s output is particularly important as deep learning models are unusually susceptible to adversarial examples [331] and can output confidence scores over 99.99% for samples that resemble pure noise.

+

As the concept of interpretability is quite broad, many methods described as improving the interpretability of deep learning models take disparate and often complementary approaches.

+

Assigning example-specific importance scores

+

Several approaches ascribe importance on an example-specific basis to the parts of the input that are responsible for a particular output. These can be broadly divided into perturbation-based approaches and backpropagation-based approaches.

+

Pertubration-based approaches change parts of the input and observe the impact on the output of the network. Alipanahi et al. [166] and Zhou & Troyanskaya [170] scored genomic sequences by introducing virtual mutations at individual positions in the sequence and quantifying the change in the output. Umarov et al. [174] used a similar strategy, but with sliding windows where the sequence within each sliding window was substituted with a random sequence. Kelley et al. [179] inserted known protein-binding motifs into the centers of sequences and assessed the change in predicted accessibility. Ribeiro et al. [332] introduced LIME, which constructs a linear model to locally approximate the output of the network on perturbed versions of the input and assigns importance scores accordingly. For analyzing images, Zeiler and Fergus [333] applied constant-value masks to different input patches. More recently, marginalizing over the plausible values of an input has been suggested as a way to more accurately estimate contributions [334].

+

A common drawback to perturbation-based approaches is computational efficiency: each perturbed version of an input requires a separate forward propagation through the network to compute the output. As noted by Shrikumar et al. [171], such methods may also underestimate the impact of features that have saturated their contribution to the output, as can happen when multiple redundant features are present. To reduce the computational overhead of perturbation-based approaches, Fong and Vedaldi [335] solve an optimization problem using gradient descent to discover a minimal subset of inputs to perturb in order to decrease the predicted probability of a selected class. Their method converges in many fewer iterations but requires the perturbation to have a differentiable form.

+

Backpropagation-based methods, in which the signal from a target output neuron is propagated backwards to the input layer, are another way to interpret deep networks that sidestep inefficiencies of the perturbastion-basd methods. A classic example of this is calculating the gradients of the output with respect to the input [336] to compute a “saliency map”. Bach et al. [337] proposed a strategy called Layerwise Relevance Propagation, which was shown to be equivalent to the element-wise product of the gradient and input [171,338]. Networks with Rectified Linear Units (ReLUs) create nonlinearities that must be addressed. Several variants exist for handling this [333,339]. Backpropagation-based methods are a highly active area of research. Researchers are still actively identifying weaknesses [340], and new methods are being developed to address them [171,341,342]. Lundberg and Lee [343] noted that several importance scoring methods including integrated gradients and LIME could all be considered approximations to Shapely values [344], which have a long history in game theory for assigning contributions to players in cooperative games.

+

Matching or exaggerating the hidden representation

+

Another approach to understanding the network’s predictions is to find artificial inputs that produce similar hidden representations to a chosen example. This can elucidate the features that the network uses for prediction and drop the features that the network is insensitive to. In the context of natural images, Mahendran and Vedaldi [345] introduced the “inversion” visualization, which uses gradient descent and backpropagation to reconstruct the input from its hidden representation. The method required placing a prior on the input to favor results that resemble natural images. For genomic sequence, Finnegan and Song [346] used a Markov chain Monte Carlo algorithm to find the maximum-entropy distribution of inputs that produced a similar hidden representation to the chosen input.

+

A related idea is “caricaturization”, where an initial image is altered to exaggerate patterns that the network searches for [347]. This is done by maximizing the response of neurons that are active in the network, subject to some regularizing constraints. Mordvintsev et al. [348] leveraged caricaturization to generate aesthetically pleasing images using neural networks.

+

Activation maximization

+

Activation maximization can reveal patterns detected by an individual neuron in the network by generating images which maximally activate that neuron, subject to some regularizing constraints. This technique was first introduced in Ehran et al. [349] and applied in subsequent work [336,347,348,350]. Lanchantin et al. [167] applied activation maximization to genomic sequence data. One drawback of this approach is that neural networks often learn highly distributed representations where several neurons cooperatively describe a pattern of interest. Thus, visualizing patterns learned by individual neurons may not always be informative.

+

RNN-specific approaches

+

Several interpretation methods are specifically tailored to recurrent neural network architectures. The most common form of interpretability provided by RNNs is through attention mechanisms, which have been used in diverse problems such as image captioning and machine translation to select portions of the input to focus on generating a particular output [351,352]. Deming et al. [353] applied the attention mechanism to models trained on genomic sequence. Attention mechanisms provide insight into the model’s decision-making process by revealing which portions of the input are used by different outputs. In the clinical domain, Choi et al. [354] leveraged attention mechanisms to highlight which aspects of a patient’s medical history were most relevant for making diagnoses. Choi et al. [355] later extended this work to take into account the structure of disease ontologies and found that the concepts represented by the model aligned with medical knowledge. Note that interpretation strategies that rely on an attention mechanism do not provide insight into the logic used by the attention layer.

+

Visualizing the activation patterns of the hidden state of a recurrent neural network can also be instructive. Early work by Ghosh and Karamcheti [356] used cluster analysis to study hidden states of comparatively small networks trained to recognize strings from a finite state machine. More recently, Karpathy et al. [357] showed the existence of individual cells in LSTMs that kept track of quotes and brackets in character-level language models. To facilitate such analyses, LSTMVis [358] allows interactive exploration of the hidden state of LSTMs on different inputs.

+

Another strategy, adopted by Lanchatin et al. [167] looks at how the output of a recurrent neural network changes as longer and longer subsequences are supplied as input to the network, where the subsequences begin with just the first position and end with the entire sequence. In a binary classification task, this can identify those positions which are responsible for flipping the output of the network from negative to positive. If the RNN is bidirectional, the same process can be repeated on the reverse sequence. As noted by the authors, this approach was less effective at identifying motifs compared to the gradient-based backpropagation approach of Simonyan et al. [336], illustrating the need for more sophisticated strategies to assign importance scores in recurrent neural networks.

+

Murdoch and Szlam [359] showed that the output of an LSTM can be decomposed into a product of factors, where each factor can be interpreted as the contribution at a particular timestep. The contribution scores were then used to identify key phrases from a model trained for sentiment analysis and obtained superior results compared to scores derived via a gradient-based approach.

+

Miscellaneous approaches

+

Toward quantifying the uncertainty of predictions, there has been a renewed interest in confidence intervals for deep neural networks. Early work from Chryssolouris et al. [360] provided confidence intervals under the assumption of normally-distributed error. A more recent technique known as test-time dropout [361] can also be used to obtain a probabilistic interpretation of a network’s outputs.

+

It can often be informative to understand how the training data affects model learning. Toward this end, Koh and Liang [362] used influence functions, a technique from robust statistics, to trace a model’s predictions back through the learning algorithm to identify the datapoints in the training set that had the most impact on a given prediction. A more free-form approach to interpretability is to visualize the activation patterns of the network on individual inputs and on subsets of the data. ActiVis and CNNvis [363,364] are two frameworks that enable interactive visualization and exploration of large-scale deep learning models. An orthogonal strategy is to use a knowledge distillation approach to replace a deep learning model with a more interpretable model that achieves comparable performance. Towards this end, Che et al. [365] used gradient boosted trees to learn interpretable healthcare features from trained deep models.

+

Finally, it is sometimes possible to train the model to provide justifications for its predictions. Lei et al. [366] used a generator to identify “rationales”, which are short and coherent pieces of the input text that produce similar results to the whole input when passed through an encoder. The authors applied their approach to a sentiment analysis task and obtained substantially superior results compared to an attention-based method.

+

Future outlook

+

While deep learning lags behind most Bayesian models in terms of interpretability, the interpretability of deep learning is comparable to or exceeds that of many other widely-used machine learning methods such as random forests or SVMs. While it is possible to obtain importance scores for different inputs in a random forest, the same is true for deep learning. Similarly, SVMs trained with a nonlinear kernel are not easily interpretable because the use of the kernel means that one does not obtain an explicit weight matrix. Finally, it is worth noting that some simple machine learning methods are less interpretable in practice than one might expect. A linear model trained on heavily engineered features might be difficult to interpret as the input features themselves are difficult to interpret. Similarly, a decision tree with many nodes and branches may also be difficult for a human to make sense of.

+

There are several directions that might benefit the development of interpretability techniques. The first is the introduction of gold standard benchmarks that different interpretability approaches could be compared against, similar in spirit to how datasets like ImageNet and CIFAR spurred the development of deep learning for computer vision. It would also be helpful if the community placed more emphasis on domains outside of computer vision. Computer vision is often used as the example application of interpretability methods, but it is not the domain with the most pressing need. Finally, closer integration of interpretability approaches with popular deep learning frameworks would make it easier for practitioners to apply and experiment with different approaches to understanding their deep learning models.

+

Data limitations

+

A lack of large-scale, high-quality, correctly labeled training data has impacted deep learning in nearly all applications we have discussed. The challenges of training complex, high-parameter neural networks from few examples are obvious, but uncertainty in the labels of those examples can be just as problematic. In genomics labeled data may be derived from an experimental assay with known and unknown technical artifacts, biases, and error profiles. It is possible to weight training examples or construct Bayesian models to account for uncertainty or non-independence in the data, as described in the TF binding example above. As another example, Park et al. [367] estimated shared non-biological signal between datasets to correct for non-independence related to assay platform or other factors in a Bayesian integration of many datasets. However, such techniques are rarely placed front and center in any description of methods and may be easily overlooked.

+

For some types of data, especially images, it is straightforward to augment training datasets by splitting a single labeled example into multiple examples. For example, an image can easily be rotated, flipped, or translated and retain its label [60]. 3D MRI and 4D fMRI (with time as a dimension) data can be decomposed into sets of 2D images [368]. This can greatly expand the number of training examples but artificially treats such derived images as independent instances and sacrifices the structure inherent in the data. CellCnn trains a model to recognize rare cell populations in single-cell data by creating training instances that consist of subsets of cells that are randomly sampled with replacement from the full dataset [221].

+

Simulated or semi-synthetic training data has been employed in multiple biomedical domains, though many of these ideas are not specific to deep learning. Training and evaluating on simulated data, for instance, generating synthetic TF binding sites with position weight matrices [169] or RNA-seq reads for predicting mRNA transcript boundaries [369], is a standard practice in bioinformatics. This strategy can help benchmark algorithms when the available gold standard dataset is imperfect, but it should be paired with an evaluation on real data, as in the prior examples [169,369]. In rare cases, models trained on simulated data have been successfully applied directly to real data [369].

+

Data can be simulated to create negative examples when only positive training instances are available. DANN [35] adopts this approach to predict the pathogenicity of genetic variants using semi-synthetic training data from Combined Annotation-Dependent Depletion (CADD) [370]. Though our emphasis here is on the training strategy, it should be noted that logistic regression outperformed DANN when distinguishing known pathogenic mutations from likely benign variants in real data. Similarly, a somatic mutation caller has been trained by injecting mutations into real sequencing datasets [261]. This method detected mutations in other semi-synthetic datasets but was not validated on real data.

+

In settings where the experimental observations are biased toward positive instances, such as MHC protein and peptide ligand binding affinity [371], or the negative instances vastly outnumber the positives, such as high-throughput chemical screening [307], training datasets have been augmented by adding additional instances and assuming they are negative. There is some evidence that this can improve performance [307], but in other cases it was only beneficial when the real training datasets were extremely small [371]. Overall, training with simulated and semi-simulated data is a valuable idea for overcoming limited sample sizes but one that requires more rigorous evaluation on real ground-truth datasets before we can recommend it for widespread use. There is a risk that a model will easily discriminate synthetic examples but not generalize to real data.

+

Multimodal, multi-task, and transfer learning, discussed in detail below, can also combat data limitations to some degree. There are also emerging network architectures, such as Diet Networks for high-dimensional SNP data [372]. These use multiple networks to drastically reduce the number of free parameters by first flipping the problem and training a network to predict parameters (weights) for each input (SNP) to learn a feature embedding. This embedding (e.g. from principal component analysis, per class histograms, or a Word2vec [78] generalization) can be learned directly from input data or take advantage of other datasets or domain knowledge. Additionally, in this task the features are the examples, an important advantage when it is typical to have 500 thousand or more SNPs and only a few thousand patients. Finally, this embedding is of a much lower dimension, allowing for a large reduction in the number of free parameters. In the example given, the number of free parameters was reduced from 30 million to 50 thousand, a factor of 600.

+

Hardware limitations and scaling

+

Efficiently scaling deep learning is challenging, and there is a high computational cost (e.g. time, memory, and energy) associated with training neural networks and using them to make predictions. This is one of the reasons why neural networks have only recently found widespread use [373].

+

Many have sought to curb these costs, with methods ranging from the very applied (e.g. reduced numerical precision [374–377]) to the exotic and theoretic ( e.g. training small networks to mimic large networks and ensembles [330,378]). The largest gains in efficiency have come from computation with graphics processing units (GPUs) [373,379–383], which excel at the matrix and vector operations so central to deep learning. The massively parallel nature of GPUs allows additional optimizations, such as accelerated mini-batch gradient descent [380,381,384,385]. However, GPUs also have limited memory, making networks of useful size and complexity difficult to implement on a single GPU or machine [69,379]. This restriction has sometimes forced computational biologists to use workarounds or limit the size of an analysis. Chen et al. [148] inferred the expression level of all genes with a single neural network, but due to memory restrictions they randomly partitioned genes into two separately analyzed halves. In other cases, researchers limited the size of their neural network [29] or the total number of training instances [315]. Some have also chosen to use standard central processing unit (CPU) implementations rather than sacrifice network size or performance [386].

+

While steady improvements in GPU hardware may alleviate this issue, it is unclear whether advances will occur quickly enough to keep pace with the growing biological datasets and increasingly complex neural networks. Much has been done to minimize the memory requirements of neural networks [330,374–377,387,388], but there is also growing interest in specialized hardware, such as field-programmable gate arrays (FPGAs) [383,389] and application-specific integrated circuits (ASICs) [390]. Less software is available for such highly specialized hardware [389]. But specialized hardware promises improvements in deep learning at reduced time, energy, and memory [383]. Specialized hardware may be a difficult investment for those not solely interested in deep learning, but for those with a deep learning focus these solutions may become popular.

+

Distributed computing is a general solution to intense computational requirements and has enabled many large-scale deep learning efforts. Some types of distributed computation [391,392] are not suitable for deep learning [393], but much progress has been made. There now exist a number of algorithms [376,393,394], tools [395–397], and high-level libraries [398,399] for deep learning in a distributed environment, and it is possible to train very complex networks with limited infrastructure [400]. Besides handling very large networks, distributed or parallelized approaches offer other advantages, such as improved ensembling [401] or accelerated hyperparameter optimization [402,403].

+

Cloud computing, which has already seen wide adoption in genomics [404], could facilitate easier sharing of the large datasets common to biology [405,406], and may be key to scaling deep learning. Cloud computing affords researchers flexibility, and enables the use of specialized hardware (e.g. FPGAs, ASICs, GPUs) without major investment. As such, it could be easier to address the different challenges associated with the multitudinous layers and architectures available [407]. Though many are reluctant to store sensitive data (e.g. patient electronic health records) in the cloud, secure, regulation-compliant cloud services do exist [408].

+

Data, code, and model sharing

+

A robust culture of data, code, and model sharing would speed advances in this domain. The cultural barriers to data sharing in particular are perhaps best captured by the use of the term “research parasite” to describe scientists who use data from other researchers [409]. A field that honors only discoveries and not the hard work of generating useful data will have difficulty encouraging scientists to share their hard-won data. It’s precisely those data that would help to power deep learning in the domain. Efforts are underway to recognize those who promote an ecosystem of rigorous sharing and analysis [410].

+

The sharing of high-quality, labeled datasets will be especially valuable. In addition, researchers who invest time to preprocess datasets to be suitable for deep learning can make the preprocessing code (e.g. Basset [179] and variationanalysis [259]) and cleaned data (e.g. MoleculeNet [313]) publicly available to catalyze further research. However, there are complex privacy and legal issues involved in sharing patient data. In some domains high-quality training data has been generated privately, i.e. high-throughput chemical screening data at pharmaceutical companies. One may think that there is little incentive for this private data to be shared. However, data are not inherently valuable. Instead, the insights that we glean from them are where the value lies. Organizations may establish a competitive advantage by releasing data sufficient for improved methods to be developed.

+

Code sharing and open source licensing is essential for continued progress in this domain. We strongly advocate following established best practices for sharing source code, archiving code in repositories that generate digital object identifiers, and open licensing [411] regardless of the minimal requirements, or lack thereof, set by journals, conferences, or preprint servers. In addition, it is important for authors to share not only code for their core models but also scripts and code used for data cleaning (see above) and hyperparameter optimization. These improve reproducibility and serve as documentation of the detailed decisions that impact model performance but may not be exhaustively captured in a manuscript’s methods text.

+

Because many deep learning models are often built using one of several popular software frameworks, it is also possible to directly share trained predictive models. The availability of pre-trained models can accelerate research, with image classifiers as an apt example. A pre-trained neural network can be quickly fine-tuned on new data and used in transfer learning, as discussed below. Taking this idea to the extreme, genomic data has been artificially encoded as images in order to benefit from pre-trained image classifiers [258]. “Model zoos” – collections of pre-trained models – are not yet common in biomedical domains but have started to appear in genomics applications [217,412]. Sharing models for patient data requires great care because deep learning models can be attacked to identify examples used in training. We discuss this issue as well as recent techniques to mitigate these concerns in the patient categorization section.

+

DeepChem [312–314] and DragoNN [412] exemplify the benefits of sharing pre-trained models and code under an open source license. DeepChem, which targets drug discovery and quantum chemistry, has actively encouraged and received community contributions of learning algorithms and benchmarking datasets. As a consequence, it now supports a large suite of machine learning approaches, both deep learning and competing strategies, that can be run on diverse test cases. This realistic, continual evaluation will play a critical role in assessing which techniques are most promising for chemical screening and drug discovery. Like formal, organized challenges such as the ENCODE-DREAM in vivo Transcription Factor Binding Site Prediction Challenge [413], DeepChem provides a forum for the fair, critical evaluations that are not always conducted in individual methodological papers, which can be biased toward favoring a new proposed algorithm. Likewise DragoNN (Deep RegulAtory GenOmic Neural Networks) offers not only code and a model zoo but also a detailed tutorial and partner package for simulating training data. These resources, especially the ability to simulate datasets that are sufficiently complex to demonstrate the challenges of training neural networks but small enough to train quickly on a CPU, are important for training students and attracting machine learning researchers to problems in genomics and healthcare.

+

Multimodal, multi-task, and transfer learning

+

The fact that biomedical datasets often contain a limited number of instances or labels can cause poor performance of deep learning algorithms. These models are particularly prone to overfitting due to their high representational power. However, transfer learning techniques, also known as domain adaptation, enable transfer of extracted patterns between different datasets and even domains. This approach consists of training a model for the base task and subsequently reusing the trained model for the target problem. The first step allows a model to take advantage of a larger amount of data and/or labels to extract better feature representations. Transferring learned features in deep neural networks improves performance compared to randomly initialized features even when pre-training and target sets are dissimilar. However, transferability of features decreases as the distance between the base task and target task increases [414].

+

In image analysis, previous examples of deep transfer learning applications proved large-scale natural image sets [43] to be useful for pre-training models that serve as generic feature extractors for various types of biological images [15,207,415,416]. More recently, deep learning models predicted protein sub-cellular localization for proteins not originally present in a training set [417]. Moreover, learned features performed reasonably well even when applied to images obtained using different fluorescent labels, imaging techniques, and different cell types [418]. However, there are no established theoretical guarantees for feature transferability between distant domains such as natural images and various modalities of biological imaging. Because learned patterns are represented in deep neural networks in a layer-wise hierarchical fashion, this issue is usually addressed by fixing an empirically chosen number of layers that preserve generic characteristics of both training and target datasets. The model is then fine-tuned by re-training top layers on the specific dataset in order to re-learn domain-specific high level concepts (e.g. fine-tuning for radiology image classification [55]). Fine-tuning on specific biological datasets enables more focused predictions.

+

In genomics, the Basset package [179] for predicting chromatin accessibility was shown to rapidly learn and accurately predict on new data by leveraging a model pre-trained on available public data. To simulate this scenario, authors put aside 15 of 164 cell type datasets and trained the Basset model on the remaining 149 datasets. Then, they fine-tuned the model with one training pass of each of the remaining datasets and achieved results close to the model trained on all 164 datasets together. In another example, Min et al. [180] demonstrated how training on the experimentally-validated FANTOM5 permissive enhancer dataset followed by fine-tuning on ENCODE enhancer datasets improved cell type-specific predictions, outperforming state-of-the-art results. In drug design, general RNN models trained to generate molecules from the ChEMBL database have been fine-tuned to produce drug-like compounds for specific targets [324,327].

+

Related to transfer learning, multimodal learning assumes simultaneous learning from various types of inputs, such as images and text. It can capture features that describe common concepts across input modalities. Generative graphical models like RBMs, deep Boltzmann machines, and DBNs, demonstrate successful extraction of more informative features for one modality (images or video) when jointly learned with other modalities (audio or text) [419]. Deep graphical models such as DBNs are well-suited for multimodal learning tasks because they learn a joint probability distribution from inputs. They can be pre-trained in an unsupervised fashion on large unlabeled data and then fine-tuned on a smaller number of labeled examples. When labels are available, convolutional neural networks are ubiquitously used because they can be trained end-to-end with backpropagation and demonstrate state-of-the-art performance in many discriminative tasks [15].

+

Jha et al. [156] showed that integrated training delivered better performance than individual networks. They compared a number of feed-forward architectures trained on RNA-seq data with and without an additional set of CLIP-seq, knockdown, and over-expression based input features. The integrative deep model generalized well for combined data, offering a large performance improvement for alternative splicing event estimation. Chaudhary et al. [420] trained a deep autoencoder model jointly on RNA-seq, miRNA-seq, and methylation data from The Cancer Genome Atlas to predict survival subgroups of hepatocellular carcinoma patients. This multimodal approach that treated different omic data types as different modalities outperformed both traditional methods (principal component analysis) and single-omic models. Interestingly, multi-omic model performance did not improve when combined with clinical information, suggesting that the model was able to capture redundant contributions of clinical features through their correlated genomic features. Chen et al. [143] used deep belief networks to learn phosphorylation states of a common set of signaling proteins in primary cultured bronchial cells collected from rats and humans treated with distinct stimuli. By interpreting species as different modalities representing similar high-level concepts, they showed that DBNs were able to capture cross-species representation of signaling mechanisms in response to a common stimuli. Another application used DBNs for joint unsupervised feature learning from cancer datasets containing gene expression, DNA methylation, and miRNA expression data [150]. This approach allowed for the capture of intrinsic relationships in different modalities and for better clustering performance over conventional k-means.

+

Multimodal learning with CNNs is usually implemented as a collection of individual networks in which each learns representations from single data type. These individual representations are further concatenated before or within fully-connected layers. FIDDLE [421] is an example of a multimodal CNN that represents an ensemble of individual networks that take NET-seq, MNase-seq, ChIP-seq, RNA-seq, and raw DNA sequence as input to predict transcription start sites. The combined model radically improves performance over separately trained datatype-specific networks, suggesting that it learns the synergistic relationship between datasets.

+

Multi-task learning is an approach related to transfer learning. In a multi-task learning framework, a model learns a number of tasks simultaneously such that features are shared across them. DeepSEA [170] implemented multi-task joint learning of diverse chromatin factors from raw DNA sequence. This allowed a sequence feature that was effective in recognizing binding of a specific TF to be simultaneously used by another predictor for a physically interacting TF. Similarly, TFImpute [157] learned information shared across transcription factors and cell lines to predict cell-specific TF binding for TF-cell line combinations. Yoon et al. [77] demonstrated that predicting the primary cancer site from cancer pathology reports together with its laterality substantially improved the performance for the latter task, indicating that multi-task learning can effectively leverage the commonality between two tasks using a shared representation. Many studies employed multi-task learning to predict chemical bioactivity [300,303] and drug toxicity [304,422]. Kearnes et al. [297] systematically compared single-task and multi-task models for ADMET properties and found that multi-task learning generally improved performance. Smaller datasets tended to benefit more than larger datasets.

+

Multi-task learning is complementary to multimodal and transfer learning. All three techniques can be used together in the same model. For example, Zhang et al. [415] combined deep model-based transfer and multi-task learning for cross-domain image annotation. One could imagine extending that approach to multimodal inputs as well. A common characteristic of these methods is better generalization of extracted features at various hierarchical levels of abstraction, which is attained by leveraging relationships between various inputs and task objectives.

+

Despite demonstrated improvements, transfer learning approaches pose challenges. There are no theoretically sound principles for pre-training and fine-tuning. Best practice recommendations are heuristic and must account for additional hyper-parameters that depend on specific deep architectures, sizes of the pre-training and target datasets, and similarity of domains. However, similarity of datasets and domains in transfer learning and relatedness of tasks in multi-task learning is difficult to access. Most studies address these limitations by empirical evaluation of the model. Unfortunately, negative results are typically not reported. Rajkomar et al. [55] showed that a deep CNN trained on natural images can boost radiology image classification performance. However, due to differences in imaging domains, the target task required either re-training the initial model from scratch with special pre-processing or fine-tuning of the whole network on radiographs with heavy data augmentation to avoid overfitting. Exclusively fine-tuning top layers led to much lower validation accuracy (81.4 versus 99.5). Fine-tuning the aforementioned Basset model with more than one pass resulted in overfitting [179]. DeepChem successfully improved results for low-data drug discovery with one-shot learning for related tasks. However, it clearly demonstrated the limitations of cross-task generalization across unrelated tasks in one-shot models, specifically nuclear receptor assays and patient adverse reactions [312].

+

In the medical domain, multimodal, multi-task and transfer learning strategies not only inherit most methodological issues from natural image, text, and audio domains, but also pose domain-specific challenges. There is a compelling need for the development of privacy-preserving transfer learning algorithms, such as Private Aggregation of Teacher Ensembles [124]. We suggest that these types of models deserve deeper investigation to establish sound theoretical guarantees and determine limits for the transferability of features between various closely related and distant learning tasks.

+

Conclusions

+

Deep learning-based methods now match or surpass the previous state of the art in a diverse array of tasks in patient and disease categorization, fundamental biological study, genomics, and treatment development. Returning to our central question: given this rapid progress, has deep learning transformed the study of human disease? Though the answer is highly dependent on the specific domain and problem being addressed, we conclude that deep learning has not yet realized its transformative potential or induced a strategic inflection point. Despite its dominance over competing machine learning approaches in many of the areas reviewed here and quantitative improvements in predictive performance, deep learning has not yet definitively “solved” these problems.

+

As an analogy, consider recent progress in conversational speech recognition. Since 2009 there have been drastic performance improvements with error rates dropping from more than 20% to less than 6% [423] and finally approaching or exceeding human performance in the past year [424,425]. The phenomenal improvements on benchmark datasets are undeniable, but greatly reducing the error rate on these benchmarks did not fundamentally transform the domain. Widespread adoption of conversational speech technologies will require solving the problem, i.e. methods that surpass human performance, and persuading users to adopt them [423]. We see parallels in healthcare, where achieving the full potential of deep learning will require outstanding predictive performance as well as acceptance and adoption by biologists and clinicians. These experts will rightfully demand rigorous evidence that deep learning has impacted their respective disciplines – elucidated new biological mechanisms and improved patient outcomes – to be convinced that the promises of deep learning are more substantive than those of previous generations of artificial intelligence.

+

Some of the areas we have discussed are closer to surpassing this lofty bar than others, generally those that are more similar to the non-biomedical tasks that are now monopolized by deep learning. In medical imaging, diabetic retinopathy [47], diabetic macular edema [47], tuberculosis [56], and skin lesion [4] classifiers are highly accurate and comparable to clinician performance.

+

In other domains, perfect accuracy will not be required because deep learning will primarily prioritize experiments and assist discovery. For example, in chemical screening for drug discovery, a deep learning system that successfully identifies dozens or hundreds of target-specific, active small molecules from a massive search space would have immense practical value even if its overall precision is modest. In medical imaging, deep learning can point an expert to the most challenging cases that require manual review [56], though the risk of false negatives must be addressed. In protein structure prediction, errors in individual residue-residue contacts can be tolerated when using the contacts jointly for 3D structure modeling. Improved contact map predictions [29] have led to notable improvements in fold and 3D structure prediction for some of the most challenging proteins, such as membrane proteins [202].

+

Conversely, the most challenging tasks may be those in which predictions are used directly for downstream modeling or decision-making, especially in the clinic. As an example, errors in sequence variant calling will be amplified if they are used directly for GWAS. In addition, the stochasticity and complexity of biological systems implies that for some problems, for instance predicting gene regulation in disease, perfect accuracy will be unattainable.

+

We are witnessing deep learning models achieving human-level performance across a number of biomedical domains. However, machine learning algorithms, including deep neural networks, are also prone to mistakes that humans are much less likely to make, such as misclassification of adversarial examples [426,427], a reminder that these algorithms do not understand the semantics of the objects presented. It may be impossible to guarantee that a model is not susceptible to adversarial examples, but work in this area is continuing [428,429]. Cooperation between human experts and deep learning algorithms addresses many of these challenges and can achieve better performance than either individually [67]. For sample and patient classification tasks, we expect deep learning methods to augment clinicians and biomedical researchers.

+

We are optimistic about the future of deep learning in biology and medicine. It is by no means inevitable that deep learning will revolutionize these domains, but given how rapidly the field is evolving, we are confident that its full potential in biomedicine has not been explored. We have highlighted numerous challenges beyond improving training and predictive accuracy, such as preserving patient privacy and interpreting models. Ongoing research has begun to address these problems and shown that they are not insurmountable. Deep learning offers the flexibility to model data in its most natural form, for example, longer DNA sequences instead of k-mers for transcription factor binding prediction and molecular graphs instead of pre-computed bit vectors for drug discovery. These flexible input feature representations have spurred creative modeling approaches that would be infeasible with other machine learning techniques. Unsupervised methods are currently less-developed than their supervised counterparts, but they may have the most potential because of how expensive and time-consuming it is to label large amounts of biomedical data. If future deep learning algorithms can summarize very large collections of input data into interpretable models that spur scientists to ask questions that they did not know how to ask, it will be clear that deep learning has transformed biology and medicine.

+

Methods

+

Continuous collaborative manuscript drafting

+

We recognized that deep learning in precision medicine is a rapidly developing area. Hence, diverse expertise was required to provide a forward-looking perspective. Accordingly, we collaboratively wrote this review in the open, enabling anyone with expertise to contribute. We wrote the manuscript in markdown and tracked changes using git. Contributions were handled through GitHub, with individuals submitting “pull requests” to suggest additions to the manuscript.

+

To facilitate citation, we defined a markdown citation syntax. We supported citations to the following identifier types (in order of preference): DOIs, PubMed IDs, arXiv IDs, and URLs. References were automatically generated from citation metadata by querying APIs to generate Citation Style Language (CSL) JSON items for each reference. Pandoc and pandoc-citeproc converted the markdown to HTML and PDF, while rendering the formatted citations and references. In total, referenced works consisted of 280 DOIs, 5 PubMed records, 108 arXiv manuscripts, and 40 URLs (webpages as well as manuscripts lacking standardized identifiers).

+

We implemented continuous analysis so the manuscript was automatically regenerated whenever the source changed [115]. We configured Travis CI – a continuous integration service – to fetch new citation metadata and rebuild the manuscript for every commit. Accordingly, formatting or citation errors in pull requests would cause the Travis CI build to fail, automating quality control. In addition, the build process renders templated variables, such as the reference counts mentioned above, to automate the updating of dynamic content. When contributions were merged into the master branch, Travis CI deployed the built manuscript by committing back to the GitHub repository. As a result, the latest manuscript version is always available at https://greenelab.github.io/deep-review. To ensure a consistent software environment, we defined a versioned conda environment of the software dependencies.

+

In addition, we instructed the Travis CI deployment script to perform blockchain timestamping [430,431]. Using OpenTimestamps, we submitted hashes for the manuscript and the source git commit for timestamping in the Bitcoin blockchain [432]. These timestamps attest that a given version of this manuscript (and its history) existed at a given point in time. The ability to irrefutably prove manuscript existence at a past time could be important to establish scientific precedence and enforce an immutable record of authorship.

+

Author contributions

+

We created an open repository on the GitHub version control platform (greenelab/deep-review) [433]. Here, we engaged with numerous authors from papers within and outside of the area. The manuscript was drafted via GitHub commits by 27 individuals who met the ICMJE standards of authorship. These were individuals who contributed to the review of the literature; drafted the manuscript or provided substantial critical revisions; approved the final manuscript draft; and agreed to be accountable in all aspects of the work. Individuals who did not contribute in all of these ways, but who did participate, are acknowledged below. We grouped authors into the following four classes of approximately equal contributions and randomly ordered authors within each contribution class. Drafted multiple sub-sections along with extensive editing, pull request reviews, or discussion: A.A.K., B.K.B., B.T.D., D.S.H., E.F., G.P.W., P.A., T.C. Drafted one or more sub-sections: A.E.C., A.S., B.J.L., E.M.C., G.L.R., J.I., J.L., J.X., S.W., W.X. Revised specific sub-sections or supervised drafting one or more sub-sections: A.K., D.D., D.J.H., L.K.W., M.H.S.S., Y.P., Y.Q. Drafted sub-sections, edited the manuscript, reviewed pull requests, and coordinated co-authors: A.G., C.S.G.

+

Competing interests

+

A.K. is on the Advisory Board of Deep Genomics Inc. E.F. is a full-time employee of GlaxoSmithKline. The remaining authors have no competing interests to declare.

+

Acknowledgements

+

We gratefully acknowledge Christof Angermueller, Kumardeep Chaudhary, Gökcen Eraslan, Michael M. Hoffman, Mikael Huss, Bharath Ramsundar and Xun Zhu for their discussion of the manuscript and reviewed papers on GitHub. We would like to thank Zhiyong Lu for revisions to the text that were not captured on GitHub as well as GitHub users aaronsheldon and swamidass who contributed text but did not formally approve the manuscript. Finally, we acknowledge funding from the Gordon and Betty Moore Foundation awards GBMF4552 (C.S.G. and D.S.H.) and GBMF4563 (D.J.H.); the National Institutes of Health awards DP2GM123485 (A.K.), R01AI116794 (B.K.B.), R01GM089652 (A.E.C.), R01GM089753 (J.X.), T32GM007753 (B.T.D.), and U54AI117924 (A.G.); the National Science Foundation awards 1245632 (G.L.R.), 1531594 (E.M.C.), and 1564955 (J.X.); and the National Institutes of Health Intramural Research Program and National Library of Medicine (Y.P.).

+

References

+ +
+
+

1. Big Data: Astronomical or Genomical?
+Zachary D. Stephens, Skylar Y. Lee, Faraz Faghri, Roy H. Campbell, Chengxiang Zhai, Miles J. Efron, Ravishankar Iyer, Michael C. Schatz, Saurabh Sinha, Gene E. Robinson
+PLOS Biology (2015-07-07) https://doi.org/10.1371/journal.pbio.1002195

+
+
+

2. Deep learning
+Yann LeCun, Yoshua Bengio, Geoffrey Hinton
+Nature (2015-05-27) https://doi.org/10.1038/nature14539

+
+
+

3. Searching for exotic particles in high-energy physics with deep learning
+P. Baldi, P. Sadowski, D. Whiteson
+Nature Communications (2014-07-02) https://doi.org/10.1038/ncomms5308

+
+
+

4. Dermatologist-level classification of skin cancer with deep neural networks
+Andre Esteva, Brett Kuprel, Roberto A. Novoa, Justin Ko, Susan M. Swetter, Helen M. Blau, Sebastian Thrun
+Nature (2017-01-25) https://doi.org/10.1038/nature21056

+
+
+

5. Google’s Neural Machine Translation System: Bridging the Gap between Human and Machine Translation
+Yonghui Wu, Mike Schuster, Zhifeng Chen, Quoc V. Le, Mohammad Norouzi, Wolfgang Macherey, Maxim Krikun, Yuan Cao, Qin Gao, Klaus Macherey, … Jeffrey Dean
+arXiv (2016-09-26) https://arxiv.org/abs/1609.08144v2

+
+
+

6. A logical calculus of the ideas immanent in nervous activity
+Warren S. McCulloch, Walter Pitts
+The Bulletin of Mathematical Biophysics (1943-12) https://doi.org/10.1007/bf02478259

+
+
+

7. Analysis of a Four-Layer Series-Coupled Perceptron. II
+H. D. Block, B. W. Knight, F. Rosenblatt
+Reviews of Modern Physics (1962-01-01) https://doi.org/10.1103/revmodphys.34.135

+
+
+

8. Machine learning applications in genetics and genomics
+Maxwell W. Libbrecht, William Stafford Noble
+Nature Reviews Genetics (2015-05-07) https://doi.org/10.1038/nrg3920

+
+
+

9. Google Research Publication: Building High-level Features Using Large Scale Unsupervised Learning(2016-12-15) http://research.google.com/archive/unsupervised_icml2012.html

+
+
+

10. HOGWILD!: A Lock-Free Approach to Parallelizing Stochastic Gradient Descent
+Feng Niu, Benjamin Recht, Christopher Re, Stephen J. Wright
+arXiv (2011-06-28) https://arxiv.org/abs/1106.5730v2

+
+
+

11. Deep Learning
+Ian Goodfellow, Yoshua Bengio, Aaron Courville
+(2016) http://www.deeplearningbook.org/

+
+
+

12. Academy of Management: Andrew S. Grove
+Andrew S. Grove
+(1998-08-09) http://www.intel.com/pressroom/archive/speeches/ag080998.htm

+
+
+

13. Deep learning for regulatory genomics
+Yongjin Park, Manolis Kellis
+Nature Biotechnology (2015-08-07) https://doi.org/10.1038/nbt.3313

+
+
+

14. Applications of Deep Learning in Biomedicine
+Polina Mamoshina, Armando Vieira, Evgeny Putin, Alex Zhavoronkov
+Molecular Pharmaceutics (2016-05-02) https://doi.org/10.1021/acs.molpharmaceut.5b00982

+
+
+

15. Deep learning for computational biology
+Christof Angermueller, Tanel Pärnamaa, Leopold Parts, Oliver Stegle
+Molecular Systems Biology (2016-07) https://doi.org/10.15252/msb.20156651

+
+
+

16. Deep learning in bioinformatics
+Seonwoo Min, Byunghan Lee, Sungroh Yoon
+Briefings in Bioinformatics (2016-07-29) https://doi.org/10.1093/bib/bbw068

+
+
+

17. Computer vision for high content screening
+Oren Z. Kraus, Brendan J. Frey
+Critical Reviews in Biochemistry and Molecular Biology (2016-01-24) https://doi.org/10.3109/10409238.2015.1135868

+
+
+

18. Deep learning for healthcare: review, opportunities and challenges
+Riccardo Miotto, Fei Wang, Shuang Wang, Xiaoqian Jiang, Joel T. Dudley
+Briefings in Bioinformatics (2017-05-06) https://doi.org/10.1093/bib/bbx044

+
+
+

19. A Survey on Deep Learning in Medical Image Analysis
+Geert Litjens, Thijs Kooi, Babak Ehteshami Bejnordi, Arnaud Arindra Adiyoso Setio, Francesco Ciompi, Mohsen Ghafoorian, Jeroen A. W. M. van der Laak, Bram van Ginneken, Clara I. Sánchez
+arXiv (2017-02-19) https://arxiv.org/abs/1702.05747v2

+
+
+

20. Deep Learning in Drug Discovery
+Erik Gawehn, Jan A. Hiss, Gisbert Schneider
+Molecular Informatics (2015-12-30) https://doi.org/10.1002/minf.201501008

+
+
+

21. Deep learning for computational chemistry
+Garrett B. Goh, Nathan O. Hodas, Abhinav Vishnu
+Journal of Computational Chemistry (2017-03-08) https://doi.org/10.1002/jcc.24764

+
+
+

22. Virtual Screening: A Challenge for Deep Learning
+Javier Pérez-Sianes, Horacio Pérez-Sánchez, Fernando Díaz
+Advances in Intelligent Systems and Computing (2016) https://doi.org/10.1007/978-3-319-40126-3_2

+
+
+

23. A renaissance of neural networks in drug discovery
+Igor I. Baskin, David Winkler, Igor V. Tetko
+Expert Opinion on Drug Discovery (2016-07-04) https://doi.org/10.1080/17460441.2016.1201262

+
+
+

24. Supervised Risk Predictor of Breast Cancer Based on Intrinsic Subtypes
+Joel S. Parker, Michael Mullins, Maggie C.U. Cheang, Samuel Leung, David Voduc, Tammi Vickery, Sherri Davies, Christiane Fauron, Xiaping He, Zhiyuan Hu, … Philip S. Bernard
+Journal of Clinical Oncology (2009-03-10) https://doi.org/10.1200/jco.2008.18.1370

+
+
+

25. New Strategies for Triple-Negative Breast Cancer–Deciphering the Heterogeneity
+I. A. Mayer, V. G. Abramson, B. D. Lehmann, J. A. Pietenpol
+Clinical Cancer Research (2014-02-15) https://doi.org/10.1158/1078-0432.ccr-13-0583

+
+
+

26. UNSUPERVISED FEATURE CONSTRUCTION AND KNOWLEDGE EXTRACTION FROM GENOME-WIDE ASSAYS OF BREAST CANCER WITH DENOISING AUTOENCODERS
+JIE TAN, MATTHEW UNG, CHAO CHENG, CASEY S GREENE
+Biocomputing 2015 (2014-11) https://doi.org/10.1142/9789814644730_0014

+
+
+

27. Mitosis Detection in Breast Cancer Histology Images with Deep Neural Networks
+Dan C. Cireşan, Alessandro Giusti, Luca M. Gambardella, Jürgen Schmidhuber
+Medical Image Computing and Computer-Assisted Intervention – MICCAI 2013 (2013) https://doi.org/10.1007/978-3-642-40763-5_51

+
+
+

28. End effector target position learning using feedforward with error back-propagation and recurrent neural networks
+J. Zurada
+Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN’94) (1994) https://doi.org/10.1109/icnn.1994.374637

+
+
+

29. Accurate De Novo Prediction of Protein Contact Map by Ultra-Deep Learning Model
+Sheng Wang, Siqi Sun, Zhen Li, Renyu Zhang, Jinbo Xu
+PLOS Computational Biology (2017-01-05) https://doi.org/10.1371/journal.pcbi.1005324

+
+
+

30. A Deep Learning Network Approach to ab initio Protein Secondary Structure Prediction
+Matt Spencer, Jesse Eickholt, Jianlin Cheng
+IEEE/ACM Transactions on Computational Biology and Bioinformatics (2015-01-01) https://doi.org/10.1109/tcbb.2014.2343960

+
+
+

31. Protein Secondary Structure Prediction Using Deep Convolutional Neural Fields
+Sheng Wang, Jian Peng, Jianzhu Ma, Jinbo Xu
+Scientific Reports (2016-01-11) https://doi.org/10.1038/srep18962

+
+
+

32. PEDLA: predicting enhancers with a deep learning-based algorithmic framework
+Feng Liu, Hao Li, Chao Ren, Xiaochen Bo, Wenjie Shu
+Cold Spring Harbor Laboratory (2016-01-07) https://doi.org/10.1101/036129

+
+
+

33. Deep Feature Selection: Theory and Application to Identify Enhancers and Promoters
+Yifeng Li, Chih-Yu Chen, Wyeth W. Wasserman
+Lecture Notes in Computer Science (2015) https://doi.org/10.1007/978-3-319-16706-0_20

+
+
+

34. DEEP: a general computational framework for predicting enhancers
+Dimitrios Kleftogiannis, Panos Kalnis, Vladimir B. Bajic
+Nucleic Acids Research (2014-11-05) https://doi.org/10.1093/nar/gku1058

+
+
+

35. DANN: a deep learning approach for annotating the pathogenicity of genetic variants
+Daniel Quang, Yifei Chen, Xiaohui Xie
+Bioinformatics (2014-10-22) https://doi.org/10.1093/bioinformatics/btu703

+
+
+

36. AtomNet: A Deep Convolutional Neural Network for Bioactivity Prediction in Structure-based Drug Discovery
+Izhar Wallach, Michael Dzamba, Abraham Heifets
+arXiv (2015-10-10) https://arxiv.org/abs/1510.02855v1

+
+
+

37. Deep Learning Applications for Predicting Pharmacological Properties of Drugs and Drug Repurposing Using Transcriptomic Data
+Alexander Aliper, Sergey Plis, Artem Artemov, Alvaro Ulloa, Polina Mamoshina, Alex Zhavoronkov
+Molecular Pharmaceutics (2016-07-05) https://doi.org/10.1021/acs.molpharmaceut.6b00248

+
+
+

38. Predicting drug-target interactions using restricted Boltzmann machines
+Yuhao Wang, Jianyang Zeng
+Bioinformatics (2013-06-19) https://doi.org/10.1093/bioinformatics/btt234

+
+
+

39. Deep-Learning-Based Drug–Target Interaction Prediction
+Ming Wen, Zhimin Zhang, Shaoyu Niu, Haozhi Sha, Ruihan Yang, Yonghuan Yun, Hongmei Lu
+Journal of Proteome Research (2017-03-13) https://doi.org/10.1021/acs.jproteome.6b00618

+
+
+

40. Latent Autoimmune Diabetes in Adults: Definition, Prevalence,  -Cell Function, and Treatment
+G. Stenstrom, A. Gottsater, E. Bakhtadze, B. Berger, G. Sundkvist
+Diabetes (2005-11-23) https://doi.org/10.2337/diabetes.54.suppl_2.s68

+
+
+

41. Islet Cell Antibodies Identify Latent Type I Diabetes in Patients Aged 35-75 Years at Diagnosis
+L. C. Groop, G. F. Bottazzo, D. Doniach
+Diabetes (1986-02-01) https://doi.org/10.2337/diab.35.2.237

+
+
+

42. Deep Learning in Medical Image Analysis
+Dinggang Shen, Guorong Wu, Heung-Il Suk
+Annual Review of Biomedical Engineering (2017-06-21) https://doi.org/10.1146/annurev-bioeng-071516-044442

+
+
+

43. ImageNet Large Scale Visual Recognition Challenge
+Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, … Li Fei-Fei
+International Journal of Computer Vision (2015-04-11) https://doi.org/10.1007/s11263-015-0816-y

+
+
+

44. Convolutional Neural Networks for Diabetic Retinopathy
+Harry Pratt, Frans Coenen, Deborah M. Broadbent, Simon P. Harding, Yalin Zheng
+Procedia Computer Science (2016) https://doi.org/10.1016/j.procs.2016.07.014

+
+
+

45. Improved Automated Detection of Diabetic Retinopathy on a Publicly Available Dataset Through Integration of Deep Learning
+Michael David Abràmoff, Yiyue Lou, Ali Erginay, Warren Clarida, Ryan Amelon, James C. Folk, Meindert Niemeijer
+Investigative Opthalmology & Visual Science (2016-10-04) https://doi.org/10.1167/iovs.16-19964

+
+
+

46. Leveraging uncertainty information from deep neural networks for disease detection
+Christian Leibig, Vaneeda Allken, Murat Seckin Ayhan, Philipp Berens, Siegfried Wahl
+Cold Spring Harbor Laboratory (2016-10-28) https://doi.org/10.1101/084210

+
+
+

47. Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs
+Varun Gulshan, Lily Peng, Marc Coram, Martin C. Stumpe, Derek Wu, Arunachalam Narayanaswamy, Subhashini Venugopalan, Kasumi Widner, Tom Madams, Jorge Cuadros, … Dale R. Webster
+JAMA (2016-12-13) https://doi.org/10.1001/jama.2016.17216

+
+
+

48. Deep Learning Ensembles for Melanoma Recognition in Dermoscopy Images
+Noel Codella, Quoc-Bao Nguyen, Sharath Pankanti, David Gutman, Brian Helba, Allan Halpern, John R. Smith
+arXiv (2016-10-14) https://arxiv.org/abs/1610.04662v2

+
+
+

49. Automated Melanoma Recognition in Dermoscopy Images via Very Deep Residual Networks
+Lequan Yu, Hao Chen, Qi Dou, Jing Qin, Pheng-Ann Heng
+IEEE Transactions on Medical Imaging (2017-04) https://doi.org/10.1109/tmi.2016.2642839

+
+
+

50. Extraction of Skin Lesions from Non-Dermoscopic Images Using Deep Learning
+Mohammad H. Jafari, Ebrahim Nasr-Esfahani, Nader Karimi, S. M. Reza Soroushmehr, Shadrokh Samavi, Kayvan Najarian
+arXiv (2016-09-08) https://arxiv.org/abs/1609.02374v1

+
+
+

51. Melanoma detection by analysis of clinical images using convolutional neural network
+E. Nasr-Esfahani, S. Samavi, N. Karimi, S.M.R. Soroushmehr, M.H. Jafari, K. Ward, K. Najarian
+2016 38th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) (2016-08) https://doi.org/10.1109/embc.2016.7590963

+
+
+

52. Detection of age-related macular degeneration via deep learning
+P. Burlina, D. E. Freund, N. Joshi, Y. Wolfson, N. M. Bressler
+2016 IEEE 13th International Symposium on Biomedical Imaging (ISBI) (2016-04) https://doi.org/10.1109/isbi.2016.7493240

+
+
+

53. Deep learning with non-medical training used for chest pathology identification
+Yaniv Bar, Idit Diamant, Lior Wolf, Hayit Greenspan
+Medical Imaging 2015: Computer-Aided Diagnosis (2015-03-20) https://doi.org/10.1117/12.2083124

+
+
+

54. Deep Convolutional Neural Networks for Computer-Aided Detection: CNN Architectures, Dataset Characteristics and Transfer Learning
+Hoo-Chang Shin, Holger R. Roth, Mingchen Gao, Le Lu, Ziyue Xu, Isabella Nogues, Jianhua Yao, Daniel Mollura, Ronald M. Summers
+IEEE Transactions on Medical Imaging (2016-05) https://doi.org/10.1109/tmi.2016.2528162

+
+
+

55. High-Throughput Classification of Radiographs Using Deep Convolutional Neural Networks
+Alvin Rajkomar, Sneha Lingam, Andrew G. Taylor, Michael Blum, John Mongan
+Journal of Digital Imaging (2016-10-11) https://doi.org/10.1007/s10278-016-9914-9

+
+
+

56. Deep Learning at Chest Radiography: Automated Classification of Pulmonary Tuberculosis by Using Convolutional Neural Networks
+Paras Lakhani, Baskaran Sundaram
+Radiology (2017-08) https://doi.org/10.1148/radiol.2017162326

+
+
+

57. Classification of breast MRI lesions using small-size training sets: comparison of deep learning approaches
+Guy Amit, Rami Ben-Ari, Omer Hadad, Einat Monovich, Noa Granot, Sharbell Hashoul
+Medical Imaging 2017: Computer-Aided Diagnosis (2017-03-03) https://doi.org/10.1117/12.2249981

+
+
+

58. Deep Learning and Structured Prediction for the Segmentation of Mass in Mammograms
+Neeraj Dhungel, Gustavo Carneiro, Andrew P. Bradley
+Lecture Notes in Computer Science (2015) https://doi.org/10.1007/978-3-319-24553-9_74

+
+
+

59. The Automated Learning of Deep Features for Breast Mass Classification from Mammograms
+Neeraj Dhungel, Gustavo Carneiro, Andrew P. Bradley
+Medical Image Computing and Computer-Assisted Intervention – MICCAI 2016 (2016) https://doi.org/10.1007/978-3-319-46723-8_13

+
+
+

60. Deep Multi-instance Networks with Sparse Label Assignment for Whole Mammogram Classification
+Wentao Zhu, Qi Lou, Yeeleng Scott Vang, Xiaohui Xie
+Cold Spring Harbor Laboratory (2016-12-20) https://doi.org/10.1101/095794

+
+
+

61. Adversarial Deep Structural Networks for Mammographic Mass Segmentation
+Wentao Zhu, Xiaohui Xie
+Cold Spring Harbor Laboratory (2016-12-20) https://doi.org/10.1101/095786

+
+
+

62. A deep learning approach for the analysis of masses in mammograms with minimal user intervention
+Neeraj Dhungel, Gustavo Carneiro, Andrew P. Bradley
+Medical Image Analysis (2017-04) https://doi.org/10.1016/j.media.2017.01.009

+
+
+

63. Improving Computer-Aided Detection Using_newlineConvolutional Neural Networks and Random View Aggregation
+Holger R. Roth, Le Lu, Jiamin Liu, Jianhua Yao, Ari Seff, Kevin Cherry, Lauren Kim, Ronald M. Summers
+IEEE Transactions on Medical Imaging (2016-05) https://doi.org/10.1109/tmi.2015.2482920

+
+
+

64. 3D Deep Learning for Multi-modal Imaging-Guided Survival Time Prediction of Brain Tumor Patients
+Dong Nie, Han Zhang, Ehsan Adeli, Luyan Liu, Dinggang Shen
+Medical Image Computing and Computer-Assisted Intervention – MICCAI 2016 (2016) https://doi.org/10.1007/978-3-319-46723-8_25

+
+
+

65. Large scale deep learning for computer aided detection of mammographic lesions
+Thijs Kooi, Geert Litjens, Bram van Ginneken, Albert Gubern-Mérida, Clara I. Sánchez, Ritse Mann, Ard den Heeten, Nico Karssemeijer
+Medical Image Analysis (2017-01) https://doi.org/10.1016/j.media.2016.07.007

+
+
+

66. Deep learning as a tool for increased accuracy and efficiency of histopathological diagnosis
+Geert Litjens, Clara I. Sánchez, Nadya Timofeeva, Meyke Hermsen, Iris Nagtegaal, Iringo Kovacs, Christina Hulsbergen - van de Kaa, Peter Bult, Bram van Ginneken, Jeroen van der Laak
+Scientific Reports (2016-05-23) https://doi.org/10.1038/srep26286

+
+
+

67. Deep Learning for Identifying Metastatic Breast Cancer
+Dayong Wang, Aditya Khosla, Rishab Gargeya, Humayun Irshad, Andrew H. Beck
+arXiv (2016-06-18) https://arxiv.org/abs/1606.05718v1

+
+
+

68. Deep learning is effective for the classification of OCT images of normal versus Age-related Macular Degeneration
+Cecilia S Lee, Doug M Baughman, Aaron Y Lee
+Cold Spring Harbor Laboratory (2016-12-14) https://doi.org/10.1101/094276

+
+
+

69. ImageNet Classification with Deep Convolutional Neural Networks
+Alex Krizhevsky, Ilya Sutskever, Geoffrey E. Hinton
+Proceedings of the 25th International Conference on Neural Information Processing Systems (2012) http://dl.acm.org/citation.cfm?id=2999134.2999257

+
+
+

70. Unsupervised Joint Mining of Deep Features and Image Labels for Large-scale Radiology Image Categorization and Scene Recognition
+Xiaosong Wang, Le Lu, Hoo-chang Shin, Lauren Kim, Mohammadhadi Bagheri, Isabella Nogues, Jianhua Yao, Ronald M. Summers
+arXiv (2017-01-23) https://arxiv.org/abs/1701.06599v1

+
+
+

71. ChestX-ray8: Hospital-scale Chest X-ray Database and Benchmarks on Weakly-Supervised Classification and Localization of Common Thorax Diseases
+Xiaosong Wang, Yifan Peng, Le Lu, Zhiyong Lu, Mohammadhadi Bagheri, Ronald M. Summers
+arXiv (2017-05-05) https://arxiv.org/abs/1705.02315v4

+
+
+

72. Points of Significance: Classification evaluation
+Jake Lever, Martin Krzywinski, Naomi Altman
+Nature Methods (2016-07-28) https://doi.org/10.1038/nmeth.3945

+
+
+

73. NIH Chest X-ray Dataset
+NIH Clinical Center
+(2017-09-07) https://nihcc.app.box.com/v/ChestXray-NIHCC

+
+
+

74. Realizing the full potential of electronic health records: the role of natural language processing
+Lucila Ohno-Machado
+Journal of the American Medical Informatics Association (2011-09) https://doi.org/10.1136/amiajnl-2011-000501

+
+
+

75. Machine-learned solutions for three stages of clinical information extraction: the state of the art at i2b2 2010
+Berry de Bruijn, Colin Cherry, Svetlana Kiritchenko, Joel Martin, Xiaodan Zhu
+Journal of the American Medical Informatics Association (2011-09) https://doi.org/10.1136/amiajnl-2011-000150

+
+
+

76. Bidirectional LSTM-CRF for Clinical Concept Extraction
+Raghavendra Chalapathy, Ehsan Zare Borzeshi, Massimo Piccardi
+arXiv (2016-11-25) https://arxiv.org/abs/1611.08373v1

+
+
+

77. Multi-task Deep Neural Networks for Automated Extraction of Primary Site and Laterality Information from Cancer Pathology Reports
+Hong-Jun Yoon, Arvind Ramanathan, Georgia Tourassi
+Advances in Big Data (2016-10-08) https://doi.org/10.1007/978-3-319-47898-2_21

+
+
+

78. Efficient Estimation of Word Representations in Vector Space
+Tomas Mikolov, Kai Chen, Greg Corrado, Jeffrey Dean
+arXiv (2013-01-16) https://arxiv.org/abs/1301.3781v3

+
+
+

79. Medical Semantic Similarity with a Neural Language Model
+Lance De Vine, Guido Zuccon, Bevan Koopman, Laurianne Sitbon, Peter Bruza
+Proceedings of the 23rd ACM International Conference on Conference on Information and Knowledge Management - CIKM ’14 (2014) https://doi.org/10.1145/2661829.2661974

+
+
+

80. Multi-layer Representation Learning for Medical Concepts
+Edward Choi, Mohammad Taha Bahadori, Elizabeth Searles, Catherine Coffey, Michael Thompson, James Bost, Javier Tejedor-Sojo, Jimeng Sun
+Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining - KDD ’16 (2016) https://doi.org/10.1145/2939672.2939823

+
+
+

81. Large-Scale Discovery of Disease-Disease and Disease-Gene Associations
+Djordje Gligorijevic, Jelena Stojanovic, Nemanja Djuric, Vladan Radosavljevic, Mihajlo Grbovic, Rob J. Kulathinal, Zoran Obradovic
+Scientific Reports (2016-08-31) https://doi.org/10.1038/srep32404

+
+
+

82. Computational Phenotype Discovery Using Unsupervised Feature Learning over Noisy, Sparse, and Irregular Clinical Data
+Thomas A. Lasko, Joshua C. Denny, Mia A. Levy
+PLoS ONE (2013-06-24) https://doi.org/10.1371/journal.pone.0066341

+
+
+

83. Semi-supervised learning of the electronic health record for phenotype stratification
+Brett K. Beaulieu-Jones, Casey S. Greene
+Journal of Biomedical Informatics (2016-12) https://doi.org/10.1016/j.jbi.2016.10.007

+
+
+

84. Deep Patient: An Unsupervised Representation to Predict the Future of Patients from the Electronic Health Records
+Riccardo Miotto, Li Li, Brian A. Kidd, Joel T. Dudley
+Scientific Reports (2016-05-17) https://doi.org/10.1038/srep26094

+
+
+

85. Multi-task Prediction of Disease Onsets from Longitudinal Lab Tests
+Narges Razavian, Jake Marcus, David Sontag
+arXiv (2016-08-02) https://arxiv.org/abs/1608.00647v3

+
+
+

86. Deep Survival Analysis
+Rajesh Ranganath, Adler Perotte, Noémie Elhadad, David Blei
+arXiv (2016-08-06) https://arxiv.org/abs/1608.02158v2

+
+
+

87. Comparison of the performance of neural network methods and Cox regression for censored survival data
+Anny Xiang, Pablo Lapuerta, Alex Ryutov, Jonathan Buckley, Stanley Azen
+Computational Statistics & Data Analysis (2000-08) https://doi.org/10.1016/s0167-9473(99)00098-5

+
+
+

88. DeepSurv: Personalized Treatment Recommender System Using A Cox Proportional Hazards Deep Neural Network
+Jared Katzman, Uri Shaham, Jonathan Bates, Alexander Cloninger, Tingting Jiang, Yuval Kluger
+arXiv (2016-06-02) https://arxiv.org/abs/1606.00931v3

+
+
+

89. Deep Exponential Families
+Rajesh Ranganath, Linpeng Tang, Laurent Charlin, David M. Blei
+arXiv (2014-11-10) https://arxiv.org/abs/1411.2581v1

+
+
+

90. Stochastic Variational Inference
+Matt Hoffman, David M. Blei, Chong Wang, John Paisley
+arXiv (2012-06-29) https://arxiv.org/abs/1206.7051v3

+
+
+

91. Hierarchical Variational Models
+Rajesh Ranganath, Dustin Tran, David M. Blei
+arXiv (2015-11-07) https://arxiv.org/abs/1511.02386v2

+
+
+

92. A machine learning-based framework to identify type 2 diabetes through electronic health records
+Tao Zheng, Wei Xie, Liling Xu, Xiaoying He, Ya Zhang, Mingrong You, Gong Yang, You Chen
+International Journal of Medical Informatics (2017-01) https://doi.org/10.1016/j.ijmedinf.2016.09.014

+
+
+

93. Implementations by Phenotype | PheKB(2017-05-17) https://phekb.org/implementations

+
+
+

94. Electronic medical record phenotyping using the anchor and learn framework
+Yoni Halpern, Steven Horng, Youngduck Choi, David Sontag
+Journal of the American Medical Informatics Association (2016-04-23) https://doi.org/10.1093/jamia/ocw011

+
+
+

95. Data Programming: Creating Large Training Sets, Quickly
+Alexander Ratner, Christopher De Sa, Sen Wu, Daniel Selsam, Christopher Ré
+arXiv (2016-05-25) https://arxiv.org/abs/1605.07723v3

+
+
+

96. Data is the New Oil
+Michael Palmer
+ANA Marketing Maestros (2006-11) http://ana.blogs.com/maestros/2006/11/data_is_the_new.html

+
+
+

97. “Data is the New Oil” — A Ludicrous Proposition – Twenty One Hundred – Medium
+Michael Haupt
+Medium (2016-05-02) https://medium.com/twenty-one-hundred/data-is-the-new-oil-a-ludicrous-proposition-1d91bba4f294

+
+
+

98. Data Programming: Machine Learning with Weak Supervision
+Alex Ratner, Stephen Bach, Chris Ré
+(2016-09-19) http://hazyresearch.github.io/snorkel/blog/weak_supervision.html

+
+
+

99. Mining electronic health records: towards better research applications and clinical care
+Peter B. Jensen, Lars J. Jensen, Søren Brunak
+Nature Reviews Genetics (2012-05-02) https://doi.org/10.1038/nrg3208

+
+
+

100. Methods and dimensions of electronic health record data quality assessment: enabling reuse for clinical research
+N. G. Weiskopf, C. Weng
+Journal of the American Medical Informatics Association (2013-01-01) https://doi.org/10.1136/amiajnl-2011-000681

+
+
+

101. Impact of Electronic Health Record Systems on Information Integrity: Quality and Safety Implications
+Sue Bowman
+Perspectives in Health Information Management (2013) https://www.ncbi.nlm.nih.gov/pubmed/24159271

+
+
+

102. Secondary Use of EHR: Data Quality Issues and Informatics Opportunities
+Taxiarchis Botsis, Gunnar Hartvigsen, Fei Chen, Chunhua Weng
+Summit on Translational Bioinformatics (2010) https://www.ncbi.nlm.nih.gov/pubmed/21347133

+
+
+

103. Have DRG-based prospective payment systems influenced the number of secondary diagnoses in health care administrative data?
+Lisbeth Serdén, Rikard Lindqvist, Måns Rosén
+Health Policy (2003-08) https://doi.org/10.1016/s0168-8510(02)00208-7

+
+
+

104. Why Patient Matching Is a Challenge: Research on Master Patient Index (MPI) Data Discrepancies in Key Identifying Fields
+Beth Haenke Just, David Marc, Megan Munns, Ryan Sandefer
+Perspectives in Health Information Management (2016) https://www.ncbi.nlm.nih.gov/pubmed/27134610

+
+
+

105. Identifying and mitigating biases in EHR laboratory tests
+Rimma Pivovarov, David J. Albers, Jorge L. Sepulveda, Noémie Elhadad
+Journal of Biomedical Informatics (2014-10) https://doi.org/10.1016/j.jbi.2014.03.016

+
+
+

106. Using electronic health records for clinical research: The case of the EHR4CR project
+Georges De Moor, Mats Sundgren, Dipak Kalra, Andreas Schmidt, Martin Dugas, Brecht Claerhout, Töresin Karakoyun, Christian Ohmann, Pierre-Yves Lastic, Nadir Ammour, … Pascal Coorevits
+Journal of Biomedical Informatics (2015-02) https://doi.org/10.1016/j.jbi.2014.10.006

+
+
+

107. Healthcare Interoperability Standards Compliance Handbook
+Frank Oemig, Robert Snelick
+Springer International Publishing (2016) https://doi.org/10.1007/978-3-319-44839-8

+
+
+

108. How sample size influences research outcomes
+Jorge Faber, Lilian Martins Fonseca
+Dental Press Journal of Orthodontics (2014-08) https://doi.org/10.1590/2176-9451.19.4.027-029.ebo

+
+
+

109. A review of approaches to identifying patient phenotype cohorts using electronic health records
+Chaitanya Shivade, Preethi Raghavan, Eric Fosler-Lussier, Peter J Embi, Noemie Elhadad, Stephen B Johnson, Albert M Lai
+Journal of the American Medical Informatics Association (2014-03) https://doi.org/10.1136/amiajnl-2013-001935

+
+
+

110. STRATEGIES FOR EQUITABLE PHARMACOGENOMIC-GUIDED WARFARIN DOSING AMONG EUROPEAN AND AFRICAN AMERICAN INDIVIDUALS IN A CLINICAL POPULATION
+LAURA K. WILEY, JACOB P. VANHOUTEN, DAVID C. SAMUELS, MELINDA C. ALDRICH, DAN M. RODEN, JOSH F. PETERSON, JOSHUA C. DENNY
+Biocomputing 2017 (2016-11-22) https://doi.org/10.1142/9789813207813_0050

+
+
+

111. Epidemiological research labelled as a violation of privacy: the case of Estonia
+M. Rahu, M. McKee
+International Journal of Epidemiology (2008-02-26) https://doi.org/10.1093/ije/dyn022

+
+
+

112. Harnessing next-generation informatics for personalizing medicine: a report from AMIA’s 2014 Health Policy Invitational Meeting
+Laura K Wiley, Peter Tarczy-Hornoch, Joshua C Denny, Robert R Freimuth, Casey L Overby, Nigam Shah, Ross D Martin, Indra Neil Sarkar
+Journal of the American Medical Informatics Association (2016-02-05) https://doi.org/10.1093/jamia/ocv111

+
+
+

113. DataSHIELD: taking the analysis to the data, not the data to the analysis
+Amadou Gaye, Yannick Marcon, Julia Isaeva, Philippe LaFlamme, Andrew Turner, Elinor M Jones, Joel Minion, Andrew W Boyd, Christopher J Newby, Marja-Liisa Nuotio, … Paul R Burton
+International Journal of Epidemiology (2014-09-27) https://doi.org/10.1093/ije/dyu188

+
+
+

114. ViPAR: a software platform for the Virtual Pooling and Analysis of Research Data
+Kim W Carter, KW Carter, RW Francis, M Bresnahan, M Gissler, TK Grønborg, R Gross, N Gunnes, G Hammond, M Hornig, … Z Yusof
+International Journal of Epidemiology (2015-10-08) https://doi.org/10.1093/ije/dyv193

+
+
+

115. Reproducibility of computational workflows is automated using continuous analysis
+Brett K Beaulieu-Jones, Casey S Greene
+Nature Biotechnology (2017-03-13) https://doi.org/10.1038/nbt.3780

+
+
+

116. Stealing Machine Learning Models via Prediction APIs
+Florian Tramèr, Fan Zhang, Ari Juels, Michael K. Reiter, Thomas Ristenpart
+arXiv (2016-09-09) https://arxiv.org/abs/1609.02943v2

+
+
+

117. The Algorithmic Foundations of Differential Privacy
+Cynthia Dwork, Aaron Roth
+Foundations and Trends® in Theoretical Computer Science (2013) https://doi.org/10.1561/0400000042

+
+
+

118. Membership Inference Attacks against Machine Learning Models
+Reza Shokri, Marco Stronati, Congzheng Song, Vitaly Shmatikov
+arXiv (2016-10-18) https://arxiv.org/abs/1610.05820v2

+
+
+

119. Generating Multi-label Discrete Electronic Health Records using Generative Adversarial Networks
+Edward Choi, Siddharth Biswal, Bradley Malin, Jon Duke, Walter F. Stewart, Jimeng Sun
+arXiv (2017-03-19) https://arxiv.org/abs/1703.06490v1

+
+
+

120. Enabling Privacy-Preserving GWASs in Heterogeneous Human Populations
+Sean Simmons, Cenk Sahinalp, Bonnie Berger
+Cell Systems (2016-07) https://doi.org/10.1016/j.cels.2016.04.013

+
+
+

121. Deep Learning with Differential Privacy
+Martín Abadi, Andy Chu, Ian Goodfellow, H. Brendan McMahan, Ilya Mironov, Kunal Talwar, Li Zhang
+arXiv (2016-07-01) https://arxiv.org/abs/1607.00133v2

+
+
+

122. Communication-Efficient Learning of Deep Networks from Decentralized Data
+Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, Blaise Aguera y Arcas
+(2017-04-10) http://proceedings.mlr.press/v54/mcmahan17a.html

+
+
+

123. Practical Secure Aggregation for Privacy Preserving Machine Learning
+Keith Bonawitz, Vladimir Ivanov, Ben Kreuter, Antonio Marcedone, H. Brendan McMahan, Sarvar Patel, Daniel Ramage, Aaron Segal, Karn Seth
+(2017) https://eprint.iacr.org/2017/281

+
+
+

124. Semi-supervised Knowledge Transfer for Deep Learning from Private Training Data
+Nicolas Papernot, Martín Abadi, Úlfar Erlingsson, Ian Goodfellow, Kunal Talwar
+(2016-11-02) https://openreview.net/forum?id=HkwoSDPgg

+
+
+

125. European Union regulations on algorithmic decision-making and a “right to explanation”
+Bryce Goodman, Seth Flaxman
+arXiv (2016-06-28) https://arxiv.org/abs/1606.08813v3

+
+
+

126. Overcoming the Winner’s Curse: Estimating Penetrance Parameters from Case-Control Data
+Sebastian Zöllner, Jonathan K. Pritchard
+The American Journal of Human Genetics (2007-04) https://doi.org/10.1086/512821

+
+
+

127. Sex bias in neuroscience and biomedical research
+Annaliese K. Beery, Irving Zucker
+Neuroscience & Biobehavioral Reviews (2011-01) https://doi.org/10.1016/j.neubiorev.2010.07.002

+
+
+

128. Generalization and Dilution of Association Results from European GWAS in Populations of Non-European Ancestry: The PAGE Study
+Christopher S. Carlson, Tara C. Matise, Kari E. North, Christopher A. Haiman, Megan D. Fesinmeyer, Steven Buyske, Fredrick R. Schumacher, Ulrike Peters, Nora Franceschini, Marylyn D. Ritchie, …
+PLoS Biology (2013-09-17) https://doi.org/10.1371/journal.pbio.1001661

+
+
+

129. New approaches to population stratification in genome-wide association studies
+Alkes L. Price, Noah A. Zaitlen, David Reich, Nick Patterson
+Nature Reviews Genetics (2010-06-15) https://doi.org/10.1038/nrg2813

+
+
+

130. Retraction
+P. Sebastiani, N. Solovieff, A. Puca, S. W. Hartley, E. Melista, S. Andersen, D. A. Dworkis, J. B. Wilk, R. H. Myers, M. H. Steinberg, … T. T. Perls
+Science (2011-07-21) https://doi.org/10.1126/science.333.6041.404-a

+
+
+

131. Leakage in data mining
+Shachar Kaufman, Saharon Rosset, Claudia Perlich, Ori Stitelman
+ACM Transactions on Knowledge Discovery from Data (2012-12-01) https://doi.org/10.1145/2382577.2382579

+
+
+

132. To predict and serve?
+Kristian Lum, William Isaac
+Significance (2016-10) https://doi.org/10.1111/j.1740-9713.2016.00960.x

+
+
+

133. Equality of Opportunity in Supervised Learning
+Moritz Hardt, Eric Price, Nathan Srebro
+arXiv (2016-10-07) https://arxiv.org/abs/1610.02413v1

+
+
+

134. Fair Algorithms for Infinite and Contextual Bandits
+Matthew Joseph, Michael Kearns, Jamie Morgenstern, Seth Neel, Aaron Roth
+arXiv (2016-10-29) https://arxiv.org/abs/1610.09559v4

+
+
+

135. The Framingham Heart Study and the epidemiology of cardiovascular disease: a historical perspective
+Syed S Mahmood, Daniel Levy, Ramachandran S Vasan, Thomas J Wang
+The Lancet (2014-03) https://doi.org/10.1016/s0140-6736(13)61752-3

+
+
+

136. Children of the 90s: Coming of age
+Helen Pearson
+Nature (2012-04-11) https://doi.org/10.1038/484155a

+
+
+

137. Nonparametric Estimation from Incomplete Observations
+E. L. Kaplan, Paul Meier
+Journal of the American Statistical Association (1958-06) https://doi.org/10.2307/2281868

+
+
+

138. Temporal disease trajectories condensed from population-wide registry data covering 6.2 million patients
+Anders Boeck Jensen, Pope L. Moseley, Tudor I. Oprea, Sabrina Gade Ellesøe, Robert Eriksson, Henriette Schmock, Peter Bjødstrup Jensen, Lars Juhl Jensen, Søren Brunak
+Nature Communications (2014-06-24) https://doi.org/10.1038/ncomms5022

+
+
+

139. Deepr: A Convolutional Net for Medical Records
+Phuoc Nguyen, Truyen Tran, Nilmini Wickramasinghe, Svetha Venkatesh
+arXiv (2016-07-26) https://arxiv.org/abs/1607.07519v1

+
+
+

140. DeepCare: A Deep Dynamic Memory Model for Predictive Medicine
+Trang Pham, Truyen Tran, Dinh Phung, Svetha Venkatesh
+arXiv (2016-02-01) https://arxiv.org/abs/1602.00357v2

+
+
+

141. Curiosity Creates Cures: The Value and Impact of Basic Research
+NIH
+(2012-05) https://www.nigms.nih.gov/Education/Documents/curiosity.pdf

+
+
+

142. Multi-omics integration accurately predicts cellular state in unexplored conditions for Escherichia coli
+Minseung Kim, Navneet Rai, Violeta Zorraquino, Ilias Tagkopoulos
+Nature Communications (2016-10-07) https://doi.org/10.1038/ncomms13090

+
+
+

143. Trans-species learning of cellular signaling systems with bimodal deep belief networks
+Lujia Chen, Chunhui Cai, Vicky Chen, Xinghua Lu
+Bioinformatics (2015-05-20) https://doi.org/10.1093/bioinformatics/btv315

+
+
+

144. Learning structure in gene expression data using deep architectures, with an application to gene clustering
+Aman Gupta, Haohan Wang, Madhavi Ganapathiraju
+Cold Spring Harbor Laboratory (2015-11-16) https://doi.org/10.1101/031906

+
+
+

145. Learning a hierarchical representation of the yeast transcriptomic machinery using an autoencoder model
+Lujia Chen, Chunhui Cai, Vicky Chen, Xinghua Lu
+BMC Bioinformatics (2016-01-11) https://doi.org/10.1186/s12859-015-0852-1

+
+
+

146. ADAGE-Based Integration of Publicly AvailablePseudomonas aeruginosaGene Expression Data with Denoising Autoencoders Illuminates Microbe-Host Interactions
+Jie Tan, John H. Hammond, Deborah A. Hogan, Casey S. Greene
+mSystems (2016-01-19) https://doi.org/10.1128/msystems.00025-15

+
+
+

147. Unsupervised extraction of stable expression signatures from public compendia with eADAGE
+Jie Tan, Georgia Doing, Kimberley A Lewis, Courtney E Price, Kathleen M Chen, Kyle C Cady, Barret Perchuk, Michael T Laub, Deborah A Hogan, Casey S Greene
+Cold Spring Harbor Laboratory (2016-10-03) https://doi.org/10.1101/078659

+
+
+

148. Gene expression inference with deep learning
+Yifei Chen, Yi Li, Rajiv Narayan, Aravind Subramanian, Xiaohui Xie
+Bioinformatics (2016-02-11) https://doi.org/10.1093/bioinformatics/btw074

+
+
+

149. DeepChrome: Deep-learning for predicting gene expression from histone modifications
+Ritambhara Singh, Jack Lanchantin, Gabriel Robins, Yanjun Qi
+arXiv (2016-07-07) https://arxiv.org/abs/1607.02078v1

+
+
+

150. Integrative Data Analysis of Multi-Platform Cancer Data with a Multimodal Deep Learning Approach
+Muxuan Liang, Zhizhong Li, Ting Chen, Jianyang Zeng
+IEEE/ACM Transactions on Computational Biology and Bioinformatics (2015-07-01) https://doi.org/10.1109/tcbb.2014.2377729

+
+
+

151. RNA mis-splicing in disease
+Marina M. Scotti, Maurice S. Swanson
+Nature Reviews Genetics (2015-11-23) https://doi.org/10.1038/nrg.2015.3

+
+
+

152. RNA splicing is a primary link between genetic variation and disease
+Y. I. Li, B. van de Geijn, A. Raj, D. A. Knowles, A. A. Petti, D. Golan, Y. Gilad, J. K. Pritchard
+Science (2016-04-28) https://doi.org/10.1126/science.aad9417

+
+
+

153. Deciphering the splicing code
+Yoseph Barash, John A. Calarco, Weijun Gao, Qun Pan, Xinchen Wang, Ofer Shai, Benjamin J. Blencowe, Brendan J. Frey
+Nature (2010-05-06) https://doi.org/10.1038/nature09000

+
+
+

154. Bayesian prediction of tissue-regulated splicing using RNA sequence and cellular context
+Hui Yuan Xiong, Yoseph Barash, Brendan J. Frey
+Bioinformatics (2011-07-29) https://doi.org/10.1093/bioinformatics/btr444

+
+
+

155. The human splicing code reveals new insights into the genetic determinants of disease
+H. Y. Xiong, B. Alipanahi, L. J. Lee, H. Bretschneider, D. Merico, R. K. C. Yuen, Y. Hua, S. Gueroussov, H. S. Najafabadi, T. R. Hughes, … B. J. Frey
+Science (2014-12-18) https://doi.org/10.1126/science.1254806

+
+
+

156. Integrative Deep Models for Alternative Splicing
+Anupama Jha, Matthew R Gazzara, Yoseph Barash
+Cold Spring Harbor Laboratory (2017-01-31) https://doi.org/10.1101/104869

+
+
+

157. Imputation for transcription factor binding predictions based on deep learning
+Qian Qin, Jianxing Feng
+PLOS Computational Biology (2017-02-24) https://doi.org/10.1371/journal.pcbi.1005403

+
+
+

158. Learning the Sequence Determinants of Alternative Splicing from Millions of Random Sequences
+Alexander B. Rosenberg, Rupali P. Patwardhan, Jay Shendure, Georg Seelig
+Cell (2015-10) https://doi.org/10.1016/j.cell.2015.09.054

+
+
+

159. MECHANISMS IN ENDOCRINOLOGY: Alternative splicing: the new frontier in diabetes research
+Jonàs Juan-Mateu, Olatz Villate, Décio L Eizirik
+European Journal of Endocrinology (2015-12-01) https://doi.org/10.1530/eje-15-0916

+
+
+

160. RNA-protein binding motifs mining with a new hybrid deep learning based cross-domain knowledge integration approach
+Xiaoyong Pan, Hong-Bin Shen
+BMC Bioinformatics (2017-02-28) https://doi.org/10.1186/s12859-017-1561-8

+
+
+

161. An integrated encyclopedia of DNA elements in the human genome
+Ian Dunham, Anshul Kundaje, Shelley F. Aldred, Patrick J. Collins, Carrie A. Davis, Francis Doyle, Charles B. Epstein, Seth Frietze, Jennifer Harrow, Rajinder Kaul, … Ewan Birney
+Nature (2012-09-05) https://doi.org/10.1038/nature11247

+
+
+

162. DNA binding sites: representation and discovery
+G. D. Stormo
+Bioinformatics (2000-01-01) https://doi.org/10.1093/bioinformatics/16.1.16

+
+
+

163. An assessment of neural network and statistical approaches for prediction of E.coli Promoter sites
+Paul B. Horton, Minoru Kanehisa
+Nucleic Acids Research (1992) https://doi.org/10.1093/nar/20.16.4331

+
+
+

164. Enhanced Regulatory Sequence Prediction Using Gapped k-mer Features
+Mahmoud Ghandi, Dongwon Lee, Morteza Mohammad-Noori, Michael A. Beer
+PLoS Computational Biology (2014-07-17) https://doi.org/10.1371/journal.pcbi.1003711

+
+
+

165. SeqGL Identifies Context-Dependent Binding Signals in Genome-Wide Regulatory Element Maps
+Manu Setty, Christina S. Leslie
+PLOS Computational Biology (2015-05-27) https://doi.org/10.1371/journal.pcbi.1004271

+
+
+

166. Predicting the sequence specificities of DNA- and RNA-binding proteins by deep learning
+Babak Alipanahi, Andrew Delong, Matthew T Weirauch, Brendan J Frey
+Nature Biotechnology (2015-07-27) https://doi.org/10.1038/nbt.3300

+
+
+

167. Deep Motif Dashboard: Visualizing and Understanding Genomic Sequences Using Deep Neural Networks
+Jack Lanchantin, Ritambhara Singh, Beilun Wang, Yanjun Qi
+arXiv (2016-08-12) https://arxiv.org/abs/1608.03644v4

+
+
+

168. Convolutional neural network architectures for predicting DNA–protein binding
+Haoyang Zeng, Matthew D. Edwards, Ge Liu, David K. Gifford
+Bioinformatics (2016-06-15) https://doi.org/10.1093/bioinformatics/btw255

+
+
+

169. Reverse-complement parameter sharing improves deep learning models for genomics
+Avanti Shrikumar, Peyton Greenside, Anshul Kundaje
+Cold Spring Harbor Laboratory (2017-01-27) https://doi.org/10.1101/103663

+
+
+

170. Predicting effects of noncoding variants with deep learning–based sequence model
+Jian Zhou, Olga G Troyanskaya
+Nature Methods (2015-08-24) https://doi.org/10.1038/nmeth.3547

+
+
+

171. Learning Important Features Through Propagating Activation Differences
+Avanti Shrikumar, Peyton Greenside, Anshul Kundaje
+arXiv (2017-04-10) https://arxiv.org/abs/1704.02685v1

+
+
+

172. The state of the art of mammalian promoter recognition
+T. Werner
+Briefings in Bioinformatics (2003-01-01) https://doi.org/10.1093/bib/4.1.22

+
+
+

173. Detection of RNA polymerase II promoters and polyadenylation sites in human DNA sequence
+Sherri Matis, Ying Xu, Manesh Shah, Xiaojun Guan, J.Ralph Einstein, Richard Mural, Edward Uberbacher
+Computers & Chemistry (1996-03) https://doi.org/10.1016/s0097-8485(96)80015-5

+
+
+

174. Recognition of prokaryotic and eukaryotic promoters using convolutional deep learning neural networks
+Ramzan Kh. Umarov, Victor V. Solovyev
+PLOS ONE (2017-02-03) https://doi.org/10.1371/journal.pone.0171410

+
+
+

175. Cap analysis gene expression for high-throughput analysis of transcriptional starting point and identification of promoter usage
+T. Shiraki, S. Kondo, S. Katayama, K. Waki, T. Kasukawa, H. Kawaji, R. Kodzius, A. Watahiki, M. Nakamura, T. Arakawa, … Y. Hayashizaki
+Proceedings of the National Academy of Sciences (2003-12-08) https://doi.org/10.1073/pnas.2136655100

+
+
+

176. Genome-wide characterization of transcriptional start sites in humans by integrative transcriptome analysis
+R. Yamashita, N. P. Sathira, A. Kanai, K. Tanimoto, T. Arauchi, Y. Tanaka, S.-i. Hashimoto, S. Sugano, K. Nakai, Y. Suzuki
+Genome Research (2011-03-03) https://doi.org/10.1101/gr.110254.110

+
+
+

177. Enhancers: five essential questions
+Len A. Pennacchio, Wendy Bickmore, Ann Dean, Marcelo A. Nobrega, Gill Bejerano
+Nature Reviews Genetics (2013-03-18) https://doi.org/10.1038/nrg3458

+
+
+

178. A unified architecture of transcriptional regulatory elements
+Robin Andersson, Albin Sandelin, Charles G. Danko
+Trends in Genetics (2015-08) https://doi.org/10.1016/j.tig.2015.05.007

+
+
+

179. Basset: learning the regulatory code of the accessible genome with deep convolutional neural networks
+David R. Kelley, Jasper Snoek, John L. Rinn
+Genome Research (2016-05-03) https://doi.org/10.1101/gr.200535.115

+
+
+

180. DeepEnhancer: Predicting enhancers by convolutional neural networks
+Xu Min, Ning Chen, Ting Chen, Rui Jiang
+2016 IEEE International Conference on Bioinformatics and Biomedicine (BIBM) (2016-12) https://doi.org/10.1109/bibm.2016.7822593

+
+
+

181. Genome-Wide Prediction of cis-Regulatory Regions Using Supervised Deep Learning Methods
+Yifeng Li, Wenqiang Shi, Wyeth W Wasserman
+Cold Spring Harbor Laboratory (2016-02-28) https://doi.org/10.1101/041616

+
+
+

182. Predicting Enhancer-Promoter Interaction from Genomic Sequence with Deep Neural Networks
+Shashank Singh, Yang Yang, Barnabas Poczos, Jian Ma
+Cold Spring Harbor Laboratory (2016-11-02) https://doi.org/10.1101/085241

+
+
+

183. A network-biology perspective of microRNA function and dysfunction in cancer
+Cameron P. Bracken, Hamish S. Scott, Gregory J. Goodall
+Nature Reviews Genetics (2016-10-31) https://doi.org/10.1038/nrg.2016.134

+
+
+

184. Evolution of microRNA diversity and regulation in animals
+Eugene Berezikov
+Nature Reviews Genetics (2011-11-18) https://doi.org/10.1038/nrg3079

+
+
+

185. Predicting effective microRNA target sites in mammalian mRNAs
+Vikram Agarwal, George W Bell, Jin-Wu Nam, David P Bartel
+eLife (2015-08-12) https://doi.org/10.7554/elife.05005

+
+
+

186. deepTarget: End-to-end Learning Framework for microRNA Target Prediction using Deep Recurrent Neural Networks
+Byunghan Lee, Junghwan Baek, Seunghyun Park, Sungroh Yoon
+arXiv (2016-03-30) https://arxiv.org/abs/1603.09123v2

+
+
+

187. deepMiRGene: Deep Neural Network based Precursor microRNA Prediction
+Seunghyun Park, Seonwoo Min, Hyunsoo Choi, Sungroh Yoon
+arXiv (2016-04-29) https://arxiv.org/abs/1605.00017v1

+
+
+

188. AUC-Maximized Deep Convolutional Neural Fields for Protein Sequence Labeling
+Sheng Wang, Siqi Sun, Jinbo Xu
+Machine Learning and Knowledge Discovery in Databases (2016) https://doi.org/10.1007/978-3-319-46227-1_1

+
+
+

189. MetaPSICOV: combining coevolution methods for accurate prediction of contacts and long range hydrogen bonding in proteins
+David T. Jones, Tanya Singh, Tomasz Kosciolek, Stuart Tetchner
+Bioinformatics (2014-11-26) https://doi.org/10.1093/bioinformatics/btu791

+
+
+

190. Identification of direct residue contacts in protein-protein interaction by message passing
+M. Weigt, R. A. White, H. Szurmant, J. A. Hoch, T. Hwa
+Proceedings of the National Academy of Sciences (2008-12-30) https://doi.org/10.1073/pnas.0805923106

+
+
+

191. Protein 3D Structure Computed from Evolutionary Sequence Variation
+Debora S. Marks, Lucy J. Colwell, Robert Sheridan, Thomas A. Hopf, Andrea Pagnani, Riccardo Zecchina, Chris Sander
+PLoS ONE (2011-12-07) https://doi.org/10.1371/journal.pone.0028766

+
+
+

192. A Unified Multitask Architecture for Predicting Local Protein Properties
+Yanjun Qi, Merja Oja, Jason Weston, William Stafford Noble
+PLoS ONE (2012-03-26) https://doi.org/10.1371/journal.pone.0032235

+
+
+

193. Improving prediction of secondary structure, local backbone angles and solvent accessible surface area of proteins by iterative deep learning
+Rhys Heffernan, Kuldip Paliwal, James Lyons, Abdollah Dehzangi, Alok Sharma, Jihua Wang, Abdul Sattar, Yuedong Yang, Yaoqi Zhou
+Scientific Reports (2015-06-22) https://doi.org/10.1038/srep11476

+
+
+

194. Protein secondary structure prediction based on position-specific scoring matrices 1 1Edited by G. Von Heijne
+David T Jones
+Journal of Molecular Biology (1999-09) https://doi.org/10.1006/jmbi.1999.3091

+
+
+

195. Deep Supervised and Convolutional Generative Stochastic Network for Protein Secondary Structure Prediction
+Jian Zhou, Olga G. Troyanskaya
+arXiv (2014-03-06) https://arxiv.org/abs/1403.1347v1

+
+
+

196. Protein contact prediction by integrating joint evolutionary coupling analysis and supervised learning
+Jianzhu Ma, Sheng Wang, Zhiyong Wang, Jinbo Xu
+Bioinformatics (2015-08-14) https://doi.org/10.1093/bioinformatics/btv472

+
+
+

197. Deep architectures for protein contact map prediction
+Pietro Di Lena, Ken Nagata, Pierre Baldi
+Bioinformatics (2012-07-30) https://doi.org/10.1093/bioinformatics/bts475

+
+
+

198. Predicting protein residue–residue contacts using deep networks and boosting
+Jesse Eickholt, Jianlin Cheng
+Bioinformatics (2012-10-09) https://doi.org/10.1093/bioinformatics/bts598

+
+
+

199. Improved Contact Predictions Using the Recognition of Protein Like Contact Patterns
+Marcin J. Skwark, Daniele Raimondi, Mirco Michel, Arne Elofsson
+PLoS Computational Biology (2014-11-06) https://doi.org/10.1371/journal.pcbi.1003889

+
+ +
+

201. CAMEO - Continuous Automated Model Evaluation(2017) http://www.cameo3d.org/

+
+
+

202. Predicting membrane protein contacts from non-membrane proteins by deep transfer learning
+Zhen Li, Sheng Wang, Yizhou Yu, Jinbo Xu
+arXiv (2017-04-24) https://arxiv.org/abs/1704.07207v1

+
+
+

203. Deep Learning Automates the Quantitative Analysis of Individual Cells in Live-Cell Imaging Experiments
+David A. Van Valen, Takamasa Kudo, Keara M. Lane, Derek N. Macklin, Nicolas T. Quach, Mialy M. DeFelice, Inbal Maayan, Yu Tanouchi, Euan A. Ashley, Markus W. Covert
+PLOS Computational Biology (2016-11-04) https://doi.org/10.1371/journal.pcbi.1005177

+
+
+

204. U-Net: Convolutional Networks for Biomedical Image Segmentation
+Olaf Ronneberger, Philipp Fischer, Thomas Brox
+Lecture Notes in Computer Science (2015) https://doi.org/10.1007/978-3-319-24574-4_28

+
+
+

205. Prospective identification of hematopoietic lineage choice by deep learning
+Felix Buggenthin, Florian Buettner, Philipp S Hoppe, Max Endele, Manuel Kroiss, Michael Strasser, Michael Schwarzfischer, Dirk Loeffler, Konstantinos D Kokkaliaris, Oliver Hilsenbeck, … Carsten Marr
+Nature Methods (2017-02-20) https://doi.org/10.1038/nmeth.4182

+
+
+

206. Reconstructing cell cycle and disease progression using deep learning
+Philipp Eulenberg, Niklas Koehler, Thomas Blasi, Andrew Filby, Anne E. Carpenter, Paul Rees, Fabian J. Theis, F. Alexander Wolf
+Cold Spring Harbor Laboratory (2016-10-17) https://doi.org/10.1101/081364

+
+
+

207. Automating Morphological Profiling with Generic Deep Convolutional Networks
+Nick Pawlowski, Juan C Caicedo, Shantanu Singh, Anne E Carpenter, Amos Storkey
+Cold Spring Harbor Laboratory (2016-11-02) https://doi.org/10.1101/085118

+
+
+

208. Generative Modeling with Conditional Autoencoders: Building an Integrated Cell
+Gregory R. Johnson, Rory M. Donovan-Maiye, Mary M. Maleckar
+arXiv (2017-04-28) https://arxiv.org/abs/1705.00092v1

+
+
+

209. Applications in image-based profiling of perturbations
+Juan C Caicedo, Shantanu Singh, Anne E Carpenter
+Current Opinion in Biotechnology (2016-06) https://doi.org/10.1016/j.copbio.2016.04.003

+
+
+

210. Large-scale image-based screening and profiling of cellular phenotypes
+Nicola Bougen-Zhukov, Sheng Yang Loh, Hwee Kuan Lee, Lit-Hsin Loo
+Cytometry Part A (2016-07-19) https://doi.org/10.1002/cyto.a.22909

+
+
+

211. Machine learning and computer vision approaches for phenotypic profiling
+Ben T. Grys, Dara S. Lo, Nil Sahin, Oren Z. Kraus, Quaid Morris, Charles Boone, Brenda J. Andrews
+The Journal of Cell Biology (2016-12-09) https://doi.org/10.1083/jcb.201610026

+
+
+

212. Single-cell genome sequencing: current state of the science
+Charles Gawad, Winston Koh, Stephen R. Quake
+Nature Reviews Genetics (2016-01-25) https://doi.org/10.1038/nrg.2015.16

+
+
+

213. Somatic mutation in single human neurons tracks developmental and transcriptional history
+M. A. Lodato, M. B. Woodworth, S. Lee, G. D. Evrony, B. K. Mehta, A. Karger, S. Lee, T. W. Chittenden, A. M. D’Gama, X. Cai, … C. A. Walsh
+Science (2015-10-01) https://doi.org/10.1126/science.aab1785

+
+
+

214. Single-cell transcriptome sequencing: recent advances and remaining challenges
+Serena Liu, Cole Trapnell
+F1000Research (2016-02-17) https://doi.org/10.12688/f1000research.7223.1

+
+
+

215. Single-Cell and Single-Molecule Analysis of Gene Expression Regulation
+Maria Vera, Jeetayu Biswas, Adrien Senecal, Robert H. Singer, Hye Yoon Park
+Annual Review of Genetics (2016-11-23) https://doi.org/10.1146/annurev-genet-120215-034854

+
+
+

216. Joint Profiling Of Chromatin Accessibility, DNA Methylation And Transcription In Single Cells
+Stephen J. Clark, Ricard Argelaguet, Chantriolnt-Andreas Kapourani, Thomas M. Stubbs, Heather J. Lee, Felix Krueger, Guido Sanguinetti, Gavin Kelsey, John C. Marioni, Oliver Stegle, Wolf Reik
+Cold Spring Harbor Laboratory (2017-05-17) https://doi.org/10.1101/138685

+
+
+

217. DeepCpG: accurate prediction of single-cell DNA methylation states using deep learning
+Christof Angermueller, Heather J. Lee, Wolf Reik, Oliver Stegle
+Genome Biology (2017-04-11) https://doi.org/10.1186/s13059-017-1189-z

+
+
+

218. Denoising genome-wide histone ChIP-seq with convolutional neural networks
+Pang Wei Koh, Emma Pierson, Anshul Kundaje
+Cold Spring Harbor Laboratory (2016-05-07) https://doi.org/10.1101/052118

+
+
+

219. Removal of Batch Effects using Distribution-Matching Residual Networks
+Uri Shaham, Kelly P. Stanton, Jun Zhao, Huamin Li, Khadir Raddassi, Ruth Montgomery, Yuval Kluger
+arXiv (2016-10-13) https://arxiv.org/abs/1610.04181v5

+
+
+

220. Single-Cell Genomics Unveils Critical Regulators of Th17 Cell Pathogenicity
+Jellert T. Gaublomme, Nir Yosef, Youjin Lee, Rona S. Gertner, Li V. Yang, Chuan Wu, Pier Paolo Pandolfi, Tak Mak, Rahul Satija, Alex K. Shalek, … Aviv Regev
+Cell (2015-12) https://doi.org/10.1016/j.cell.2015.11.009

+
+
+

221. Sensitive detection of rare disease-associated cell subsets via representation learning.
+Eirini Arvaniti, Manfred Claassen
+Cold Spring Harbor Laboratory (2016-03-31) https://doi.org/10.1101/046508

+
+
+

222. Deep Residual Learning for Image Recognition
+Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun
+arXiv (2015-12-10) https://arxiv.org/abs/1512.03385v1

+
+
+

223. Reversed graph embedding resolves complex single-cell developmental trajectories
+Xiaojie Qiu, Qi Mao, Ying Tang, Li Wang, Raghav Chawla, Hannah Pliner, Cole Trapnell
+Cold Spring Harbor Laboratory (2017-02-21) https://doi.org/10.1101/110668

+
+
+

224. Mastering the game of Go with deep neural networks and tree search
+David Silver, Aja Huang, Chris J. Maddison, Arthur Guez, Laurent Sifre, George van den Driessche, Julian Schrittwieser, Ioannis Antonoglou, Veda Panneershelvam, Marc Lanctot, … Demis Hassabis
+Nature (2016-01-27) https://doi.org/10.1038/nature16961

+
+
+

225. Compositional biases of bacterial genomes and evolutionary implications.
+S Karlin, J Mrázek, AM Campbell
+Journal of Bacteriology (1997-06) https://doi.org/10.1128/jb.179.12.3899-3913.1997

+
+
+

226. Accurate phylogenetic classification of variable-length DNA fragments
+Alice Carolyn McHardy, Héctor García Martín, Aristotelis Tsirigos, Philip Hugenholtz, Isidore Rigoutsos
+Nature Methods (2006-12-10) https://doi.org/10.1038/nmeth976

+
+
+

227. NBC: the Naive Bayes Classification tool webserver for taxonomic classification of metagenomic reads
+G. L. Rosen, E. R. Reichenberger, A. M. Rosenfeld
+Bioinformatics (2010-11-08) https://doi.org/10.1093/bioinformatics/btq619

+
+
+

228. Informatics for Unveiling Hidden Genome Signatures
+T. Abe
+Genome Research (2003-04-01) https://doi.org/10.1101/gr.634603

+
+
+

229. Metagenomic microbial community profiling using unique clade-specific marker genes
+Nicola Segata, Levi Waldron, Annalisa Ballarini, Vagheesh Narasimhan, Olivier Jousson, Curtis Huttenhower
+Nature Methods (2012-06-10) https://doi.org/10.1038/nmeth.2066

+
+
+

230. WGSQuikr: Fast Whole-Genome Shotgun Metagenomic Classification
+David Koslicki, Simon Foucart, Gail Rosen
+PLoS ONE (2014-03-13) https://doi.org/10.1371/journal.pone.0091784

+
+
+

231. Scalable metagenomic taxonomy classification using a reference genome database
+Sasha K. Ames, David A. Hysom, Shea N. Gardner, G. Scott Lloyd, Maya B. Gokhale, Jonathan E. Allen
+Bioinformatics (2013-07-04) https://doi.org/10.1093/bioinformatics/btt389

+
+
+

232. Large-scale machine learning for metagenomics sequence classification
+Kévin Vervier, Pierre Mahé, Maud Tournoud, Jean-Baptiste Veyrieras, Jean-Philippe Vert
+Bioinformatics (2015-11-20) https://doi.org/10.1093/bioinformatics/btv683

+
+
+

233. Combining gene prediction methods to improve metagenomic gene annotation
+Non G Yok, Gail L Rosen
+BMC Bioinformatics (2011) https://doi.org/10.1186/1471-2105-12-20

+
+
+

234. Machine learning for metagenomics: methods and tools
+Hayssam Soueidan, Macha Nikolski
+Metagenomics (2017-01-01) https://doi.org/10.1515/metgen-2016-0001

+
+
+

235. Utilizing Machine Learning Approaches to Understand the Interrelationship of Diet, the Human Gastrointestinal Microbiome, and Health
+Heather Guetterman, Loretta Auvil, Nate Russell, Michael Welge, Matt Berry, Lisa Gatzke, Colleen Bushell, Hannah Holscher
+The FASEB Journal (2016-04-01) http://www.fasebj.org/content/30/1_Supplement/406.3

+
+
+

236. Supervised classification of human microbiota
+Dan Knights, Elizabeth K. Costello, Rob Knight
+FEMS Microbiology Reviews (2011-03) https://doi.org/10.1111/j.1574-6976.2010.00251.x

+
+
+

237. A comprehensive evaluation of multicategory classification methods for microbiomic data
+Alexander Statnikov, Mikael Henaff, Varun Narendra, Kranti Konganti, Zhiguo Li, Liying Yang, Zhiheng Pei, Martin J Blaser, Constantin F Aliferis, Alexander V Alekseyenko
+Microbiome (2013) https://doi.org/10.1186/2049-2618-1-11

+
+
+

238. Machine Learning Meta-analysis of Large Metagenomic Datasets: Tools and Biological Insights
+Edoardo Pasolli, Duy Tin Truong, Faizan Malik, Levi Waldron, Nicola Segata
+PLOS Computational Biology (2016-07-11) https://doi.org/10.1371/journal.pcbi.1004977

+
+
+

239. DectICO: an alignment-free supervised metagenomic classification method based on feature extraction and dynamic selection
+Xiao Ding, Fudong Cheng, Changchang Cao, Xiao Sun
+BMC Bioinformatics (2015-10-07) https://doi.org/10.1186/s12859-015-0753-3

+
+
+

240. Correction: Class Prediction and Feature Selection with Linear Optimization for Metagenomic Count Data
+Zhenqiu Liu, Dechang Chen, Li Sheng, Amy Y. Liu
+PLoS ONE (2014-05-12) https://doi.org/10.1371/journal.pone.0097958

+
+
+

241. Fizzy: feature subset selection for metagenomics
+Gregory Ditzler, J. Calvin Morrison, Yemin Lan, Gail L. Rosen
+BMC Bioinformatics (2015-11-04) https://doi.org/10.1186/s12859-015-0793-8

+
+
+

242. A Bootstrap Based Neyman-Pearson Test for Identifying Variable Importance
+Gregory Ditzler, Robi Polikar, Gail Rosen
+IEEE Transactions on Neural Networks and Learning Systems (2015-04) https://doi.org/10.1109/tnnls.2014.2320415

+
+
+

243. Orphelia: predicting genes in metagenomic sequencing reads
+Katharina J. Hoff, Thomas Lingner, Peter Meinicke, Maike Tech
+Nucleic Acids Research (2009-05-08) https://doi.org/10.1093/nar/gkp327

+
+
+

244. FragGeneScan: predicting genes in short and error-prone reads
+Mina Rho, Haixu Tang, Yuzhen Ye
+Nucleic Acids Research (2010-08-28) https://doi.org/10.1093/nar/gkq747

+
+
+

245. Continuous Distributed Representation of Biological Sequences for Deep Proteomics and Genomics
+Ehsaneddin Asgari, Mohammad R. K. Mofrad
+PLOS ONE (2015-11-10) https://doi.org/10.1371/journal.pone.0141287

+
+
+

246. Fast model-based protein homology detection without alignment
+S. Hochreiter, M. Heusel, K. Obermayer
+Bioinformatics (2007-05-08) https://doi.org/10.1093/bioinformatics/btm247

+
+
+

247. Convolutional LSTM Networks for Subcellular Localization of Proteins
+Søren Kaae Sønderby, Casper Kaae Sønderby, Henrik Nielsen, Ole Winther
+arXiv (2015-03-06) https://arxiv.org/abs/1503.01919v1

+
+
+

248. Neural network-based taxonomic clustering for metagenomics
+Steven D. Essinger, Robi Polikar, Gail L. Rosen
+The 2010 International Joint Conference on Neural Networks (IJCNN) (2010-07) https://doi.org/10.1109/ijcnn.2010.5596644

+
+
+

249. Clustering metagenomic sequences with interpolated Markov models
+David R Kelley, Steven L Salzberg
+BMC Bioinformatics (2010) https://doi.org/10.1186/1471-2105-11-544

+
+
+

250. METAGENOMIC TAXONOMIC CLASSIFICATION USING EXTREME LEARNING MACHINES
+ZEEHASHAM RASHEED, HUZEFA RANGWALA
+Journal of Bioinformatics and Computational Biology (2012-10) https://doi.org/10.1142/s0219720012500151

+
+
+

251. Globoko ucenje na genomskih in filogenetskih podatkih
+Nina Mrzelj
+Univerza v Ljubljani, Fakulteta za racunalništvo in informatiko (2016) https://repozitorij.uni-lj.si/IzpisGradiva.php?id=85515

+
+
+

252. Influence of microbiome species in hard-to-heal wounds on disease severity and treatment duration
+Dagmar Chudobova, Kristyna Cihalova, Roman Guran, Simona Dostalova, Kristyna Smerkova, Radek Vesely, Jaromir Gumulec, Michal Masarik, Zbynek Heger, Vojtech Adam, Rene Kizek
+The Brazilian Journal of Infectious Diseases (2015-11) https://doi.org/10.1016/j.bjid.2015.08.013

+
+
+

253. Multi-Layer and Recursive Neural Networks for Metagenomic Classification
+Gregory Ditzler, Robi Polikar, Gail Rosen
+IEEE Transactions on NanoBioscience (2015-09) https://doi.org/10.1109/tnb.2015.2461219

+
+
+

254. TensorFlow vs. scikit-learn : The Microbiome Challenge
+Ali A. Faruqi
+Ali A. Faruqi (2016-07-09) http://alifar76.github.io/sklearn-metrics/

+
+
+

255. Advances in Optimizing Recurrent Networks
+Yoshua Bengio, Nicolas Boulanger-Lewandowski, Razvan Pascanu
+arXiv (2012-12-04) https://arxiv.org/abs/1212.0901v2

+
+
+

256. DeepNano: Deep Recurrent Neural Networks for Base Calling in MinION Nanopore Reads
+Vladimír Boža, Broňa Brejová, Tomáš Vinař
+arXiv (2016-03-30) https://arxiv.org/abs/1603.09195v1

+
+
+

257. Sequence to Sequence Learning with Neural Networks
+Ilya Sutskever, Oriol Vinyals, Quoc V. Le
+arXiv (2014-09-10) https://arxiv.org/abs/1409.3215v3

+
+
+

258. Creating a universal SNP and small indel variant caller with deep neural networks
+Ryan Poplin, Dan Newburger, Jojo Dijamco, Nam Nguyen, Dion Loy, Sam S. Gross, Cory Y. McLean, Mark A. DePristo
+Cold Spring Harbor Laboratory (2016-12-14) https://doi.org/10.1101/092890

+
+
+

259. Training Genotype Callers with Neural Networks
+Rémi Torracinta, Fabien Campagne
+Cold Spring Harbor Laboratory (2016-12-30) https://doi.org/10.1101/097469

+
+
+

260. Xception: Deep Learning with Depthwise Separable Convolutions
+François Chollet
+arXiv (2016-10-07) https://arxiv.org/abs/1610.02357v3

+
+
+

261. Adaptive Somatic Mutations Calls with Deep Learning and Semi-Simulated Data
+Remi Torracinta, Laurent Mesnard, Susan Levine, Rita Shaknovich, Maureen Hanson, Fabien Campagne
+Cold Spring Harbor Laboratory (2016-10-04) https://doi.org/10.1101/079087

+
+
+

262. The Path to Personalized Medicine
+Margaret A. Hamburg, Francis S. Collins
+New England Journal of Medicine (2010-07-22) https://doi.org/10.1056/nejmp1006304

+
+
+

263. Biomedical Informatics for Computer-Aided Decision Support Systems: A Survey
+Ashwin Belle, Mark A. Kon, Kayvan Najarian
+The Scientific World Journal (2013) https://doi.org/10.1155/2013/769639

+
+
+

264. Advantages and disadvantages of using artificial neural networks versus logistic regression for predicting medical outcomes
+Jack V. Tu
+Journal of Clinical Epidemiology (1996-11) https://doi.org/10.1016/s0895-4356(96)00002-9

+
+
+

265. Use of an Artificial Neural Network for the Diagnosis of Myocardial Infarction
+William G. Baxt
+Annals of Internal Medicine (1991-12-01) https://doi.org/10.7326/0003-4819-115-11-843

+
+
+

266. Clinical Prediction Rules
+John H. Wasson, Harold C. Sox, Raymond K. Neff, Lee Goldman
+New England Journal of Medicine (1985-09-26) https://doi.org/10.1056/nejm198509263131306

+
+
+

267. The use of artificial neural networks in decision support in cancer: A systematic review
+Paulo J. Lisboa, Azzam F.G. Taktak
+Neural Networks (2006-05) https://doi.org/10.1016/j.neunet.2005.10.007

+
+
+

268. Estimating causal effects of treatments in randomized and nonrandomized studies.
+Donald B. Rubin
+Journal of Educational Psychology (1974) https://doi.org/10.1037/h0037350

+
+
+

269. Learning Representations for Counterfactual Inference
+Fredrik D. Johansson, Uri Shalit, David Sontag
+arXiv (2016-05-12) https://arxiv.org/abs/1605.03661v2

+
+
+

270. Causal Phenotype Discovery via Deep Networks
+David C. Kale, Zhengping Che, Mohammad Taha Bahadori, Wenzhe Li, Yan Liu, Randall Wetzel
+AMIA Annual Symposium Proceedings (2015) https://www.ncbi.nlm.nih.gov/pubmed/26958203

+
+
+

271. Modeling Missing Data in Clinical Time Series with RNNs
+Zachary C. Lipton, David C. Kale, Randall Wetzel
+arXiv (2016-06-13) https://arxiv.org/abs/1606.04130v5

+
+
+

272. Recurrent Neural Networks for Multivariate Time Series with Missing Values
+Zhengping Che, Sanjay Purushotham, Kyunghyun Cho, David Sontag, Yan Liu
+arXiv (2016-06-06) https://arxiv.org/abs/1606.01865v2

+
+
+

273. Predicting Complications in Critical Care Using Heterogeneous Clinical Data
+Vijay Huddar, Bapu Koundinya Desiraju, Vaibhav Rajan, Sakyajit Bhattacharya, Shourya Roy, Chandan K. Reddy
+IEEE Access (2016) https://doi.org/10.1109/access.2016.2618775

+
+
+

274. Phenotyping of Clinical Time Series with LSTM Recurrent Neural Networks
+Zachary C. Lipton, David C. Kale, Randall C. Wetzel
+arXiv (2015-10-26) https://arxiv.org/abs/1510.07641v2

+
+
+

275. Optimal medication dosing from suboptimal clinical examples: A deep reinforcement learning approach
+Shamim Nemati, Mohammad M. Ghassemi, Gari D. Clifford
+2016 38th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) (2016-08) https://doi.org/10.1109/embc.2016.7591355

+
+
+

276. From vital signs to clinical outcomes for patients with sepsis: a machine learning basis for a clinical decision support system
+Eren Gultepe, Jeffrey P Green, Hien Nguyen, Jason Adams, Timothy Albertson, Ilias Tagkopoulos
+Journal of the American Medical Informatics Association (2014-03) https://doi.org/10.1136/amiajnl-2013-001815

+
+
+

277. Imaging-based enrichment criteria using deep learning algorithms for efficient clinical trials in mild cognitive impairment
+Vamsi K. Ithapu, Vikas Singh, Ozioma C. Okonkwo, Richard J. Chappell, N. Maritza Dowling, Sterling C. Johnson
+Alzheimer’s & Dementia (2015-12) https://doi.org/10.1016/j.jalz.2015.01.010

+
+
+

278. Integrated deep learned transcriptomic and structure-based predictor of clinical trials outcomes
+Artem V Artemov, Evgeny Putin, Quentin Vanhaelen, Alexander Aliper, Ivan V Ozerov, Alex Zhavoronkov
+Cold Spring Harbor Laboratory (2016-12-20) https://doi.org/10.1101/095653

+
+
+

279. Innovation in the pharmaceutical industry: New estimates of R&D costs
+Joseph A. DiMasi, Henry G. Grabowski, Ronald W. Hansen
+Journal of Health Economics (2016-05) https://doi.org/10.1016/j.jhealeco.2016.01.012

+
+
+

280. An analysis of the attrition of drug candidates from four major pharmaceutical companies
+Michael J. Waring, John Arrowsmith, Andrew R. Leach, Paul D. Leeson, Sam Mandrell, Robert M. Owen, Garry Pairaudeau, William D. Pennie, Stephen D. Pickett, Jibo Wang, … Alex Weir
+Nature Reviews Drug Discovery (2015-06-19) https://doi.org/10.1038/nrd4609

+
+
+

281. The Connectivity Map: Using Gene-Expression Signatures to Connect Small Molecules, Genes, and Disease
+J. Lamb
+Science (2006-09-29) https://doi.org/10.1126/science.1132939

+
+
+

282. A survey of current trends in computational drug repositioning
+Jiao Li, Si Zheng, Bin Chen, Atul J. Butte, S. Joshua Swamidass, Zhiyong Lu
+Briefings in Bioinformatics (2015-03-31) https://doi.org/10.1093/bib/bbv020

+
+
+

283. A review of connectivity map and computational approaches in pharmacogenomics
+Aliyu Musa, Laleh Soltan Ghoraie, Shu-Dong Zhang, Galina Galzko, Olli Yli-Harja, Matthias Dehmer, Benjamin Haibe-Kains, Frank Emmert-Streib
+Briefings in Bioinformatics (2017-01-09) https://doi.org/10.1093/bib/bbw112

+
+
+

284. A review of validation strategies for computational drug repositioning
+Adam S. Brown, Chirag J. Patel
+Briefings in Bioinformatics (2016-11-22) https://academic.oup.com/bib/article/doi/10.1093/bib/bbw110/2562646/A-review-of-validation-strategies-for

+
+
+

285. Drug Repositioning: A Machine-Learning Approach through Data Integration
+Francesco Napolitano, Yan Zhao, Vania M Moreira, Roberto Tagliaferri, Juha Kere, Mauro D’Amato, Dario Greco
+Journal of Cheminformatics (2013) https://doi.org/10.1186/1758-2946-5-30

+
+
+

286. Drug–Disease Association and Drug-Repositioning Predictions in Complex Diseases Using Causal Inference–Probabilistic Matrix Factorization
+Jihong Yang, Zheng Li, Xiaohui Fan, Yiyu Cheng
+Journal of Chemical Information and Modeling (2014-09-22) https://doi.org/10.1021/ci500340n

+
+
+

287. Drug repositioning for non-small cell lung cancer by using machine learning algorithms and topological graph theory
+Chien-Hung Huang, Peter Mu-Hsin Chang, Chia-Wei Hsu, Chi-Ying F. Huang, Ka-Lok Ng
+BMC Bioinformatics (2016-01-11) https://doi.org/10.1186/s12859-015-0845-0

+
+
+

288. Machine Learning Prediction of Cancer Cell Sensitivity to Drugs Based on Genomic and Chemical Properties
+Michael P. Menden, Francesco Iorio, Mathew Garnett, Ultan McDermott, Cyril H. Benes, Pedro J. Ballester, Julio Saez-Rodriguez
+PLoS ONE (2013-04-30) https://doi.org/10.1371/journal.pone.0061318

+
+
+

289. Large-scale integration of small molecule-induced genome-wide transcriptional responses, Kinome-wide binding affinities and cell-growth inhibition profiles reveal global trends characterizing systems-level drug action
+Dušica Vidović, Amar Koleti, Stephan C. Schürer
+Frontiers in Genetics (2014-09-30) https://doi.org/10.3389/fgene.2014.00342

+
+
+

290. Computational Discovery of Putative Leads for Drug Repositioning through Drug-Target Interaction Prediction
+Edgar D. Coelho, Joel P. Arrais, José Luís Oliveira
+PLOS Computational Biology (2016-11-28) https://doi.org/10.1371/journal.pcbi.1005219

+
+
+

291. Large-Scale Off-Target Identification Using Fast and Accurate Dual Regularized One-Class Collaborative Filtering and Its Application to Drug Repurposing
+Hansaim Lim, Aleksandar Poleksic, Yuan Yao, Hanghang Tong, Di He, Luke Zhuang, Patrick Meng, Lei Xie
+PLOS Computational Biology (2016-10-07) https://doi.org/10.1371/journal.pcbi.1005135

+
+
+

292. Pairwise input neural network for target-ligand interaction prediction
+Caihua Wang, Juan Liu, Fei Luo, Yafang Tan, Zixin Deng, Qian-Nan Hu
+2014 IEEE International Conference on Bioinformatics and Biomedicine (BIBM) (2014-11) https://doi.org/10.1109/bibm.2014.6999129

+
+
+

293. L1000CDS2: LINCS L1000 characteristic direction signatures search engine
+Qiaonan Duan, St Patrick Reid, Neil R Clark, Zichen Wang, Nicolas F Fernandez, Andrew D Rouillard, Ben Readhead, Sarah R Tritsch, Rachel Hodos, Marc Hafner, … Avi Ma’ayan
+npj Systems Biology and Applications (2016-08-04) https://doi.org/10.1038/npjsba.2016.15

+
+
+

294. A guide to drug discovery: Hit and lead generation: beyond high-throughput screening
+Konrad H. Bleicher, Hans-Joachim Böhm, Klaus Müller, Alexander I. Alanine
+Nature Reviews Drug Discovery (2003-05) https://doi.org/10.1038/nrd1086

+
+
+

295. Hit discovery and hit-to-lead approaches
+György M. Keserű, Gergely M. Makara
+Drug Discovery Today (2006-08) https://doi.org/10.1016/j.drudis.2006.06.016

+
+
+

296. Influence Relevance Voting: An Accurate And Interpretable Virtual High Throughput Screening Method
+S. Joshua Swamidass, Chloé-Agathe Azencott, Ting-Wan Lin, Hugo Gramajo, Shiou-Chuan Tsai, Pierre Baldi
+Journal of Chemical Information and Modeling (2009-04-27) https://doi.org/10.1021/ci8004379

+
+
+

297. Modeling Industrial ADMET Data with Multitask Networks
+Steven Kearnes, Brian Goldman, Vijay Pande
+arXiv (2016-06-28) https://arxiv.org/abs/1606.08793v3

+
+
+

298. XenoSite: Accurately Predicting CYP-Mediated Sites of Metabolism with Neural Networks
+Jed Zaretzki, Matthew Matlock, S. Joshua Swamidass
+Journal of Chemical Information and Modeling (2013-12-23) https://doi.org/10.1021/ci400518g

+
+
+

299. Molecular Descriptors for ChemoinformaticsMethods and Principles in Medicinal Chemistry (2009-07-15) https://doi.org/10.1002/9783527628766

+
+
+

300. Multi-task Neural Networks for QSAR Predictions
+George E. Dahl, Navdeep Jaitly, Ruslan Salakhutdinov
+arXiv (2014-06-04) https://arxiv.org/abs/1406.1231v1

+
+
+

301. Deep Neural Nets as a Method for Quantitative Structure–Activity Relationships
+Junshui Ma, Robert P. Sheridan, Andy Liaw, George E. Dahl, Vladimir Svetnik
+Journal of Chemical Information and Modeling (2015-02-23) https://doi.org/10.1021/ci500747n

+
+
+

302. Deep learning as an opportunity in virtual screening
+Thomas Unterthiner, Andreas Mayr, Günter Klambauer, Marvin Steijaert, Jörg K. Wegner, Hugo Ceulemans, Sepp Hochreiter
+Neural Information Processing Systems 2014: Deep Learning and Representation Learning Workshop (2014) http://www.dlworkshop.org/23.pdf?attredirects=0

+
+
+

303. Massively Multitask Networks for Drug Discovery
+Bharath Ramsundar, Steven Kearnes, Patrick Riley, Dale Webster, David Konerding, Vijay Pande
+arXiv (2015-02-06) https://arxiv.org/abs/1502.02072v1

+
+
+

304. DeepTox: Toxicity Prediction using Deep Learning
+Andreas Mayr, Günter Klambauer, Thomas Unterthiner, Sepp Hochreiter
+Frontiers in Environmental Science (2016-02-02) https://doi.org/10.3389/fenvs.2015.00080

+
+
+

305. Computational Modeling of β-Secretase 1 (BACE-1) Inhibitors Using Ligand Based Approaches
+Govindan Subramanian, Bharath Ramsundar, Vijay Pande, Rajiah Aldrin Denny
+Journal of Chemical Information and Modeling (2016-10-24) https://doi.org/10.1021/acs.jcim.6b00290

+
+
+

306. The enumeration of chemical space
+Jean-Louis Reymond, Lars Ruddigkeit, Lorenz Blum, Ruud van Deursen
+Wiley Interdisciplinary Reviews: Computational Molecular Science (2012-04-18) https://doi.org/10.1002/wcms.1104

+
+
+

307. Accurate and efficient target prediction using a potency-sensitive influence-relevance voter
+Alessandro Lusci, David Fooshee, Michael Browning, Joshua Swamidass, Pierre Baldi
+Journal of Cheminformatics (2015-12) https://doi.org/10.1186/s13321-015-0110-6

+
+
+

308. Extended-Connectivity Fingerprints
+David Rogers, Mathew Hahn
+Journal of Chemical Information and Modeling (2010-05-24) https://doi.org/10.1021/ci100050t

+
+
+

309. Convolutional Networks on Graphs for Learning Molecular Fingerprints
+David K. Duvenaud, Dougal Maclaurin, Jorge Iparraguirre, Rafael Bombarell, Timothy Hirzel, Alan Aspuru-Guzik, Ryan P. Adams
+(2015) http://papers.nips.cc/paper/5954-convolutional-networks-on-graphs-for-learning-molecular-fingerprints

+
+
+

310. Deep Architectures and Deep Learning in Chemoinformatics: The Prediction of Aqueous Solubility for Drug-Like Molecules
+Alessandro Lusci, Gianluca Pollastri, Pierre Baldi
+Journal of Chemical Information and Modeling (2013-07-22) https://doi.org/10.1021/ci400187y

+
+
+

311. Molecular graph convolutions: moving beyond fingerprints
+Steven Kearnes, Kevin McCloskey, Marc Berndl, Vijay Pande, Patrick Riley
+Journal of Computer-Aided Molecular Design (2016-08) https://doi.org/10.1007/s10822-016-9938-8

+
+
+

312. Low Data Drug Discovery with One-Shot Learning
+Han Altae-Tran, Bharath Ramsundar, Aneesh S. Pappu, Vijay Pande
+ACS Central Science (2017-04-03) https://doi.org/10.1021/acscentsci.6b00367

+
+
+

313. MoleculeNet: A Benchmark for Molecular Machine Learning
+Zhenqin Wu, Bharath Ramsundar, Evan N. Feinberg, Joseph Gomes, Caleb Geniesse, Aneesh S. Pappu, Karl Leswing, Vijay Pande
+arXiv (2017-03-02) https://arxiv.org/abs/1703.00564v2

+
+
+

314. deepchem/deepchemGitHub (2017) https://github.com/deepchem/deepchem

+
+
+

315. Automatic chemical design using a data-driven continuous representation of molecules
+Rafael Gómez-Bombarelli, David Duvenaud, José Miguel Hernández-Lobato, Jorge Aguilera-Iparraguirre, Timothy D. Hirzel, Ryan P. Adams, Alán Aspuru-Guzik
+arXiv (2016-10-07) https://arxiv.org/abs/1610.02415v2

+
+
+

316. Structure-Based Virtual Screening for Drug Discovery: a Problem-Centric Review
+Tiejun Cheng, Qingliang Li, Zhigang Zhou, Yanli Wang, Stephen H. Bryant
+The AAPS Journal (2012-01-27) https://doi.org/10.1208/s12248-012-9322-0

+
+
+

317. Atomic Convolutional Networks for Predicting Protein-Ligand Binding Affinity
+Joseph Gomes, Bharath Ramsundar, Evan N. Feinberg, Vijay S. Pande
+arXiv (2017-03-30) https://arxiv.org/abs/1703.10603v1

+
+
+

318. The PDBbind Database:  Methodologies and Updates
+Renxiao Wang, Xueliang Fang, Yipin Lu, Chao-Yie Yang, Shaomeng Wang
+Journal of Medicinal Chemistry (2005-06) https://doi.org/10.1021/jm048957q

+
+
+

319. Boosting Docking-Based Virtual Screening with Deep Learning
+Janaina Cruz Pereira, Ernesto Raúl Caffarena, Cicero Nogueira dos Santos
+Journal of Chemical Information and Modeling (2016-12-27) https://doi.org/10.1021/acs.jcim.6b00355

+
+
+

320. Protein-Ligand Scoring with Convolutional Neural Networks
+Matthew Ragoza, Joshua Hochuli, Elisa Idrobo, Jocelyn Sunseri, David Ryan Koes
+arXiv (2016-12-08) https://arxiv.org/abs/1612.02751v1

+
+
+

321. Enabling future drug discovery byde novodesign
+Markus Hartenfeller, Gisbert Schneider
+Wiley Interdisciplinary Reviews: Computational Molecular Science (2011-04-25) https://doi.org/10.1002/wcms.49

+
+
+

322. De Novo Design at the Edge of Chaos
+Petra Schneider, Gisbert Schneider
+Journal of Medicinal Chemistry (2016-05-12) https://doi.org/10.1021/acs.jmedchem.5b01849

+
+
+

323. Generating Sequences With Recurrent Neural Networks
+Alex Graves
+arXiv (2013-08-04) https://arxiv.org/abs/1308.0850v5

+
+
+

324. Generating Focussed Molecule Libraries for Drug Discovery with Recurrent Neural Networks
+Marwin H. S. Segler, Thierry Kogej, Christian Tyrchan, Mark P. Waller
+arXiv (2017-01-05) https://arxiv.org/abs/1701.01329v1

+
+
+

325. Grammar Variational Autoencoder
+Matt J. Kusner, Brooks Paige, José Miguel Hernández-Lobato
+arXiv (2017-03-06) https://arxiv.org/abs/1703.01925v1

+
+
+

326. ChEMBL: a large-scale bioactivity database for drug discovery
+A. Gaulton, L. J. Bellis, A. P. Bento, J. Chambers, M. Davies, A. Hersey, Y. Light, S. McGlinchey, D. Michalovich, B. Al-Lazikani, J. P. Overington
+Nucleic Acids Research (2011-09-23) https://doi.org/10.1093/nar/gkr777

+
+
+

327. Molecular De Novo Design through Deep Reinforcement Learning
+Marcus Olivecrona, Thomas Blaschke, Ola Engkvist, Hongming Chen
+arXiv (2017-04-25) https://arxiv.org/abs/1704.07555v2

+
+
+

328. Sequence Tutor: Conservative Fine-Tuning of Sequence Generation Models with KL-control
+Natasha Jaques, Shixiang Gu, Dzmitry Bahdanau, José Miguel Hernández-Lobato, Richard E. Turner, Douglas Eck
+arXiv (2016-11-09) https://arxiv.org/abs/1611.02796v9

+
+
+

329. The relationship between Precision-Recall and ROC curves
+Jesse Davis, Mark Goadrich
+Proceedings of the 23rd international conference on Machine learning - ICML ’06 (2006) https://doi.org/10.1145/1143844.1143874

+
+
+

330. Do Deep Nets Really Need to be Deep?
+Lei Jimmy Ba, Rich Caruana
+arXiv (2013-12-21) https://arxiv.org/abs/1312.6184v7

+
+
+

331. Deep Neural Networks are Easily Fooled: High Confidence Predictions for Unrecognizable Images
+Anh Nguyen, Jason Yosinski, Jeff Clune
+arXiv (2014-12-05) https://arxiv.org/abs/1412.1897v4

+
+
+

332. “Why Should I Trust You?”: Explaining the Predictions of Any Classifier
+Marco Tulio Ribeiro, Sameer Singh, Carlos Guestrin
+arXiv (2016-02-16) https://arxiv.org/abs/1602.04938v3

+
+
+

333. Visualizing and Understanding Convolutional Networks
+Matthew D Zeiler, Rob Fergus
+arXiv (2013-11-12) https://arxiv.org/abs/1311.2901v3

+
+
+

334. Visualizing Deep Neural Network Decisions: Prediction Difference Analysis
+Luisa M Zintgraf, Taco S Cohen, Tameem Adel, Max Welling
+arXiv (2017-02-15) https://arxiv.org/abs/1702.04595v1

+
+
+

335. Interpretable Explanations of Black Boxes by Meaningful Perturbation
+Ruth Fong, Andrea Vedaldi
+arXiv (2017-04-11) https://arxiv.org/abs/1704.03296v1

+
+
+

336. Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps
+Karen Simonyan, Andrea Vedaldi, Andrew Zisserman
+arXiv (2013-12-20) https://arxiv.org/abs/1312.6034v2

+
+
+

337. On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer-Wise Relevance Propagation
+Sebastian Bach, Alexander Binder, Grégoire Montavon, Frederick Klauschen, Klaus-Robert Müller, Wojciech Samek
+PLOS ONE (2015-07-10) https://doi.org/10.1371/journal.pone.0130140

+
+
+

338. Investigating the influence of noise and distractors on the interpretation of neural networks
+Pieter-Jan Kindermans, Kristof Schütt, Klaus-Robert Müller, Sven Dähne
+arXiv (2016-11-22) https://arxiv.org/abs/1611.07270v1

+
+
+

339. Striving for Simplicity: The All Convolutional Net
+Jost Tobias Springenberg, Alexey Dosovitskiy, Thomas Brox, Martin Riedmiller
+arXiv (2014-12-21) https://arxiv.org/abs/1412.6806v3

+
+
+

340. Salient Deconvolutional Networks
+Aravindh Mahendran, Andrea Vedaldi
+Computer Vision – ECCV 2016 (2016) https://doi.org/10.1007/978-3-319-46466-4_8

+
+
+

341. Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization
+Ramprasaath R. Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, Dhruv Batra
+arXiv (2016-10-07) https://arxiv.org/abs/1610.02391v3

+
+
+

342. Axiomatic Attribution for Deep Networks
+Mukund Sundararajan, Ankur Taly, Qiqi Yan
+arXiv (2017-03-04) https://arxiv.org/abs/1703.01365v2

+
+
+

343. An unexpected unity among methods for interpreting model predictions
+Scott Lundberg, Su-In Lee
+arXiv (2016-11-22) https://arxiv.org/abs/1611.07478v3

+
+
+

344. 17. A Value for n-Person Games
+L. S. Shapley
+Contributions to the Theory of Games (AM-28), Volume II (1953) https://doi.org/10.1515/9781400881970-018

+
+
+

345. Understanding Deep Image Representations by Inverting Them
+Aravindh Mahendran, Andrea Vedaldi
+arXiv (2014-11-26) https://arxiv.org/abs/1412.0035v1

+
+
+

346. Maximum Entropy Methods for Extracting the Learned Features of Deep Neural Networks
+Alex I Finnegan, Jun S Song
+Cold Spring Harbor Laboratory (2017-02-03) https://doi.org/10.1101/105957

+
+
+

347. Visualizing Deep Convolutional Neural Networks Using Natural Pre-images
+Aravindh Mahendran, Andrea Vedaldi
+International Journal of Computer Vision (2016-05-18) https://doi.org/10.1007/s11263-016-0911-8

+
+
+

348. Inceptionism: Going Deeper into Neural Networks
+Alexander Mordvintsev, Christopher Olah, Mike Tyka
+Google Research Blog (2015-06) http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html

+
+
+

349. Visualizing Higher-Layer Features of a Deep Network
+Dumitru Erhan, Yoshua Bengio, Aaron Courville, Pascal Vincent
+University of Montreal (2009-06) http://www.iro.umontreal.ca/~lisa/publications2/index.php/publications/show/247

+
+
+

350. Understanding Neural Networks Through Deep Visualization
+Jason Yosinski, Jeff Clune, Anh Nguyen, Thomas Fuchs, Hod Lipson
+arXiv (2015-06-22) https://arxiv.org/abs/1506.06579v1

+
+
+

351. Neural Machine Translation by Jointly Learning to Align and Translate
+Dzmitry Bahdanau, Kyunghyun Cho, Yoshua Bengio
+arXiv (2014-09-01) https://arxiv.org/abs/1409.0473v7

+
+
+

352. Show, Attend and Tell: Neural Image Caption Generation with Visual Attention
+Kelvin Xu, Jimmy Ba, Ryan Kiros, Kyunghyun Cho, Aaron Courville, Ruslan Salakhutdinov, Richard Zemel, Yoshua Bengio
+arXiv (2015-02-10) https://arxiv.org/abs/1502.03044v3

+
+
+

353. Genetic Architect: Discovering Genomic Structure with Learned Neural Architectures
+Laura Deming, Sasha Targ, Nate Sauder, Diogo Almeida, Chun Jimmie Ye
+arXiv (2016-05-23) https://arxiv.org/abs/1605.07156v1

+
+
+

354. RETAIN: An Interpretable Predictive Model for Healthcare using Reverse Time Attention Mechanism
+Edward Choi, Mohammad Taha Bahadori, Joshua A. Kulas, Andy Schuetz, Walter F. Stewart, Jimeng Sun
+arXiv (2016-08-19) https://arxiv.org/abs/1608.05745v4

+
+
+

355. GRAM: Graph-based Attention Model for Healthcare Representation Learning
+Edward Choi, Mohammad Taha Bahadori, Le Song, Walter F. Stewart, Jimeng Sun
+arXiv (2016-11-21) https://arxiv.org/abs/1611.07012v3

+
+
+

356. Sequence learning with recurrent networks: analysis of internal representations
+Joydeep Ghosh, Vijay Karamcheti
+Science of Artificial Neural Networks (1992-07-01) https://doi.org/10.1117/12.140112

+
+
+

357. Visualizing and Understanding Recurrent Networks
+Andrej Karpathy, Justin Johnson, Li Fei-Fei
+arXiv (2015-06-05) https://arxiv.org/abs/1506.02078v2

+
+
+

358. LSTMVis: A Tool for Visual Analysis of Hidden State Dynamics in Recurrent Neural Networks
+Hendrik Strobelt, Sebastian Gehrmann, Hanspeter Pfister, Alexander M. Rush
+arXiv (2016-06-23) https://arxiv.org/abs/1606.07461v2

+
+
+

359. Automatic Rule Extraction from Long Short Term Memory Networks
+W. James Murdoch, Arthur Szlam
+arXiv (2017-02-08) https://arxiv.org/abs/1702.02540v2

+
+
+

360. Confidence interval prediction for neural network models
+G. Chryssolouris, M. Lee, A. Ramsey
+IEEE Transactions on Neural Networks (1996) https://doi.org/10.1109/72.478409

+
+
+

361. Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning
+Yarin Gal, Zoubin Ghahramani
+arXiv (2015-06-06) https://arxiv.org/abs/1506.02142v6

+
+
+

362. Understanding Black-box Predictions via Influence Functions
+Pang Wei Koh, Percy Liang
+arXiv (2017-03-14) https://arxiv.org/abs/1703.04730v2

+
+
+

363. ActiVis: Visual Exploration of Industry-Scale Deep Neural Network Models
+Minsuk Kahng, Pierre Y. Andrews, Aditya Kalro, Duen Horng Chau
+arXiv (2017-04-06) https://arxiv.org/abs/1704.01942v2

+
+
+

364. Towards Better Analysis of Deep Convolutional Neural Networks
+Mengchen Liu, Jiaxin Shi, Zhen Li, Chongxuan Li, Jun Zhu, Shixia Liu
+arXiv (2016-04-24) https://arxiv.org/abs/1604.07043v3

+
+
+

365. Distilling Knowledge from Deep Networks with Applications to Healthcare Domain
+Zhengping Che, Sanjay Purushotham, Robinder Khemani, Yan Liu
+arXiv (2015-12-11) https://arxiv.org/abs/1512.03542v1

+
+
+

366. Rationalizing Neural Predictions
+Tao Lei, Regina Barzilay, Tommi Jaakkola
+arXiv (2016-06-13) https://arxiv.org/abs/1606.04155v2

+
+
+

367. Functional Knowledge Transfer for High-accuracy Prediction of Under-studied Biological Processes
+Christopher Y. Park, Aaron K. Wong, Casey S. Greene, Jessica Rowland, Yuanfang Guan, Lars A. Bongo, Rebecca D. Burdine, Olga G. Troyanskaya
+PLoS Computational Biology (2013-03-14) https://doi.org/10.1371/journal.pcbi.1002957

+
+
+

368. DeepAD: Alzheimer′s Disease Classification via Deep Convolutional Neural Networks using MRI and fMRI
+Saman Sarraf, Danielle D. DeSouza, John Anderson, Ghassem Tofighi,
+Cold Spring Harbor Laboratory (2016-08-21) https://doi.org/10.1101/070441

+
+
+

369. DeepBound: Accurate Identification of Transcript Boundaries via Deep Convolutional Neural Fields
+Mingfu Shao, Jianzhu Ma, Sheng Wang
+Cold Spring Harbor Laboratory (2017-04-07) https://doi.org/10.1101/125229

+
+
+

370. A general framework for estimating the relative pathogenicity of human genetic variants
+Martin Kircher, Daniela M Witten, Preti Jain, Brian J O’Roak, Gregory M Cooper, Jay Shendure
+Nature Genetics (2014-02-02) https://doi.org/10.1038/ng.2892

+
+
+

371. Predicting Peptide-MHC Binding Affinities With Imputed Training Data
+Alex Rubinsteyn, Timothy O’Donnell, Nandita Damaraju, Jeffrey Hammerbacher
+Cold Spring Harbor Laboratory (2016-05-22) https://doi.org/10.1101/054775

+
+
+

372. Diet Networks: Thin Parameters for Fat Genomics
+Adriana Romero, Pierre Luc Carrier, Akram Erraqabi, Tristan Sylvain, Alex Auvolat, Etienne Dejoie, Marc-André Legault, Marie-Pierre Dubé, Julie G. Hussin, Yoshua Bengio
+International Conference on Learning Representations 2017 (2016-11-04) https://openreview.net/forum?id=Sk-oDY9ge&noteId=Sk-oDY9ge

+
+
+

373. Deep learning in neural networks: An overview
+Jürgen Schmidhuber
+Neural Networks (2015-01) https://doi.org/10.1016/j.neunet.2014.09.003

+
+
+

374. Deep Learning with Limited Numerical Precision
+Suyog Gupta, Ankur Agrawal, Kailash Gopalakrishnan, Pritish Narayanan
+arXiv (2015-02-09) https://arxiv.org/abs/1502.02551v1

+
+
+

375. Training deep neural networks with low precision multiplications
+Matthieu Courbariaux, Yoshua Bengio, Jean-Pierre David
+arXiv (2014-12-22) https://arxiv.org/abs/1412.7024v5

+
+
+

376. Taming the Wild: A Unified Analysis of Hogwild!-Style Algorithms
+Christopher De Sa, Ce Zhang, Kunle Olukotun, Christopher Ré
+arXiv (2015-06-22) https://arxiv.org/abs/1506.06438v2

+
+
+

377. Quantized Neural Networks: Training Neural Networks with Low Precision Weights and Activations
+Itay Hubara, Matthieu Courbariaux, Daniel Soudry, Ran El-Yaniv, Yoshua Bengio
+arXiv (2016-09-22) https://arxiv.org/abs/1609.07061v1

+
+
+

378. Distilling the Knowledge in a Neural Network
+Geoffrey Hinton, Oriol Vinyals, Jeff Dean
+arXiv (2015-03-09) https://arxiv.org/abs/1503.02531v1

+
+
+

379. Large-scale deep unsupervised learning using graphics processors
+Rajat Raina, Anand Madhavan, Andrew Y. Ng
+Proceedings of the 26th Annual International Conference on Machine Learning - ICML ’09 (2009) https://doi.org/10.1145/1553374.1553486

+
+
+

380. Improving the speed of neural networks on CPUs
+Vincent Vanhoucke, Andrew Senior, Mark Z. Mao
+(2011) https://research.google.com/pubs/pub37631.html

+
+
+

381. On parallelizability of stochastic gradient descent for speech DNNS
+Frank Seide, Hao Fu, Jasha Droppo, Gang Li, Dong Yu
+2014 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) (2014-05) https://doi.org/10.1109/icassp.2014.6853593

+
+
+

382. Caffe con Troll: Shallow Ideas to Speed Up Deep Learning
+Stefan Hadjis, Firas Abuzaid, Ce Zhang, Christopher Ré
+arXiv (2015-04-16) https://arxiv.org/abs/1504.04343v2

+
+
+

383. Growing pains for deep learning
+Chris Edwards
+Communications of the ACM (2015-06-25) https://doi.org/10.1145/2771283

+
+
+

384. Experiments on Parallel Training of Deep Neural Network using Model Averaging
+Hang Su, Haoyu Chen
+arXiv (2015-07-05) https://arxiv.org/abs/1507.01239v2

+
+
+

385. Efficient mini-batch training for stochastic optimization
+Mu Li, Tong Zhang, Yuqiang Chen, Alexander J. Smola
+Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining - KDD ’14 (2014) https://doi.org/10.1145/2623330.2623612

+
+
+

386. CGBVS-DNN: Prediction of Compound-protein Interactions Based on Deep Learning
+Masatoshi Hamanaka, Kei Taneishi, Hiroaki Iwata, Jun Ye, Jianguo Pei, Jinlong Hou, Yasushi Okuno
+Molecular Informatics (2016-08-12) https://doi.org/10.1002/minf.201600045

+
+
+

387. cuDNN: Efficient Primitives for Deep Learning
+Sharan Chetlur, Cliff Woolley, Philippe Vandermersch, Jonathan Cohen, John Tran, Bryan Catanzaro, Evan Shelhamer
+arXiv (2014-10-03) https://arxiv.org/abs/1410.0759v3

+
+
+

388. Compressing Neural Networks with the Hashing Trick
+Wenlin Chen, James T. Wilson, Stephen Tyree, Kilian Q. Weinberger, Yixin Chen
+arXiv (2015-04-19) https://arxiv.org/abs/1504.04788v1

+
+
+

389. Deep Learning on FPGAs: Past, Present, and Future
+Griffin Lacey, Graham W. Taylor, Shawki Areibi
+arXiv (2016-02-13) https://arxiv.org/abs/1602.04283v1

+
+
+

390. In-Datacenter Performance Analysis of a Tensor Processing Unit
+Norman P. Jouppi, Cliff Young, Nishant Patil, David Patterson, Gaurav Agrawal, Raminder Bajwa, Sarah Bates, Suresh Bhatia, Nan Boden, Al Borchers, … Doe Hyun Yoon
+arXiv (2017-04-16) https://arxiv.org/abs/1704.04760v1

+
+
+

391. MapReduce
+Jeffrey Dean, Sanjay Ghemawat
+Communications of the ACM (2008-01-01) https://doi.org/10.1145/1327452.1327492

+
+
+

392. Distributed GraphLab
+Yucheng Low, Danny Bickson, Joseph Gonzalez, Carlos Guestrin, Aapo Kyrola, Joseph M. Hellerstein
+Proceedings of the VLDB Endowment (2012-04-01) https://doi.org/10.14778/2212351.2212354

+
+
+

393. Large Scale Distributed Deep Networks
+Jeffrey Dean, Greg S Corrado, Rajat Monga, Kai Chen, Matthieu Devin, Quoc V Le, Mark Z Mao, Marc’Aurelio Ranzato, Andrew Senior, Paul Tucker, … Andrew Y Ng
+Neural Information Processing Systems 2012 (2012-12) http://research.google.com/archive/large_deep_networks_nips2012.html

+
+
+

394. Taming the Wild: A Unified Analysis of Hogwild!-Style Algorithms
+Christopher De Sa, Ce Zhang, Kunle Olukotun, Christopher Ré
+Advances in neural information processing systems (2015-12) https://www.ncbi.nlm.nih.gov/pubmed/27330264

+
+
+

395. SparkNet: Training Deep Networks in Spark
+Philipp Moritz, Robert Nishihara, Ion Stoica, Michael I. Jordan
+arXiv (2015-11-19) https://arxiv.org/abs/1511.06051v4

+
+
+

396. MLlib: Machine Learning in Apache Spark
+Xiangrui Meng, Joseph Bradley, Burak Yavuz, Evan Sparks, Shivaram Venkataraman, Davies Liu, Jeremy Freeman, DB Tsai, Manish Amde, Sean Owen, … Ameet Talwalkar
+arXiv (2015-05-26) https://arxiv.org/abs/1505.06807v1

+
+
+

397. TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems
+Martín Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S. Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, … Xiaoqiang Zheng
+arXiv (2016-03-14) https://arxiv.org/abs/1603.04467v2

+
+
+

398. fchollet/kerasGitHub (2017) https://github.com/fchollet/keras

+
+
+

399. maxpumperla/elephasGitHub (2017) https://github.com/maxpumperla/elephas

+
+
+

400. Deep learning with COTS HPC systems
+Adam Coates, Brody Huval, Tao Wang, David Wu, Bryan Catanzaro, Ng Andrew
+(2013-02-13) http://www.jmlr.org/proceedings/papers/v28/coates13.html

+
+
+

401. Ensemble-Compression: A New Method for Parallel Training of Deep Neural Networks
+Shizhao Sun, Wei Chen, Jiang Bian, Xiaoguang Liu, Tie-Yan Liu
+arXiv (2016-06-02) https://arxiv.org/abs/1606.00575v2

+
+
+

402. Algorithms for Hyper-parameter Optimization
+James Bergstra, Rémi Bardenet, Yoshua Bengio, Balázs Kégl
+Proceedings of the 24th International Conference on Neural Information Processing Systems (2011) http://dl.acm.org/citation.cfm?id=2986459.2986743

+
+
+

403. Random Search for Hyper-Parameter Optimization
+James Bergstra, Yoshua Bengio
+Journal of Machine Learning Research (2012) http://www.jmlr.org/papers/v13/bergstra12a.html

+
+
+

404. Cloud computing and the DNA data race
+Michael C Schatz, Ben Langmead, Steven L Salzberg
+Nature Biotechnology (2010-07) https://doi.org/10.1038/nbt0710-691

+
+
+

405. The real cost of sequencing: scaling computation to keep pace with data generation
+Paul Muir, Shantao Li, Shaoke Lou, Daifeng Wang, Daniel J Spakowicz, Leonidas Salichos, Jing Zhang, George M. Weinstock, Farren Isaacs, Joel Rozowsky, Mark Gerstein
+Genome Biology (2016-03-23) https://doi.org/10.1186/s13059-016-0917-0

+
+
+

406. The case for cloud computing in genome informatics
+Lincoln D Stein
+Genome Biology (2010) https://doi.org/10.1186/gb-2010-11-5-207

+
+
+

407. One weird trick for parallelizing convolutional neural networks
+Alex Krizhevsky
+arXiv (2014-04-23) https://arxiv.org/abs/1404.5997v2

+
+
+

408. A view of cloud computing
+Michael Armbrust, Ion Stoica, Matei Zaharia, Armando Fox, Rean Griffith, Anthony D. Joseph, Randy Katz, Andy Konwinski, Gunho Lee, David Patterson, Ariel Rabkin
+Communications of the ACM (2010-04-01) https://doi.org/10.1145/1721654.1721672

+
+
+

409. Data Sharing
+Dan L. Longo, Jeffrey M. Drazen
+New England Journal of Medicine (2016-01-21) https://doi.org/10.1056/nejme1516564

+
+
+

410. Celebrating parasites
+Casey S Greene, Lana X Garmire, Jack A Gilbert, Marylyn D Ritchie, Lawrence E Hunter
+Nature Genetics (2017-03-30) https://doi.org/10.1038/ng.3830

+
+
+

411. Enhancing reproducibility for computational methods
+V. Stodden, M. McNutt, D. H. Bailey, E. Deelman, Y. Gil, B. Hanson, M. A. Heroux, J. P. A. Ioannidis, M. Taufer
+Science (2016-12-08) https://doi.org/10.1126/science.aah6168

+
+
+

412. DragoNN(2016-11-06) http://kundajelab.github.io/dragonn/

+
+
+

413. ENCODE-DREAM in vivo Transcription Factor Binding Site Prediction Challenge(2017) https://www.synapse.org/#!Synapse:syn6131484/wiki/402026

+
+
+

414. How transferable are features in deep neural networks?
+Jason Yosinski, Jeff Clune, Yoshua Bengio, Hod Lipson
+(2014) https://papers.nips.cc/paper/5347-how-transferable-are-features-in-deep-neural-networks

+
+
+

415. Deep Model Based Transfer and Multi-Task Learning for Biological Image Analysis
+Wenlu Zhang, Rongjian Li, Tao Zeng, Qian Sun, Sudhir Kumar, Jieping Ye, Shuiwang Ji
+Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining - KDD ’15 (2015) https://doi.org/10.1145/2783258.2783304

+
+
+

416. Deep convolutional neural networks for annotating gene expression patterns in the mouse brain
+Tao Zeng, Rongjian Li, Ravi Mukkamala, Jieping Ye, Shuiwang Ji
+BMC Bioinformatics (2015-05-07) https://doi.org/10.1186/s12859-015-0553-9

+
+
+

417. Accurate Classification of Protein Subcellular Localization from High-Throughput Microscopy Images Using Deep Learning
+Tanel Pärnamaa, Leopold Parts
+G3: Genes|Genomes|Genetics (2017-04-08) https://doi.org/10.1534/g3.116.033654

+
+
+

418. Automated analysis of high‐content microscopy data with deep learning
+Oren Z Kraus, Ben T Grys, Jimmy Ba, Yolanda Chong, Brendan J Frey, Charles Boone, Brenda J Andrews
+Molecular Systems Biology (2017-04) https://doi.org/10.15252/msb.20177551

+
+
+

419. Multimodal Deep Learning
+Jiquan Ngiam, Aditya Khosla, Mingyu Kim, Juhan Nam, Honglak Lee, Andrew Y. Ng
+Proceedings of the 28th International Conference on Machine Learning (2011) https://ccrma.stanford.edu/~juhan/pubs/NgiamKhoslaKimNamLeeNg2011.pdf

+
+
+

420. Deep Learning based multi-omics integration robustly predicts survival in liver cancer
+Kumardeep Chaudhary, Olivier B. Poirion, Liangqun Lu, Lana X. Garmire
+Cold Spring Harbor Laboratory (2017-03-08) https://doi.org/10.1101/114892

+
+
+

421. FIDDLE: An integrative deep learning framework for functional genomic data inference
+Umut Eser, L. Stirling Churchman
+Cold Spring Harbor Laboratory (2016-10-17) https://doi.org/10.1101/081380

+
+
+

422. Modeling Reactivity to Biological Macromolecules with a Deep Multitask Network
+Tyler B. Hughes, Na Le Dang, Grover P. Miller, S. Joshua Swamidass
+ACS Central Science (2016-08-24) https://doi.org/10.1021/acscentsci.6b00162

+
+
+

423. IBM edges closer to human speech recognition
+BI Intelligence
+Business Insider (2017-03-14) http://www.businessinsider.com/ibm-edges-closer-to-human-speech-recognition-2017-3

+
+
+

424. Achieving Human Parity in Conversational Speech Recognition
+W. Xiong, J. Droppo, X. Huang, F. Seide, M. Seltzer, A. Stolcke, D. Yu, G. Zweig
+arXiv (2016-10-17) https://arxiv.org/abs/1610.05256v2

+
+
+

425. English Conversational Telephone Speech Recognition by Humans and Machines
+George Saon, Gakuto Kurata, Tom Sercu, Kartik Audhkhasi, Samuel Thomas, Dimitrios Dimitriadis, Xiaodong Cui, Bhuvana Ramabhadran, Michael Picheny, Lynn-Li Lim, … Phil Hall
+arXiv (2017-03-06) https://arxiv.org/abs/1703.02136v1

+
+
+

426. Intriguing properties of neural networks
+Christian Szegedy, Wojciech Zaremba, Ilya Sutskever, Joan Bruna, Dumitru Erhan, Ian Goodfellow, Rob Fergus
+arXiv (2013-12-21) https://arxiv.org/abs/1312.6199v4

+
+
+

427. Explaining and Harnessing Adversarial Examples
+Ian J. Goodfellow, Jonathon Shlens, Christian Szegedy
+arXiv (2014-12-20) https://arxiv.org/abs/1412.6572v3

+
+
+

428. Towards the Science of Security and Privacy in Machine Learning
+Nicolas Papernot, Patrick McDaniel, Arunesh Sinha, Michael Wellman
+arXiv (2016-11-11) https://arxiv.org/abs/1611.03814v1

+
+
+

429. Feature Squeezing: Detecting Adversarial Examples in Deep Neural Networks
+Weilin Xu, David Evans, Yanjun Qi
+arXiv (2017-04-04) https://arxiv.org/abs/1704.01155v1

+
+
+

430. The Grey Literature — Proof of prespecified endpoints in medical research with the bitcoin blockchain
+Benjamin Gregory Carlisle
+(2014-08-25) https://www.bgcarlisle.com/blog/2014/08/25/proof-of-prespecified-endpoints-in-medical-research-with-the-bitcoin-blockchain/

+
+
+

431. The most interesting case of scientific irreproducibility?
+Daniel Himmelstein
+Satoshi Village (2017-03-08) http://blog.dhimmel.com/irreproducible-timestamps/

+
+
+

432. OpenTimestamps: a timestamping proof standard(2017-05-16) https://opentimestamps.org/

+
+
+

433. greenelab/deep-reviewGitHub (2017) https://github.com/greenelab/deep-review

+
+
+ + + diff --git a/all-sections.md b/manuscript.md similarity index 91% rename from all-sections.md rename to manuscript.md index 416a2037..a8e046c3 100644 --- a/all-sections.md +++ b/manuscript.md @@ -1,67 +1,319 @@ -# Opportunities and obstacles for deep learning in biology and medicine - -This article is also available as a [preprint on bioRxiv](https://doi.org/10.1101/142760). - -Travers Ching1,*, -Daniel S. Himmelstein2, -Brett K. Beaulieu-Jones3, -Alexandr A. Kalinin4, -Brian T. Do5, -Gregory P. Way2, -Enrico Ferrero6, -Paul-Michael Agapow7, -Wei Xie8, -Gail L. Rosen9, -Benjamin J. Lengerich10, -Johnny Israeli11, -Jack Lanchantin12, -Stephen Woloszynek9, -Anne E. Carpenter13, -Avanti Shrikumar14, -Jinbo Xu15, -Evan M. Cofer16, -David J. Harris17, -Dave DeCaprio18, -Yanjun Qi12, -Anshul Kundaje19, -Yifan Peng20, -Laura K. Wiley21, -Marwin H.S. Segler22, -Anthony Gitter23,†, -Casey S. Greene2,† - - -* Author order was determined with a randomized algorithm - -† To whom correspondence should be addressed: gitter@biostat.wisc.edu (A.G.) and csgreene@upenn.edu (C.S.G.) - -1. Molecular Biosciences and Bioengineering Graduate Program, University of Hawaii at Manoa, Honolulu, HI -2. Department of Systems Pharmacology and Translational Therapeutics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA -3. Genomics and Computational Biology Graduate Group, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA -4. Department of Computational Medicine and Bioinformatics, University of Michigan Medical School, Ann Arbor, MI -5. Harvard Medical School, Boston, MA -6. Computational Biology and Stats, Target Sciences, GlaxoSmithKline, Stevenage, United Kingdom -7. Data Science Institute, Imperial College London, London, United Kingdom -8. Electrical Engineering and Computer Science, Vanderbilt University, Nashville, TN -9. Ecological and Evolutionary Signal-processing and Informatics Laboratory, Department of Electrical and Computer Engineering, Drexel University, Philadelphia, PA -10. Computational Biology Department, School of Computer Science, Carnegie Mellon University, Pittsburgh, PA -11. Biophysics Program, Stanford University, Stanford, CA -12. Department of Computer Science, University of Virginia, Charlottesville, VA -13. Imaging Platform, Broad Institute of Harvard and MIT, Cambridge, MA -14. Department of Computer Science, Stanford University, Stanford, CA -15. Toyota Technological Institute at Chicago, Chicago, IL -16. Department of Computer Science, Trinity University, San Antonio, TX -17. Department of Wildlife Ecology and Conservation, University of Florida, Gainesville, FL -18. ClosedLoop.ai, Austin, TX -19. Department of Genetics and Department of Computer Science, Stanford University, Stanford, CA -20. National Center for Biotechnology Information and National Library of Medicine, National Institutes of Health, Bethesda, MD -21. Division of Biomedical Informatics and Personalized Medicine, University of Colorado School of Medicine, Aurora, CO -22. Institute of Organic Chemistry, Westfälische Wilhelms-Universität Münster, Münster, Germany -23. Department of Biostatistics and Medical Informatics, University of Wisconsin-Madison and Morgridge Institute for Research, Madison, WI - - - -## Abstract +--- +author-meta: +- Travers Ching +- Daniel S. Himmelstein +- Brett K. Beaulieu-Jones +- Alexandr A. Kalinin +- Brian T. Do +- Gregory P. Way +- Enrico Ferrero +- Paul-Michael Agapow +- Wei Xie +- Gail L. Rosen +- Benjamin J. Lengerich +- Johnny Israeli +- Jack Lanchantin +- Stephen Woloszynek +- Anne E. Carpenter +- Avanti Shrikumar +- Jinbo Xu +- Evan M. Cofer +- David J. Harris +- Dave DeCaprio +- Yanjun Qi +- Anshul Kundaje +- Yifan Peng +- Laura K. Wiley +- Marwin H.S. Segler +- Anthony Gitter +- Casey S. Greene +date-meta: '2017-11-03' +keywords: +- deep learning +- review +- precision medicine +- genomics +- machine learning +- neural networks +- collaborative +- manubot +lang: en-US +title: Opportunities and obstacles for deep learning in biology and medicine +... + + + +_A DOI-citable preprint of this manuscript is available at _. + + + +This manuscript was automatically generated +from [greenelab/deep-review@8eb858a](https://github.com/greenelab/deep-review/tree/8eb858a277c7e31b6d0db5cfb10ebf7ebab59fe1) +on November 3, 2017. + + +## Authors + + + ++ **Travers Ching**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0002-5577-3516](https://orcid.org/0000-0002-5577-3516) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [traversc](https://github.com/traversc)
+ + Molecular Biosciences and Bioengineering Graduate Program, University of Hawaii at Manoa, Honolulu, HI + + ++ **Daniel S. Himmelstein**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0002-3012-7446](https://orcid.org/0000-0002-3012-7446) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [dhimmel](https://github.com/dhimmel)
+ + Department of Systems Pharmacology and Translational Therapeutics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA + · Funded by GBMF GBMF4552 + + ++ **Brett K. Beaulieu-Jones**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0002-6700-1468](https://orcid.org/0000-0002-6700-1468) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [brettbj](https://github.com/brettbj)
+ + Genomics and Computational Biology Graduate Group, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA + · Funded by NIH R01AI116794 + + ++ **Alexandr A. Kalinin**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0003-4563-3226](https://orcid.org/0000-0003-4563-3226) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [alxndrkalinin](https://github.com/alxndrkalinin)
+ + Department of Computational Medicine and Bioinformatics, University of Michigan Medical School, Ann Arbor, MI + + ++ **Brian T. Do**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0003-4992-2623](https://orcid.org/0000-0003-4992-2623) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [bdo311](https://github.com/bdo311)
+ + Harvard Medical School, Boston, MA + · Funded by NIH T32GM007753 + + ++ **Gregory P. Way**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0002-0503-9348](https://orcid.org/0000-0002-0503-9348) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [gwaygenomics](https://github.com/gwaygenomics)
+ + Department of Systems Pharmacology and Translational Therapeutics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA + + ++ **Enrico Ferrero**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0002-8362-100X](https://orcid.org/0000-0002-8362-100X) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [enricoferrero](https://github.com/enricoferrero)
+ + Computational Biology and Stats, Target Sciences, GlaxoSmithKline, Stevenage, United Kingdom + + ++ **Paul-Michael Agapow**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0003-1126-1479](https://orcid.org/0000-0003-1126-1479) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [agapow](https://github.com/agapow)
+ + Data Science Institute, Imperial College London, London, United Kingdom + + ++ **Wei Xie**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0002-1871-6846](https://orcid.org/0000-0002-1871-6846) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [xieconnect](https://github.com/xieconnect)
+ + Electrical Engineering and Computer Science, Vanderbilt University, Nashville, TN + + ++ **Gail L. Rosen**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0003-1763-5750](https://orcid.org/0000-0003-1763-5750) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [gailrosen](https://github.com/gailrosen)
+ + Ecological and Evolutionary Signal-processing and Informatics Laboratory, Department of Electrical and Computer Engineering, Drexel University, Philadelphia, PA + · Funded by NSF 1245632 + + ++ **Benjamin J. Lengerich**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0001-8690-9554](https://orcid.org/0000-0001-8690-9554) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [blengerich](https://github.com/blengerich)
+ + Computational Biology Department, School of Computer Science, Carnegie Mellon University, Pittsburgh, PA + + ++ **Johnny Israeli**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0003-1633-5780](https://orcid.org/0000-0003-1633-5780) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [jisraeli](https://github.com/jisraeli)
+ + Biophysics Program, Stanford University, Stanford, CA + + ++ **Jack Lanchantin**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0003-0811-0944](https://orcid.org/0000-0003-0811-0944) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [jacklanchantin](https://github.com/jacklanchantin)
+ + Department of Computer Science, University of Virginia, Charlottesville, VA + + ++ **Stephen Woloszynek**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0003-0568-298X](https://orcid.org/0000-0003-0568-298X) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [sw1](https://github.com/sw1)
+ + Ecological and Evolutionary Signal-processing and Informatics Laboratory, Department of Electrical and Computer Engineering, Drexel University, Philadelphia, PA + + ++ **Anne E. Carpenter**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0003-1555-8261](https://orcid.org/0000-0003-1555-8261) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [annecarpenter](https://github.com/annecarpenter)
+ + Imaging Platform, Broad Institute of Harvard and MIT, Cambridge, MA + · Funded by NIH R01GM089652 + + ++ **Avanti Shrikumar**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0002-6443-4671](https://orcid.org/0000-0002-6443-4671) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [AvantiShri](https://github.com/AvantiShri)
+ + Department of Computer Science, Stanford University, Stanford, CA + + ++ **Jinbo Xu**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0001-7111-4839](https://orcid.org/0000-0001-7111-4839) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [j3xugit](https://github.com/j3xugit)
+ + Toyota Technological Institute at Chicago, Chicago, IL + · Funded by NIH R01GM089753, NSF 1564955 + + ++ **Evan M. Cofer**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0003-3877-0433](https://orcid.org/0000-0003-3877-0433) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [evancofer](https://github.com/evancofer)
+ + Department of Computer Science, Trinity University, San Antonio, TX + · Funded by NSF 1531594 + + ++ **David J. Harris**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0003-3332-9307](https://orcid.org/0000-0003-3332-9307) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [davharris](https://github.com/davharris)
+ + Department of Wildlife Ecology and Conservation, University of Florida, Gainesville, FL + · Funded by GBMF GBMF4563 + + ++ **Dave DeCaprio**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0001-8931-9461](https://orcid.org/0000-0001-8931-9461) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [DaveDeCaprio](https://github.com/DaveDeCaprio)
+ + ClosedLoop.ai, Austin, TX + + ++ **Yanjun Qi**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0002-5796-7453](https://orcid.org/0000-0002-5796-7453) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [qiyanjun](https://github.com/qiyanjun)
+ + Department of Computer Science, University of Virginia, Charlottesville, VA + + ++ **Anshul Kundaje**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0003-3084-2287](https://orcid.org/0000-0003-3084-2287) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [akundaje](https://github.com/akundaje)
+ + Department of Genetics and Department of Computer Science, Stanford University, Stanford, CA + · Funded by NIH DP2GM123485 + + ++ **Yifan Peng**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0001-9309-8331](https://orcid.org/0000-0001-9309-8331) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [yfpeng](https://github.com/yfpeng)
+ + National Center for Biotechnology Information and National Library of Medicine, National Institutes of Health, Bethesda, MD + · Funded by OTHER National Institutes of Health Intramural Research Program and National Library of Medicine + + ++ **Laura K. Wiley**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0001-6681-9754](https://orcid.org/0000-0001-6681-9754) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [laurakwiley](https://github.com/laurakwiley)
+ + Division of Biomedical Informatics and Personalized Medicine, University of Colorado School of Medicine, Aurora, CO + + ++ **Marwin H.S. Segler**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0001-8008-0546](https://orcid.org/0000-0001-8008-0546) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [mrwns](https://github.com/mrwns)
+ + Institute of Organic Chemistry, Westfälische Wilhelms-Universität Münster, Münster, Germany + + ++ **Anthony Gitter**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0002-5324-9833](https://orcid.org/0000-0002-5324-9833) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [agitter](https://github.com/agitter)
+ + Department of Biostatistics and Medical Informatics, University of Wisconsin-Madison and Morgridge Institute for Research, Madison, WI + · Funded by NIH U54AI117924 + + ++ **Casey S. Greene**
+ ![ORCID icon](images/orcid.svg){height="13px" width="13px"} + [0000-0001-8713-9213](https://orcid.org/0000-0001-8713-9213) + · ![GitHub icon](images/github.svg){height="13px" width="13px"} + [cgreene](https://github.com/cgreene)
+ + Department of Systems Pharmacology and Translational Therapeutics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA + · Funded by GBMF GBMF4552 + + + + +## Abstract {.page_break_before} Deep learning, which describes a class of machine learning algorithms, has recently showed impressive results across a variety of domains. Biology and @@ -173,7 +425,12 @@ Here, we seek to identify whether deep learning is an innovation that can induce a Strategic Inflection Point in the practice of biology or medicine. There are already a number of reviews focused on applications of deep learning -in biology [@yXqhuueV; @1VZjheOA; @irSe12Sm; @G00xvi94; @MmRGFVUu], healthcare [@11I7bLcP3; @LL5huVs3], and drug discovery [@gJE0ExFr; @zCt6PUXj; @1DTUK3YyI; @xPkT1z7D]. Under our guiding question, we sought to highlight +in biology [@yXqhuueV; @1VZjheOA; +@irSe12Sm; @G00xvi94; +@MmRGFVUu], healthcare [@11I7bLcP3; +@LL5huVs3], and drug discovery [@gJE0ExFr; +@zCt6PUXj; @1DTUK3YyI; +@xPkT1z7D]. Under our guiding question, we sought to highlight cases where deep learning enabled researchers to solve challenges that were previously considered infeasible or simplified tedious analyses. We also identified approaches that researchers are using to address challenges @@ -223,9 +480,11 @@ deep learning, has been extensively applied is molecular target prediction. For example, deep recurrent neural networks (RNNs) have been used to predict gene targets of microRNAs [@YUms527e], and CNNs have been applied to predict protein residue-residue contacts and secondary structure -[@BhfjKSY3; @ZzaRyGuJ; @UO8L6nd]. Other recent exciting applications of deep learning +[@BhfjKSY3; @ZzaRyGuJ; +@UO8L6nd]. Other recent exciting applications of deep learning include recognition of functional genomic elements such as enhancers and -promoters [@s5sy4AOi; @17B2QAA1k; @12aqvAgz6] and prediction of the deleterious effects of +promoters [@s5sy4AOi; @17B2QAA1k; +@12aqvAgz6] and prediction of the deleterious effects of nucleotide polymorphisms [@15E5yG1Ho]. #### Treatment of patients @@ -317,21 +576,25 @@ learning models, such as ImageNet [@cBVeXnZx], for new purposes. Diagnosing diabetic retinopathy through color fundus images became an area of focus for deep learning researchers after a large labeled image set was made publicly available during a 2015 Kaggle competition [@ayTsooEM]. -Most participants trained neural networks from scratch [@ayTsooEM; @e3vyHBV2; @14Ovc5nPg], but Gulshan et al. +Most participants trained neural networks from scratch [@ayTsooEM; +@e3vyHBV2; @14Ovc5nPg], but Gulshan et al. [@1mJW6umJ] repurposed a 48-layer Inception-v3 deep architecture pre-trained on natural images and surpassed the state-of-the-art specificity and sensitivity. Such features were also repurposed to detect melanoma, the deadliest form of skin cancer, from dermoscopic [@sLPsrfbl; @phRCihNB] and -non-dermoscopic images of skin lesions [@1AJUcl1KV; @O39LDkX; @XnYNYoYB] as well as +non-dermoscopic images of skin lesions [@1AJUcl1KV; +@O39LDkX; @XnYNYoYB] as well as age-related macular degeneration [@iBPOt78R]. Pre-training on natural images can enable very deep networks to succeed without overfitting. For the melanoma task, reported performance was competitive with or better than a board -of certified dermatologists [@sLPsrfbl; @XnYNYoYB]. +of certified dermatologists [@sLPsrfbl; +@XnYNYoYB]. Reusing features from natural images is also an emerging approach for radiographic images, where datasets are often too small to train large deep neural networks without -these techniques [@1Fy5bcnCI; @1GAyqYBNZ; @x6HXFAS4; @Qve94Jra]. +these techniques [@1Fy5bcnCI; @1GAyqYBNZ; +@x6HXFAS4; @Qve94Jra]. A deep CNN trained on natural images boosts performance in radiographic images [@x6HXFAS4]. However, the target task required either re-training the initial model from scratch with special pre-processing or @@ -350,7 +613,9 @@ literature, they found a smaller network trained with data augmentation on few hundred images from a few dozen patients can outperform a pre-trained out-of-domain classifier. Data augmentation is a different strategy to deal with small training sets. The practice is exemplified by a series of papers that -analyze images from mammographies [@VFw1VXDP; @JK8NuXy3; @9G9Hv1Pp; @Xxb4t3zO; @5kfDbGhA]. To expand the number and diversity of images, +analyze images from mammographies [@VFw1VXDP; +@JK8NuXy3; @9G9Hv1Pp; @Xxb4t3zO; +@5kfDbGhA]. To expand the number and diversity of images, researchers constructed adversarial examples [@Xxb4t3zO]. Adversarial examples are constructed by applying a transformation that changes training images but not their content -- for example by rotating an image by a @@ -648,7 +913,8 @@ Methods to accomplish more with little high-quality labeled data are also being applied in other domains and may also be adapted to this challenge, e.g. data programming [@5Il3kN32]. In data programming, noisy automated labeling functions are integrated. Numerous writers have described data as the new oil -[@6fE0Vrba; @o8mib4CN]. +[@6fE0Vrba; +@o8mib4CN]. The idea behind this metaphor is that data are available in large quantities, valuable once refined, and the underlying resource that will enable a data-driven revolution in how work is done. Contrasting with this perspective, @@ -831,7 +1097,8 @@ disequilibrium (even when population stratification is explicitly controlled for [@T3GG8iJN]). As a result, many genomic findings are of limited value for people of non-European ancestry [@dKwyEWWF] and may even lead to worse treatment outcomes for them. Methods have been developed for -mitigating some of these problems in genomic studies [@10shRODux; @T3GG8iJN], but it is not clear how easily they can be adapted for +mitigating some of these problems in genomic studies [@10shRODux; +@T3GG8iJN], but it is not clear how easily they can be adapted for deep models that are designed specifically to extract subtle effects from high-dimensional data. For example, differences in the equipment that tended to be used for cases versus controls have led to spurious genetic findings (e.g. @@ -938,7 +1205,8 @@ expression microarrays into known modules representing cell cycle processes [@AnenJOuU] and to stratify yeast strains based on chemical and mutational perturbations [@yVBx9Qx4]. Shallow (one hidden layer) denoising autoencoders have also been fruitful in extracting biological insight -from thousands of _Pseudomonas aeruginosa_ experiments [@1CFhfCyWN; @zuLdSQx3] and in aggregating features relevant to specific breast +from thousands of _Pseudomonas aeruginosa_ experiments [@1CFhfCyWN; +@zuLdSQx3] and in aggregating features relevant to specific breast cancer subtypes [@PBiRSdXv]. These unsupervised approaches applied to gene expression data are powerful methods for identifying gene signatures that may otherwise be overlooked. An additional benefit of unsupervised approaches is @@ -1189,7 +1457,8 @@ But one could easily replace them with the output of one of the enhancer or prom Prediction of microRNAs (miRNAs) and miRNA targets is of great interest, as they are critical components of gene regulatory networks and -are often conserved across great evolutionary distance [@yVKIhIAf; @8lpCCppx]. While many machine learning algorithms have been +are often conserved across great evolutionary distance [@yVKIhIAf; +@8lpCCppx]. While many machine learning algorithms have been applied to these tasks, they currently require extensive feature selection and optimization. For instance, one of the most widely adopted tools for miRNA target prediction, TargetScan, trained multiple linear @@ -1204,7 +1473,8 @@ As in other applications, deep learning promises to achieve equal or better performance in predictive tasks by automatically engineering complex features to minimize an objective function. Two recently published tools use different recurrent neural network-based architectures to perform miRNA and target -prediction with solely sequence data as input [@1TeyWffV; @1GwC1ll6h]. Though the results are preliminary and still based on +prediction with solely sequence data as input [@1TeyWffV; +@1GwC1ll6h]. Though the results are preliminary and still based on a validation set rather than a completely independent test set, they were able to predict microRNA target sites with higher specificity and sensitivity than TargetScan. Excitingly, these tools seem to show that RNNs can accurately align @@ -1240,12 +1510,14 @@ basic problem and an almost essential module of any protein structure prediction package. Contact prediction is much more challenging than secondary structure prediction, but it has a much larger impact on tertiary structure prediction. In recent years, the accuracy of contact prediction has greatly improved -[@BhfjKSY3; @7atXz0r; @kboAopkh; @10dNuD89l]. +[@BhfjKSY3; @7atXz0r; +@kboAopkh; @10dNuD89l]. One can represent protein secondary structure with three different states (alpha helix, beta strand, and loop regions) or eight finer-grained states. Accuracy of a three-state prediction is called Q3, and accuracy of an 8-state prediction is called Q8. Several groups -[@1AlhRKQbe; @ZzaRyGuJ; @UpFrhdJf] applied deep learning to protein +[@1AlhRKQbe; @ZzaRyGuJ; +@UpFrhdJf] applied deep learning to protein secondary structure prediction but were unable to achieve significant improvement over the *de facto* standard method PSIPRED [@Aic7UyXM], which uses two shallow feedforward neural @@ -1319,7 +1591,8 @@ Once so-called regions of interest have been identified, it is often straightforward to measure biological properties of interest, such as fluorescence intensities, textures, and sizes. Given the dramatic successes of deep learning in biological imaging, we simply refer to articles that review -recent advancements [@MmRGFVUu; @40EG4ZEU; @TutLhFSz]. For deep +recent advancements [@MmRGFVUu; +@40EG4ZEU; @TutLhFSz]. For deep learning to become commonplace for biological image segmentation, we need user-friendly tools. @@ -1353,7 +1626,8 @@ cell population morphologies using conventional methods of segmentation and feature extraction has already proven useful for functionally annotating genes and alleles, identifying the cellular target of small molecules, and identifying disease-specific phenotypes suitable for drug screening -[@hkKO4QYl; @m3Ij21U8; @McjXFLLq]. Deep learning would bring to these new kinds of +[@hkKO4QYl; @m3Ij21U8; +@McjXFLLq]. Deep learning would bring to these new kinds of experiments -- known as image-based profiling or morphological profiling -- a higher degree of accuracy, stemming from the freedom from human-tuned feature extraction strategies. @@ -1375,7 +1649,8 @@ processes over time is not the limiting factor, single-cell techniques can provide maximal resolution compared to averaging across all cells in bulk tissue, enabling the study of transcriptional bursting with single-cell fluorescence *in situ* hybridization or the heterogeneity of epigenetic patterns -with single-cell Hi-C or ATAC-seq [@QafUwNKn; @v97iPXDw]. Joint profiling of single-cell epigenetic and +with single-cell Hi-C or ATAC-seq [@QafUwNKn; +@v97iPXDw]. Joint profiling of single-cell epigenetic and transcriptional states provides unprecedented views of regulatory processes [@1CAw3FaPI]. @@ -1437,12 +1712,15 @@ more than 99% of the genomic content. Subsequent tools aimed to classify tetranucleotide frequencies, which differ across organisms [@N9NzkOjA], using supervised [@QV551Nlx; @1HtJuEkb2] or unsupervised methods [@1HhqhBwrM]. Then, researchers began to use techniques that could estimate relative abundances from -an entire sample faster than classifying individual reads [@56wEWVIl; @RqhGD9c7; @189TQrQA9; @8DLzxOEt]. There is also great interest in +an entire sample faster than classifying individual reads [@56wEWVIl; +@RqhGD9c7; @189TQrQA9; @8DLzxOEt]. There is also great interest in identifying and annotating sequence reads [@qUGH5CX8; @yFOAeemA]. However, the focus on taxonomic and functional annotation is just the first step. Several groups have proposed methods to determine host or environment phenotypes from -the organisms that are identified [@W0cYSf89; @aI9g2UOc; @c5P9jHCg; @y9s5irW] or overall sequence composition [@5W4KMSdT]. Also, researchers have -looked into how feature selection can improve classification [@Kt9NojjR; @y9s5irW], and techniques have been proposed that are classifier-independent +the organisms that are identified [@W0cYSf89; @aI9g2UOc; @c5P9jHCg; +@y9s5irW] or overall sequence composition [@5W4KMSdT]. Also, researchers have +looked into how feature selection can improve classification [@Kt9NojjR; +@y9s5irW], and techniques have been proposed that are classifier-independent [@1AN5UPfb1; @O9D66oYa]. Most neural networks are used for @@ -1603,7 +1881,8 @@ to traditional screening methods in 21 of 27 studies. While further progress has been made in using deep learning for clinical decision making, it is hindered by a challenge common to many deep learning applications: it is much easier to predict an outcome than to suggest an action -to change the outcome. Several attempts [@1FE0F2pQ; @qXdO2aMm] at recasting the clinical decision-making problem into +to change the outcome. Several attempts [@1FE0F2pQ; +@qXdO2aMm] at recasting the clinical decision-making problem into a prediction problem (i.e. prediction of which treatment will most improve the patient's health) have accurately predicted survival patterns, but technical and medical challenges remain for clinical adoption (similar to those for @@ -1636,7 +1915,8 @@ Discussion). #### Predicting patient trajectories A common application for deep learning in this domain is the temporal structure -of healthcare records. Many studies [@4zpZxjHR; @O7Vbecm2; @fOaBA9Vc; @glyI7H6F] have used RNNs to categorize +of healthcare records. Many studies [@4zpZxjHR; @O7Vbecm2; +@fOaBA9Vc; @glyI7H6F] have used RNNs to categorize patients, but most stop short of suggesting clinical decisions. Nemati et al. [@16OQvsRqJ] used deep reinforcement learning to optimize a heparin dosing policy for intensive care patients. However, because the ideal dosing @@ -1673,16 +1953,19 @@ efficiency of clinical trials and accelerate drug development. Drug repositioning (or repurposing) is an attractive option for delivering new drugs to the market because of the high costs and failure rates associated with -more traditional drug discovery approaches [@13c9OPizf; @79Ktl2]. A decade ago, the Connectivity Map +more traditional drug discovery approaches [@13c9OPizf; +@79Ktl2]. A decade ago, the Connectivity Map [@Ot5bUkmI] had a sizeable impact. Reverse matching disease gene expression signatures with a large set of reference compound profiles allowed researchers to formulate repurposing hypotheses at scale using a simple non-parametric test. Since then, several advanced computational methods have been applied to formulate and validate drug -repositioning hypotheses [@gTwjIQqB; @1BkEtNVsj; @ir7ElHha]. Using supervised learning and collaborative filtering +repositioning hypotheses [@gTwjIQqB; @1BkEtNVsj; +@ir7ElHha]. Using supervised learning and collaborative filtering to tackle this type of problem is proving successful, especially when coupling disease or compound omic data with topological information from protein-protein -or protein-compound interaction networks [@M1EW8Rfl; @16FEYidu2; @18lqFDKRR]. +or protein-compound interaction networks [@M1EW8Rfl; +@16FEYidu2; @18lqFDKRR]. For example, Menden et al. [@QcwZC8wG] used a shallow neural network to predict sensitivity of cancer cell lines to drug treatment @@ -1698,7 +1981,8 @@ such as otenzepad and pinacidil for neurological disorders. Drug repositioning can also be approached by attempting to predict novel drug-target interactions and then repurposing the drug for the associated -indication [@tOpadZQw; @1SIuofeg]. Wang et al. [@TeIxEjqm] +indication [@tOpadZQw; +@1SIuofeg]. Wang et al. [@TeIxEjqm] devised a pairwise input neural network with two hidden layers that takes two inputs, a drug and a target binding site, and predicts whether they interact. Wang et al. [@1AU7wzPqa] trained individual RBMs for @@ -1768,7 +2052,8 @@ over a random forest baseline, remarking "we have seldom seen any method in the past 10 years that could consistently outperform [random forest] by such a margin" [@xOaTIeBY]. Subsequent work (reviewed in more detail by Goh et al. [@zCt6PUXj]) explored the effects of jointly modeling -far more targets than the Merck challenge [@F8fP2vAg; @yAoN5gTU], with Ramsundar et al. +far more targets than the Merck challenge [@F8fP2vAg; +@yAoN5gTU], with Ramsundar et al. [@yAoN5gTU] showing that the benefits of multi-task networks had not yet saturated even with 259 targets. Although DeepTox [@Y1D0SZrO], a deep learning approach, won another competition, the @@ -1886,7 +2171,8 @@ researchers without biochemistry expertise. One open question in ligand-based screening pertains to the benefits and limitations of transfer learning. Multi-task neural networks have shown the -advantages of jointly modeling many targets [@F8fP2vAg; @yAoN5gTU]. Other studies have shown the limitations of +advantages of jointly modeling many targets [@F8fP2vAg; +@yAoN5gTU]. Other studies have shown the limitations of transfer learning when the prediction tasks are insufficiently related [@uP7SgBVd; @P4ixsM8i]. This has important implications for representation learning. The typical approach to improve deep @@ -1923,7 +2209,8 @@ will be misleading during training, and the predictive performance is sensitive to the docking quality [@Gue0c5Gb]. There are two established options for representing a protein-compound complex. One option, a 3D -grid, can featurize the input complex [@Z7fd0BYf; @bNBiIiTt]. Each entry in the grid tracks the types of protein and +grid, can featurize the input complex [@Z7fd0BYf; +@bNBiIiTt]. Each entry in the grid tracks the types of protein and ligand atoms in that region of the 3D space or descriptors derived from those atoms. Alternatively, DeepVS [@Gue0c5Gb] and atomic convolutions [@17YaKNLKk] offer greater flexibility in their convolutions by eschewing @@ -2145,10 +2432,12 @@ to the input [@1YcKYTvO] to compute a "saliency map". Bach et al. which was shown to be equivalent to the element-wise product of the gradient and input [@zhmq9ktJ; @b1sc0cgP]. Networks with Rectified Linear Units (ReLUs) create nonlinearities that must be -addressed. Several variants exist for handling this [@voh0OiT2; @f2L6isRj]. Backpropagation-based methods are a highly +addressed. Several variants exist for handling this [@voh0OiT2; +@f2L6isRj]. Backpropagation-based methods are a highly active area of research. Researchers are still actively identifying weaknesses [@vjXoJqO3], and new methods are being developed to address -them [@RZsNSRDS; @WzFOJBiA; @zhmq9ktJ]. Lundberg and Lee [@DeOI1oGf] noted that +them [@RZsNSRDS;; @WzFOJBiA; +@zhmq9ktJ]. Lundberg and Lee [@DeOI1oGf] noted that several importance scoring methods including integrated gradients and LIME could all be considered approximations to Shapely values [@YBJdA6LJ], which have a long history in game theory for assigning contributions to players in @@ -2183,7 +2472,8 @@ Activation maximization can reveal patterns detected by an individual neuron in the network by generating images which maximally activate that neuron, subject to some regularizing constraints. This technique was first introduced in Ehran et al. [@UAAd9Uez] and applied in subsequent work -[@1YcKYTvO; @XLHInhc1; @17i18PMkR; @1FkT6C6oa]. Lanchantin et +[@1YcKYTvO; @XLHInhc1; +@17i18PMkR; @1FkT6C6oa]. Lanchantin et al. [@Dwi2eAvT] applied activation maximization to genomic sequence data. One drawback of this approach is that neural networks often learn highly distributed representations where several neurons cooperatively describe @@ -2196,7 +2486,8 @@ Several interpretation methods are specifically tailored to recurrent neural network architectures. The most common form of interpretability provided by RNNs is through attention mechanisms, which have been used in diverse problems such as image captioning and machine translation to select portions of the input to -focus on generating a particular output [@haHzVaaz; @yHn4SDRI]. Deming et al. [@SAvEOARL] applied the attention +focus on generating a particular output [@haHzVaaz; +@yHn4SDRI]. Deming et al. [@SAvEOARL] applied the attention mechanism to models trained on genomic sequence. Attention mechanisms provide insight into the model's decision-making process by revealing which portions of the input are used by different outputs. In the clinical domain, Choi et al. @@ -2333,7 +2624,8 @@ synthetic TF binding sites with position weight matrices transcript boundaries [@2M3zXijc], is a standard practice in bioinformatics. This strategy can help benchmark algorithms when the available gold standard dataset is imperfect, but it should be paired with an evaluation -on real data, as in the prior examples [@iEmvzeT8; @2M3zXijc]. In rare cases, models trained on simulated data have been +on real data, as in the prior examples [@iEmvzeT8; +@2M3zXijc]. In rare cases, models trained on simulated data have been successfully applied directly to real data [@2M3zXijc]. Data can be simulated to create negative examples when only positive training @@ -2387,14 +2679,18 @@ why neural networks have only recently found widespread use [@BQS8ClV0]. Many have sought to curb these costs, with methods ranging from the very applied -(e.g. reduced numerical precision [@CKcJuj03; @1G3owNNps; @w6CoVmFK; @1GUizyE8e]) to the exotic and theoretic (e.g. +(e.g. reduced numerical precision [@CKcJuj03; @1G3owNNps; +@w6CoVmFK; @1GUizyE8e]) to the exotic and theoretic (e.g. training small networks to mimic large networks and ensembles [@1AhGoHZP9; @1CRF3gAV]). The largest gains in efficiency have come from computation with graphics processing units (GPUs) -[@F3e4wfzQ; @NSgduYNT; @IULiPa6L; @13KjSCKB2; @1FocAi7N0; @BQS8ClV0], which excel at the matrix and vector +[@F3e4wfzQ; @NSgduYNT; @IULiPa6L; +@13KjSCKB2; @1FocAi7N0; +@BQS8ClV0], which excel at the matrix and vector operations so central to deep learning. The massively parallel nature of GPUs allows additional optimizations, such as accelerated mini-batch gradient descent -[@NSgduYNT; @IULiPa6L; @aClNvbyM; @fNkl8HFz]. However, GPUs also have limited memory, making networks +[@NSgduYNT; @IULiPa6L; @aClNvbyM; +@fNkl8HFz]. However, GPUs also have limited memory, making networks of useful size and complexity difficult to implement on a single GPU or machine [@F3e4wfzQ; @CCS5KSIM]. This restriction has sometimes forced computational biologists to use workarounds or limit the size @@ -2410,7 +2706,9 @@ performance [@x0M6vals]. While steady improvements in GPU hardware may alleviate this issue, it is unclear whether advances will occur quickly enough to keep pace with the growing biological datasets and increasingly complex neural networks. Much has been done -to minimize the memory requirements of neural networks [@YwdqeYZi; @1AhGoHZP9; @CKcJuj03; @1G3owNNps; @w6CoVmFK; @15lYGmZpY; @1GUizyE8e], but there is +to minimize the memory requirements of neural networks [@YwdqeYZi; +@1AhGoHZP9; @CKcJuj03; @1G3owNNps; +@w6CoVmFK; @15lYGmZpY; @1GUizyE8e], but there is also growing interest in specialized hardware, such as field-programmable gate arrays (FPGAs) [@1FocAi7N0; @9NKsJjSw] and application-specific integrated circuits (ASICs) [@ULagTifF]. Less @@ -2425,7 +2723,9 @@ Distributed computing is a general solution to intense computational requirements and has enabled many large-scale deep learning efforts. Some types of distributed computation [@xE3EYmck; @1XcexUAV] are not suitable for deep learning [@17cBimWgp], but much progress has been made. -There now exist a number of algorithms [@17cBimWgp; @188FA7whS; @w6CoVmFK], tools [@rmJZ2Aui; @rZnxDitd; @hOeUlCvS], and high-level libraries [@FwEK0msb; @y9IoEy4r] for deep +There now exist a number of algorithms [@17cBimWgp; @188FA7whS; +@w6CoVmFK], tools [@rmJZ2Aui; @rZnxDitd; +@hOeUlCvS], and high-level libraries [@FwEK0msb; @y9IoEy4r] for deep learning in a distributed environment, and it is possible to train very complex networks with limited infrastructure [@4MZ2tmZ8]. Besides handling very large networks, distributed or parallelized approaches offer other @@ -2536,7 +2836,8 @@ the distance between the base task and target task increases In image analysis, previous examples of deep transfer learning applications proved large-scale natural image sets [@cBVeXnZx] to be useful for pre-training models that serve as generic feature extractors for -various types of biological images [@HlDY7trA; @z3I2IudI; @irSe12Sm; @BMg062hc]. More recently, deep +various types of biological images [@HlDY7trA; @z3I2IudI; +@irSe12Sm; @BMg062hc]. More recently, deep learning models predicted protein sub-cellular localization for proteins not originally present in a training set [@2a7MHtAx]. Moreover, learned features performed reasonably well even when applied to images obtained using @@ -2563,7 +2864,8 @@ al. [@jV2YerUS] demonstrated how training on the experimentally-validated FANTOM enhancer datasets improved cell type-specific predictions, outperforming state-of-the-art results. In drug design, general RNN models trained to generate molecules from the ChEMBL database have been fine-tuned to produce -drug-like compounds for specific targets [@8LWFFeYg; @1EayJRsI]. +drug-like compounds for specific targets [@8LWFFeYg; +@1EayJRsI]. Related to transfer learning, multimodal learning assumes simultaneous learning from various types of inputs, such as images and text. It can capture features @@ -2629,7 +2931,9 @@ site from cancer pathology reports together with its laterality substantially improved the performance for the latter task, indicating that multi-task learning can effectively leverage the commonality between two tasks using a shared representation. Many studies employed multi-task learning to predict -chemical bioactivity [@1Dzz0P0qr; @yAoN5gTU] and drug toxicity [@Y1D0SZrO; @1BARarxfz]. Kearnes et al. [@uP7SgBVd] +chemical bioactivity [@1Dzz0P0qr; +@yAoN5gTU] and drug toxicity [@Y1D0SZrO; +@1BARarxfz]. Kearnes et al. [@uP7SgBVd] systematically compared single-task and multi-task models for ADMET properties and found that multi-task learning generally improved performance. Smaller datasets tended to benefit more than larger datasets. @@ -2821,15 +3125,15 @@ The remaining authors have no competing interests to declare. ### Acknowledgements -We gratefully acknowledge Christof Angermueller, Kumardeep Chaudhary, Gökcen -Eraslan, Michael M. Hoffman, Mikael Huss, Bharath Ramsundar and Xun Zhu for -their discussion of the manuscript and reviewed papers on GitHub. We would like -to thank Zhiyong Lu for revisions to the text that were not captured on GitHub -as well as GitHub users aaronsheldon and swamidass who contributed text but did -not formally approve the manuscript. Finally, we acknowledge funding from the -Gordon and Betty Moore Foundation awards GBMF4552 (C.S.G. and D.S.H.) and GBMF4563 (D.J.H.); the National -Institutes of Health awards DP2GM123485 (A.K.), R01AI116794 (B.K.B.), R01GM089652 (A.E.C.), R01GM089753 (J.X.), T32GM007753 (B.T.D.), and U54AI117924 (A.G.); the National Science Foundation -awards 1245632 (G.L.R.), 1531594 (E.M.C.), and 1564955 (J.X.); and the National Institutes of Health Intramural Research Program and National Library of Medicine (Y.P.). +We gratefully acknowledge Christof Angermueller, Kumardeep Chaudhary, Gökcen Eraslan, Michael M. Hoffman, Mikael Huss, Bharath Ramsundar and Xun Zhu for their discussion of the manuscript and reviewed papers on GitHub. +We would like to thank Zhiyong Lu for revisions to the text that were not captured on GitHub as well as GitHub users aaronsheldon and swamidass who contributed text but did not formally approve the manuscript. +Finally, we acknowledge funding from the Gordon and Betty Moore Foundation awards GBMF4552 (C.S.G. and D.S.H.) and GBMF4563 (D.J.H.); +the National Institutes of Health awards DP2GM123485 (A.K.), R01AI116794 (B.K.B.), R01GM089652 (A.E.C.), R01GM089753 (J.X.), T32GM007753 (B.T.D.), and U54AI117924 (A.G.); +the National Science Foundation awards 1245632 (G.L.R.), 1531594 (E.M.C.), and 1564955 (J.X.); +and the National Institutes of Health Intramural Research Program and National Library of Medicine (Y.P.). -## References \ No newline at end of file +## References {.page_break_before} + + +
diff --git a/manuscript.pdf b/manuscript.pdf new file mode 100644 index 0000000000000000000000000000000000000000..922f7588c16889290e7d1d0c43ead94daf8e42fd GIT binary patch literal 706804 zcmdqK>vAN?k)`>+pCUw7MuZe00&%~L^m3(GoF>`L?&gwgA$yA>B>`kqW>f)z8~~Zc zoR{1m`v$u&vfoiP4~GR{>F#Fkh1~6)A-l?|B2C?Hsz;9={V#v`^*>!q$Mb*qFaP)d z^MC$d|1cd*Mt4`=|HD829N+$8^y}U6u)i7o zX*~Mv;r4ca^YFMo9DgzT)&2hQar7_a(XaNq4>yPXhl_u{JMJHTG5YFe|M%VT&Hd=B z@#tT6H;1GD(}&%0^yl&D*LS}d{pNoE_U``E=nv!3Uv{5}CWUw8LEV%~4ZtiYf4Z*Oc?M-~h7 zesg%VS$@5H*niRz@oD$R{m)s(%dhYEyT`lx(a+wq8h6(}A6@>-{?m_l_irAw$Zz)h z_oJKr?w;l0553=iJnZ?(@9+25hlj)6adf+XbGSYnncUm`@$U9;jX}4&Yxczc$-j5p zGufNbasS9K|G+s&(0?`o!H-+>Nfi0XsUJLzx=yT?ThlUrv9C#{v%WWm8t*C)PHB@zccgSnfdR` z{C8&6@67ynX8t=f|DBot&dh&j?!Pnl-8%f)$h#xcjo>(bN`*W|IXZhXW_rY z2A4q#|DA>Z&Z7FAh5ydNe`n#p6BaKl+R}e#>A$n|-&y+aEUVvH`tL0Lcb5J;OP|lu z=d<$PS^4j*{CBjsnBA)Sot6L2%716&zq9h+S^4j*{dd;>J8S=)wg1k#`kl4^4jWlN zzxLl*`|qs%cQ*bz8~>e+|IWsLXH)&o#(!tyzq9e*+4%2l{CBqgJ6r#qt^dx}e`j0$ z&enft>%X)0-(inxc3c0Q>BN6%I`LncPVA@tyN=lOWNd(_SPjowln zZBFZd{PB0yzv+k7|CZf$4_tR|?hba(O()|if9U0O!NH&CisDcF6<0NvPX8bG z*N^)B`D`}caz|`uQ+C8^H5qSao8@XX=Zh~^OYYOzY_grR6<3SJcv6k&{`he9$jzZ2 z|NY_k!)a=oy8W4&I-{SPJj18yzdby@`*1bBzPr7AYwr#>|D|r>3x7^7ANjkNoBh>h zx!KLu>+AJ?KD}DKnY_7LUVnEro$Rl^Tko%SSIg~p`)PyOv*l{MUCk!b#e}oFnoq{_ z)qFNxY`7@qoB4RPSWUPp+4!p||Ku!;^<;K!^xggS;f<}@g#WqV|IaR#%k}zVzMQTa zY+yr;m&+Ab`htVLT+Fd(xth$Dlg(&0S&bLV)qJ~IvypiL#?xx7tqq@DxY~jz%l7E5 z-gU#gXUt}^na(Ej1xNR6-fX$qRQqklot8E4hAmgm{H1R{pG;>L>&2qmw_j{#X)<_HW*qJI$xB&N|=W*=dhI{n@nf7naNEe6?M0&1Fry=+m#(9DwO! zwQ1Ns@wQ(~=G<6w-BRaY9_NP(%ZKU@@3$2*t^A(ozUsc|uN1jlegBycKl4YbM-u#vgQ~r*@7$fbNhXMp*!M&H=x^opU;+KE+`K1X0kln@AJt<+sMc2 z&n{!q&qVjnzPtH5QQ;3B5jO0Y`17AlezO|0N7kF|0%tI5@&&t|1=wz97qiuD-e5z# z>t`#TsLf=xm@e^~;>&)vSaH3EvEdfr>doDJs%N(SeRDSJ6`lu95)T1RU)HP(zWnTB zvYgB>w(~`|`)0uPMc&%ZZ( z35-VYQp4nDv+a1YSS>cI6#>cFA z`|CTvrsLxSzZs(6%irAJeR%K7^)->iufb=o_kaA&uSS=@8$I5C0Dj=_g47)!-tX@B z$Lmkk>uCR?&iKCq;OzLr_!09e>Obxe`Zpdctm2RRhr18=*MJxN=^ixk!-v~v^W=w~ zgh6!_q=@RpD5zk!%`c|Y*$U@ky=_>qc(<+BQ@qhRhXZf??69ww3xcCCHr#I88DPn* z+h4P1vtBL6t0gboY6{4ZHS2;e&k^6Orx&YDM~`|v*^ZZs>2|T4ad6^mcD0?2*VWi? zU;gih{q-FX?)78Co{2Bnvzf0qm!&uqtYCd{kB1sZ?+<&u!U8XNB8U)@xxkA|7YTmNk4>s9SvWKF!_+*w=3qZ(-yTN_%+ALK{PyJaZcb^vPAhOW`>ni#2}qKGDLJz*TY=?j zb1~a)x_Q_0&4QHCl39~IiJj}Y(4jC?fB(M0l!qTb7f}jGdFoHc)8)-tPr+X8?xk7tWH$OKr*`2v7b^P@gC+yXc!K$L(^ zU;k|8>m{-8WUhA$e>rF71z&%9v7U0{FE;aT@zoUEocRLMF7S@y?N7d9yspNE`}*I{ z|Nem#2A{rtfOR(Qh;9 zf-;}aXTYVCb;D}K58N3R5oIN_u!!!Nf;YR5?~Fp#F!4B+Ih!~@&}6xpO~_WxesU0n zNOC)$UJw#?d)y+p;p}={@nDehaDZjFGcM9v#aZk0{mCaL1C8UvF2W zg_qs#*a&er|Nc`PkKtjI|8e+aJi%_?{wY#2nynx}Y@irGG?+t&$e4G*x95GEY!(-@ z*{0jYbB0Ak7f%)#(f#7t{osVmPT&4)-a@dppv;pkIfATt7kqoJocUyXv6;`i-8&KB zaJKK+M&s%5Aj+rjpxnOY%eVLH&St&bj^PI3|F1#evSwZMg|b`1@o+=y`e!;fMcH%vW#aL=Z`-j3lGET`KIl)bE}7k&QCWU?Vc+%ao` zkHl4Z!cU%3uz0-2(H1MOXIk9n|8|x(n;sA z?O!1YzYRq9ZaH}Im_HLis>~on#V^zuK!pT|+O9=S*UG2M3{sqAxVPgqq?hRoE9U$GbJDW2=>}%?jI18KM<46Yisg37eGbQYvsAr8=30OgagNthuyp`Z zn<+ss1RP2bpw6u)8;le?Vvg&v@}b&>&p&X@E3pLdXB}qSx$Wdk8`vcPyz{wkq}T=$ z++&r9V1=Q{}S&r>QC0BVy@RC3u0{9Urp&>w7#W?Xl(YOFS* zs!ongl-J)&CW#j~E!EM%Nf?u~@Ui|@Vx}B3r8iSmp=O@`IYlB1 zo>b~y*8Z`pd5)-p94j)JZB(B(kPUdUIAZH?<78Sf!`UJ!M~&)f-=C~du?ZQAP#aKa zg;(%iK6_3;Q{{QXyXRfZT?>VV_#Y5ws(oJ@vEj9()B%5Dl)9yBH^{4$FV`o{72E)vL{`M6|e(t%7o_}bM(|~wkT&StNXI3D#MFB=g zDi&Iia(MfRJt$dMo{fqRt%bffASu_E*ziWW+XBwXS)||>&?$`dw-z&k%@v*^gI$qT3S2(Zud;PJ$;-F zTsk7DGb2>-1vD?ue^Y8iPy%?i5P~!)QUv2rE+*3oFHIJYuQ4$oF4oI0C; zSwYN;0|fgpjCEg%locy#>=)AoxFTulTq~5q!Oh{_5aW>Fk{c2b6y0g0Eg%b6)`W~B z`D!2g!mJ?i!IO=Q)K>VVg=oL025!Hy{LH4T-loms%L>&k9xWvLOJGODqN3m?3975Sm3|SRRzujvU6>SnT@vuyR3vbC(&BqzbUu^3lb1N zkQJgAPVkS4FcfMO2p55S$gS}(a=LLtIMT{}X)Q#Y*&0h1uXKzj;bZ;nsF{>L5OAPY zBKK_LY~tn2_z67-XEqShlPKg4kpfkj4M|#7KRg-&_a$+yCk z$1#qW)^G^Nl?DHlL7cBSbGUBDJC9Ls7X8s8PT!i2^|~ulSbL8yW?IAS2)yVW?h+wt-;E z2E1jFzsh`aXmD(K6wOr09os-aj-m-*uY&cVY~X>N!B;8@W|haHUJUx1X`4k0AqQ=9V_MihUpI3wE=plhgJiPEcBt|E{Ec^>H-Wcnu@ zD#(gJhb|z7qmDi|H#6$C7(tN&d5N`w4nUZ)*2iiq+U}>+sUXFqdcLw#;RU<|gP<(Z z+kVO$6+@APC=6RsGs<|QLF@myuSn#hXA#^|Z1vo=Ak;(oo!ul{frpkXgoR|H8<6wl zuOxMDbW9(qm!a)uN*!y%6yO^a9c#8CE_=m&cWS6I&U_b}^>U?Ki&z%$7e<3+qZzBM zXuFwGR>(Vnlu?-YWTP*+p@7oCaFlZf^y1CX)1yiqBAX3xJ_iKgkfePWYqX*zOdT3Y zybeqq+E;Xhsdcs+VCq!7+Y6ROSrIF?VkJas;^Dw4uKK{t2^C3$lPeLa$zJ_xtZ~I! zKZ+Kf0L~U6a*=0m;t){#$d<6}BG2BmA*!L;hPF3AD%sWxrwJq-G!!h?7(cNQkpG?# zV-r@C-T-p4aOBFES-M8R+=3PgkB1nL;b8~T1)PDCVp0)A@bfU#eI-)2m|;LO*Z`m( z{4@ID?s*ST3t_mTF_COzO(4~*mmIex5to8pT&%5Zzz~e}q5fPqzJ`57F`s~jLQJxO!)_VEY-|Eu|ix1 zxPFKd5>kE1Rv?Z|wgEXs)zv~wBr(J8bOBmpYvDTb#*3dJF_>zszqKIgMf6E=UdfIE zRG{vGqiY4X5^19ff;dT;MJ_$6Qy0`QxJVnN4AoeVRw(tvo)wWhn}bZDh)8|m`0`K# ztK&c$;YKkm`$;c0G)42h!cc!k^mHs#W;I!HSrhzm(t#G>4aw6bM27{1AFSAqHR5C0 zNNqvlt#0Wdh9+gN;~`c+b0@tvE75Q85L1_!umZwqin$87`zkFWCksrXY$qM12nk_% z+Ady-dV8wHUvg(s@9kstij`i)5CFh!NE_3M1z%V{-2F{w7s=hAwigh2gyKDAc_Q@* zh9~8Z*Sk`X%LV5=hcMQk70MGXp@Ty6Q)|0SD}Z152qdt@Wf0AjI#p%73wKBCAy4!w}$pvj~!Bla&c${aFFhAob9qY7qSf%s4_L zpb=P=lyX(95_*?$r4i>E&dGJcn}I`iabX|=v4O@tHfCE@P%C7 zoEQE{Asha%^3XokxDst|DYTg5MS(A%Vn1>@!X2e~N8)G&&Mf|)LBnCJKO^L$uo2K!D!9j>odCYm z?aW?qc`TG1kqA`;D&`#TAP=KL6r5)fBYmttE0iBcxdi01fT*m{ikBQs9(&zMkwEZO z&~C6bhXp1)3!%NWwt;&739q>IR?@Rez5*btgck>T$TFfvOAKgCOYVuF-awSjg%L~> zu&6&w!Vdk}0K;~SlM3CdfEA}~Cit|Hojed?mvPyJFj19m^1LJsqU1ZFST$B#(bn5d z9cm&^m=;Ua0M^{%nO76}v^lJH@BIn@$30a=iTu1YG(w#l7ctnY2avT5rMpRaDUP%E zEyy$hq!5>iYIFbC#JyAt6Ao&B62o)xt=H7Hao6hIBeCFPwGnL>LULv=bC`M{A4>&` zR3=fADh$LMg$PVECGb#uIT~1T-(nKJ0gP}Ts%=OtL`(N7bOC>zU^nz0@0bqJsc3Mo zQtQPX-$Qm?6mZGY3GEtLw=USR;Z0{-tAB;=ND|YEA*!)@#Y%S_uABNhg{sZWvx9U4 zDMGd5<_;qX!af5R0##9V9TW<6E8uJ;-lf@7jrC`Qs_jYXZct{;c1%RncPg($@bGd7 zXJKy`9Lw4OASt7(geXck1c_m+KO=|^!RyHC5?E(jLCp_0)uie>gU7fb&*SAx>X3rp z6KcdLk;X?1W8JMtJtw`mS1QB8KMZomr~yssWUZd?z`-gyJSjGT*rB1~L$wWwTkM&0 zk~-Nqq-v8#3-*Fy7Q2a&-}Ne~$OgNKPehJRHnhBn=YhiUDmA&PW9}@dw( z>xCf{xiB%~EOKEIjI*|TF2$`;>{Cp1#E<751PQXryoluliJSs4pn4|{Nm4YM<&-)% z3ZPh{lWq%3-#FQ1fLQ&_06Kdp1T`S$;U-`f&QxbGo zm@5T5V#AQq1q1DoUnpasgLz&1=uhA18&6<5Khb((jP7h zM>H;7AS!{hL*-IYvjBKSh8_JZK%p?ypAqtO*~H`lQ47v>biwiXzCrgEgV4!B+ETJo zU=4!!LdQ9LIpPB!s!eG7NTha?&LF{DnZTDWSX|j4Y#G}C(nBRu5Q*+&Lrcd5%+C(W zt2idJ;IwwjhiWkzLd4X1$)%@uzAPU~P=Yr~NNB)B?3w7Oz?ZFf@?~WBloi4n@r5Dm zi2_!sgBhfa1l45S-;p5-RmoG?5~#@Ni7v%c_PA=3P$uPG zQ2<}Y9#1U*QK&L{h;8y*Wv>B3cS6}I!!!L1U^(I@^3$d)2LPRik=(%&jf z!OI~Z@kR-W%D#N@qCi7HM8q7AQwOXNOqWn9S>_q8Q#uX3Xl9}J!GoqmO%H zIuS&NUN4`jN-J5khLpN&4RBF-v@udB@QSOBR>}^&74*`L;H|La$V>!R2xyc5SH$;hghVL%tHb9 z6p13r8JMZi5fS1}Jbq;ixHL#VnGA@EKf_p$RwygELJ_^BRB9@rYS3tJuoRQ!aLM5X zR2BCq7Z#A^k{gOoBoQXGg~&r@E(LBAF(02odxk72(A23GL2L$kvid4fFNr7Gwie+W zwo2L{J~i8k9vB7cimHZ^fTHJT&#Vw24H?LpS!Grzrh74B>gBOhV+EEYpdGx|=zN4U zh>CT%c@YPYh%Fuy!Bf9!5QDPF@VttTs~lcu@NK3S@AE!xa@UbMByYA=cd~qHgPJXG zmg2ZeQ6{FDNQ4bQiyrD?Zl2*Whr!s}KK61CY=b!oyBET(GVe0GPMH8W>lXqUd z*yzJyPP~^Xgt6{dajGn&7yrccwogj@@V(%^Q^IPFn=ao8&S9vcv_wxB>(7c_{1Yis zaNh=rB2PP=Y9$r(LSliT&4!Wy=N6W<)lv3nMISQODo3a3WvEJsJzmZnIXA9rlR+r- zcvCw{i8aEpg3!-k5lZoR`$SWMC8~ZK((ZK{JW&Vw0irZo%}5yPZyi+o551yH42$1@ z%1@26gf3%mmuQMgaWX|#@0oCS+j@$rW$2~ei||N*YS=I~Q*r@brccKiiTEKMFK&RE zUt+9}^|usC53tyzCkkq%R8Isd92(@rR;V#2;SzYRmA?#Q1^icB22zn$8!1s*7<5O8DS^@@B@=iGRJn+RTr~kx zqd`1+YTazNsj9Ua>u*OvC4`&Gj)oqUeJ;?=P}kld%Ns*-)O^BNtW=KBvM+(;lINn2 z9Th`zk)stuaxXhtiQD)Y`Y`BzqT)-*f5}SmEIlnFmRux_&Ac01K^d{+S6;lEdf7>E zD+w(UWzaB5$4$`^F^mNTlHRHn(lj?AW;a2e0~9ns+-~@Oe(mv zzw>MvGrT)wItxf}_#22f0byCGvj9pH#Is^<+!VFCCFD~vw<))TLhmol3iX0UN8d9` zX9_K1?&DecJ6ERAk~|TX_mdhsBAVGgVstC{L~-HCG;K!Ftt9`0x59_&m1uh_rA~D( z`^iql3HMa2L_!mMNAy-53N^kpM_wToHJ3KjpAEpVa5^QQIpy-z?{;_9RTL;K0Gb&9 zPTM>oF;KdxtdmF$RxKo_2_OWnYZ$AID0ThB+Z64;K(?0Z8p3e87nw|kB?8t2N=#Y2 zGc@O%sN_ZJ8mMQV*hVcwjY-&DuL9^!f*en`P(w3r2sKhiM)Llsz0ugix~t^vH(Hj! zH3m=cKDJ~5xyX2)MD~&VFkLi)N(F!^Jqp8Ey<(;Njnl$CPdl-sG-xs>&!nuuNpdOr zAi)_bEs-^=OF`S2%U3*iiA)+Qjckq!xg*Uc;a(r>ezsHlsRvXV(mlQJ)Y6@Gh)tj^ zwslbVYWjgp=mnNGVIr7eR(2zYoIo`|>#(Q*cdC9sT!Y-W+1Pq&9TfeOM=4AFg9XdY zC5s^?Ma8DTP|hT=M#4adu!eZ$DY7KCdPXH_`A{U=A@|ZqcqZHl&Iq0e*C_9WxUKFX zLi7Q2Q>%Bemei4#ZI|gGrWV4GbukZ0Gdl7z= zvCJR>alW`KCX2=edj|O=jUb7@@h8U0FT0Z)g+^aUn#VR!T#x&O#Ed1)%NiFYUqm76 z6#bUzn!wI{(u;zV!js5TYlY!+mKs(8sZWD`xQbE!@v-hpkt!eR#ZeUAQ{Mow6=Vip6!n%W0kQ=X43MN_~oEX!&BMxB!SQ|;XL3zYN=54h_oND5bFxI0LJyc8Bt4QWQ zNZ`?}k@r2+t%z_}YYZ!bhd`8Bw^9l0KNFAX!BHfMff8ZpY@!QS#aiCXOH?^@Gf6e6 zb&CXxr2k=KjqA|zQPjPvlkzH$;wGnDVx}9a;ZaOoVg}ZgmP?RjMxT?kv^A-he4ODB z)~Xx16e3}@DwCV0#2N}GUQN@uV$WX1q!lJTLZwp)TVG;Wm=FR*J&WKFltl4ScnS+kImUpc*1HRD@iu)U#JEhk)t_qJ>E&>c$L-~D31fsOZ zR>0PvQH zYr!#;f<=%9&Rh%bAH?lc#v(%5nH8i!`9z>iLhG=n#EN9ad_QHWM7EHA0dL77$YY`B zgbFOl+^JI#C8=<%iRx#%5v!r@OOZ694_MI^V3obmQn(*N`v#-Xqf3YrhEcW@il_*- ze5|__spq1fV*&Gm5Q;X{YL4u#2(qY#;KR0q69doB;5y3T=6GQ?)M!LYFJ&D{>0iZ5 z5q(WAQn%1ud#b@pNo_m6u|!A6EpY4S-o{qlP8Qmw%UB$)Ms5UhSrjYqTmcfRJi!Ek zY@lANL@ukff6~u}gBK328{(5E^A|sp3oQgECG$^xe>wHY>FN5+e+|)-r1opeqb`0T z9~*rJHGHzhunKNZZ9F;fD!%uzdKFs2j#8w5In)(@g7&(U0^~Oiz~`uC7$o;jh6gu6 z-vv~L_zHm^$4)_6faiUzMeNfhYbYxWJgFqBGMz-BC+MPvwJ?qYIUFdtaCe2C)J4SdOrWp%SogIE7{fF4 zq!&z4lpnI>aETynSXCs=y*b}mj6UP(}3Q9U6?#K-!x0#Aoz4B!&w4cX7NdPzav z3j)F>DWGJJMIFlMka&@cg$gn;4m6#!pd;~Qw{@>xQMDSmi5&MzTdKjmO6|H{+$#WJ zb&{g(Et~jEDw!vq$G&RJ!Tn6uJy$B3AAk9IC%-S;sqAGW+bP+2E-}IrFI01329xB) z+oya4vU`zaL()O%<@5vSQd~#PFx)l0Acz4}bwz~hBwp}sT~xKO&Ly=Dy;xx{A8gXl zI){*)bp>URmW6oyTPm7}62&LAjCdI9?+!yu6PLHJ;~&TgZz?G%ymp@`7n+Nl@ysX( zA0%eHZ%`QPZbZuCHAYq?)|eYN#YW(th}7t9n$C>CKf#^%u|^|W-o&d{Bm&ZVJm)4B zhM##8TNw&3wFJAe6Il`ecgbH7VufZdtR+Mb_@-(tjQ}1h56^nJBwo&E!EWmOqhr9I zNHkZV3@hdtN?!D$uxmK7!5$i1d4N-0#4j7sHdag-VTyzxYDLPh@j@tCIrXE77ecOO zHB=v*(uF{y3i585M6YlyZeb*aRima)Per|SRQdEY_a;TUM9M|ApQL}JHRofs5v2=Z z;BVEaJ@*Rp_GUQ1ww=zbkUlzz1%KBItqAG6=YBnZ2yJ8veG@5n@n}133$`y^duNEq zTXhSWP}HgdZA+qy)mZnXNEx9lJCR*k6I_yi%6qKHf6!Dh5@Dv4`Mt;>E|UDWF4xZ+J?3#JGBxjJmPUG3)!jQ1IdD-Y*e4v z2%;=OS)rUqp!^F~v_0A>E0lXKR^R3aq|X={G5Jh|ky2<)Y5_vJ)Olb9AwEh`64_`U ztJk8f6)7L9ClE;WCOpe<=bf=g;~J`NMQ1^)4-g2<4=Nd#t{|amth*Jdwcwd8KNV8< zVv<2a+-FKAmx~GJ5B@apIyWt;P@WMY6d$Q=_##e)7$$V};dVml%qPly%e+Mb0FgLV z>cI#=A7aZi?Ehj+{BG;4^jQ+rOuZh-tnj-uMkz_1L;4Tz!;P)WIOg@*Z zq0$?be*$mU$NIB^R((8S$k}ZQdK;-9DN2!{0O`yMx#_1ixdU0DG#gAk{&q{&rL^R=_Ald8Se_Qc)&MpQ^P6w=o$VS29v>E8ZC& z>TX0zm%bM$OMslZ11Mt$mnkwrtQ4e@DwwO26)oSCMF??Ug()Cqx9&^<(F;Bgpd{(c z7A{IYiztadTU()f<*qI=61q_2)ku~C48opDRSxAsoB-PCSsOa<_mHhBm)v)K&}?*( z0WSiQDD8G;g*E=fPO^?zXzI$jRwUO#j|x?-w6ZJ^$Mm*;xQbr+7AX*0+rLvXt5AMp(pN?_MZ9liE=x@}TMz(n+~jltL}>c!QTh zjVn+4Nzxkd6tCC4l%b5^9oxXhvYzglD?yzhxM$V-AYKV{(LuCSKCRnpAR}<=h+IwGdkgFb3=sXA!wY+$p6xK~=!Fa2QGQgos$xFAxWOtUoVhDpEbn`jVyK1*p48 zDpnU!j$9|w%shW&;b{+~x?SZA@fCfjHlbzco<7*3_?AA~VlJUZ55G*sAwE;s+dLdq zJ4swH6UUQkAuL%RtF34o3#JY>KzJWLTU4e90%3KnXI3bpMV8ykUk0+Gm!|^61g=tb zC?tO3+e_jnDyDh@6A28mMy8DQw-&uT+dKqv{k)=666w@T^TXE4i1I_Cs~`do-G($t z80&6C;L|=&8}z{$hq@1XOTZ_s=TU5h>8FE?SW00aBYI&2884jnY2Ic{4!VJfc0sYv zU^P$yLeLC8s!BcZIt}%=61JU68OWlL0Y-r(idq>PAsPb@QIMV*$cla_GFC}Zy=k!( zMxUur-a;c%w`?ybn>TpOo+3_E0~TqS;4PsP2a})jBK13Qav6w{Dnd0Gd+*D@gI zHF`aecb-GtpEqmOCtvnnf+cqbhF zC+g^gfb_V=T6uY46E``+$zaTNF(Y%afi4C9FQy--a?KGRxp)1DQ)lE_wOBz8c@Qy5g3Oev3Qr1hD%B%h=tJF?B3$3kg1|U>&fHJdAdX%oVkn;L zP*57elNls3hti?w1sF9agm$kEjKa_7q6k7$;`YE&@el&i*$qB;kVYIH<#2uD$USYC}cI?E}N0(5QS_-s<=M#PY= zS^G=Q6o4F{i7JfB+Fz!vpps5O+=QmI`CuFyO?~j31|Vb{#n|BP(s~1WBJPNh6<0*o zw3zg<{$3Nw;NmQnDNNB965;qN#q8XXs-;oP??=I2bz!(+I5VBCNCKm?V%UMWMYRn0 z%c70PCV-j%YN*~tYNTZ*B)3y9ZlW|aD6?H4BPFpueQe~z@YRV&*s6rC@L?eMu%U>alxrto^s)Y|AU$9H;Q5ZK`of-BVLPho3p(x8q3{UwUl=u6>a{jEhWuZ3Ewae^lZ5=wEV+u1okT%m4+Fu#Jk7nGAc%9W$m?2nwa5h3 z+QC$Z&@UZ~A{*G72IM?*$Js+%j)g5VOz0E=Oil(nQWNEt#i5aw2Y*F9AbB$B_pRWm zS3`~K(318xea+hDX-;kPTn|Mr>VV0_(7avF$(a?9$uI(5kDL>S60@yH#(;fjD#AU)gBo{FKi6rCR%QBk0}y=GJh5Fa6Q6rmax?j z^Y4Xq47;0LGDW}Q?2W|oToJ-Dd8h)GKme7dwv8PoUL{FbAFEfP^eh+woK$jAqYQb6 z6J{!|%Sa?oT47M=BpW2bIEQ*j+_yGbCCI7`)h3i~q+ZZ}(-$Lhb^eh^6cUD5)5l}O>My#q%&qDtx5F}l$VyGpq zq9|!L)}Ixq?y&ePf_&*RX6BNK*SZzh=X)v8i_A&TlzK3q%&`Dwh`7`q3 zuvH1SRdR-47LPai3f{LFNh{90k9D^q_>u!T8v6P)@A1akK2wQYKr1zv7X64os6G2-VYd`7&x0m!KmGIp%^Y?y)PQ+&w^gus@Sc?1%#!m zOy!wd3zADXqk;bB--|;f`6rYP(HId?8vhBbGsx&tCCL3t=1fdX)MKW2Bh^@KMcey# z;C;k>bSe}FW)p5i5<@qa3vixP`ncmrocUOHE0S*^+qJm=ytqfaf-r1RcoHgtp9@b? zCBGz<)dZ3!MtoMXe8jF{th*7ZrQmakiB#4;%@_yp6lrQ@a)LaCs4{ak=@9`@R6jn6 zPlNkVE~_r7Hr8lGyGMk_%kS~mkQ_qmJm(S7a%4Uty{wpuv{7{onR0qwdDB!Ekx=Qx zT3&V)wX+AgIN6hpEB4HhOzk_BQBJDNP_8TQw9pDySC4nZT)`W_smKqC7K-x@ml_rf zXf-~jkM(B-FRz%*i0Md*H8UUq0zqT1!qv~LAa4%47|>LC&7rKIe>v%T8fj9-lxu}u zxjcE^$47*I8Xyz1+jT4vD{F1kwg(H>M_I&>Ln4x_;MLZsZ-Gl<6E_!M47yXrjs4nAjMP`q7qk;pkVJdfNnsu zL87ui5)Ec9>wuJvDT~)ig76!N>b*0dH@|DP0kfrg|b{!%7|X7 z1F5Iw^(Kd7+|Zcx4PR6>cgWWYe=j6`r>szPO?+3DiWaefRp24lO?B4IV5=?@bfzY} z;=RP){GJTS3!s;}I&=b|u5;qK0gz4OufXHH0+9Z3u*Yb&&0P(6HT1c*k)R?B=2+#P*ypaQ1L>21TUGfdl&p|!%dQSgk&XKy@~K~hAv-3VM$pT1Pb^}+ z*4VKmYgfvRfIm>>LRFEE)wgk5{2_HLKV(TnHstvN(Tm(cZ zN3hmB?V+=Td8T2kKPyE129!lQA%8>R^wWW`YBZ0CBah^%odH_s^wtK0*9`@67Y`-3 zoxvvMXZ7>AtCagxEFcjEf&lu5qiT+nBelNlEyb@Obb{%tvBsrn`MmSHuT~Dl=F55B zwHyu4duqjcA&^oRMEH-=0W1b&4p4)n#kf@AL@ZObgrJvRNF;T6VBi?~SmS~{gCCMc z2)k1w^8!&Hvd;Vl*OZ}NXGRFE!@2kFgaa92+)KsPR&0`ajgh8>BY;Yi7oSB-Awm%B zQREU;b~}(2#=R`nS3(VkfS1r#Vlgx?flr1LZ&Wkz8PIMXePhT%fRLP&kM*~oEHkZ2 zCR9bii}k}9uRD$wXs?8#ctg&t;2dx|V0S7C8_0@Y^l1fDt4A_W5MVI>7taQ01e{AW z2uV=Wj4711D8s|S_Ob43krM3dsj+p?#?HQdCV7pM~zG>rvWYa3SGu8m5K#9A}Cm zI>dael1Nflbg~X|?c`U*AGEQ~RsaROOopZLGJAOCbjRGotpWokrYs2XVXRIvBMF1D z0WT}%9>gC->Le*ER6{xAy~4r6FV4H6=9JlNsYK&&vuXGbr43MeRjX%{UxHdKtr9$U zpeND>KGxq-z}Lhl4PAAztq^^NOgT%U&mCJ~6q+qLHyETtS)oqyQ-F(-rH}wzEmapA zA)gmKK2%DTEk*KdTkeD3F6!e>-c!6{fL)coOHy#jUt*1@`qVh{cLI(rNkNWIW+<4p zLb(s9r(O(~do0igMm~u;81aflTH?B`LP4@SxK~x$fJ;{d2MZHBAB&_Fz1*|p{N$5w zz4xZ#J>H;Du?c{z`nnKBK@%NssPwlBz6@lJ7y<5CWjWR+Qkl-;rQg3&z4{;Vlfw z2ho3_-}qRMRxknDk{33P;bp=h3=K(qrg&7Q8KN7QzN3LED3=%_k&pFgg<^uzwcz9O zzKZ+C`()}EGr8YjJ$*&M}-0y7rt}AwwlS z(aKpY%5lR`y%32+?)h7heDN$S=Uyy&D=fw!*tRC;u??y@kVdn@TsAW!I{)dL`N>aZ=}6G$Pm;5G1M5$PESK z_TYrBsx=Rp17kF?1fIkZtfgvP%PMrXB7G(#-@FDLvt&i)TA17>mnffKEab=vC|Il$ z(Ul7X6j_nnQHlk5jKDDpT8u%KQA%IkR4_slrEUzaHSUy!f`C||Up0`mYRYV^^HL;@ zpdf`k%q7XSkzHM69k?JaP&gzFkJt);9g>cm7!fPQaQITKQ0dc_?h4+?JhlRqkcgSQ z38uod9>@y*5I7M8<^{bR2?M1*5?noXj-j9)FVNz2= z!j`9(n?(tk7jazI9QXN9UK8MSt6Es$l{d8(y@XF@eqqNDmSHKn(t1AY~7kNxBMK7MB zSfXHXRO(nV4Ng0e00{z>+)_j#nP7d z2%p7?zMW4+5+0rR0rCQMP^m*xc}s@-Ov!BGazT!ffDw@{|4qH8V!#QhS{v(ZMRG$C zkMU%cG%GPyHSnh<1}Y`dBMGq^JJ?v&+(xea<3(X%tUnXf>k;`H3K~!?%=OvXBNUml zO_aTnYB?LLs{4YY4{RcdT5c{li)sjxS@)G(2Wg?8jzaZHfwRkN!*dq&R4AMW`G8Ai zDkYH4wa#SE#J5!55GiUO>u)907k&$$%XAbnUpsmIFcB(2zSsz0A7Ul0Dust%un|!y zy}Vh{AS8oGn#a}javzt~P`wTbJfWqFxD1f=t7Hq7;rZ?4A}-{4eBMP&9-Q(E48#-u z2pkz=T!{N%#fZp&iK*t^Qo|yWnr5H87UUd!tX{5`phMDxKF&VuBV4GeI92a9^%E{Y zq5Vh7oZ4+sE|swd8pso%NSaW-3NTxh{^Jo*rr&3&awA$Gj*xPOu?rk(I_Cm^x1>aD ztn(@)V~t+?Lvl50wj0u&K2x|!eg&)F1H5^*50Sb=VAZ+L#yVS(TnKJXz)QNxA`O_$ z2nT>u5pwX=RjZ?vJYKCcEGtJ{wdm}*7qmchBBTbA)B;J)zn3RODh0r)?%&EIk?5xP zPvRLT4575Z{9Ab>2t@i=e~TbN6JSh|T1pP(K7{0Ee4A^#oLK?0iabYEayN_>y=6@sg(q{j(V zo84H!QhETSt3q3!vZ_I1%a@s=y*MV4XHk1fSIk%eCV+gRE4<;4McRNQ30E=!qI6UB zPlmDj8f$q23=5Y;S23?PPX_P^a0A_M1dq^)kKhod46g~x75!rTTT9 z^2xCYln>S_Wo0I$HxoG!c!Xs+f6|vKmR~AM3Im*L8Hfj`S8ROfd{E_P6eJ2iWWm&( z-iGRRXt|e?M)Xrht27!s4(ILpOiwg+nY0GRFzzKeX!Qu_o5$Wum(3|5>2@noMAP!n zOl@J1g`_Tx#0Mdh*Vs~X2XLiDW30cm=%pGsI@dnYii{tr55g~C<2~czw>;O${nU&4 zjTGje>>2yol@Zoqg`6`rXF4!1TTyx{dRYlqGhR_9r~Twr%Q$b)?_rAxW8hvQqbcqf z=O@1yIQdTWaw<%xhZF}Igd*c3GKhExni9(UavHE=4(z6)C&(8Bw_&WuQ^9k~IV)4Z zQspcUc2(UGTLDuB1caqA3kR~Imr8$m9@yaQBhXu*g{r*6JP+f)C=+FT@M^5TwIHl4 zfAT~lC$Uf7U+K%(3dqAGpEwv?!kaJJ3SdrVdJ^fAJ`9Ti|42CjlE+j$1(7~=JD>u` z$NF0ft0}1H@wv!I^jatjpf_!U>dHB~E;>dvRNGMcS_o4Bi!9UUiUabbDWh2@? z7K7&W2~@{xTFMd>!Iq==h|j%1W}4@Yc1gG*D3}pgz*!M%M(SKPRxd?cD^j6RFG~SN z$AeIbmJ9o&G63+OEFT=P8RaV+2ID2EY`hw3G@@lhuDZF`YI{l?kQ0$hnV=zClUy(* z%}Z6`)HNGqlzvJo7__Ln>bL+-DrA(0*9r#9`WPhrMM>bR-pX~EGx1%j#i=5y%4k)g zqFK|*Rcd59Sh{{yMk^V8z4SKJxa~Mnk?M95&FTdlfdfj`W{l$Clal_;uAE;A{) zoq7@I1BC)uASY&$3nELC5@cD(_?ze}= zR0b1mDD%YM5wSH;dO-AqS!ZRQQdd(yhdSy(;?tA^uVg{Y)l<<5oKYX^z6wDd=Glpm z>MG)-aO9;zmwPqgkI*o%lC{!I$;*bT322Qo5M-^ySfbsgk9D_V(6x|uRsBrZ_pWA2 zG0O7EGw~%`5jneZM<{I~&JdeCf<`PsSPj)H(ej4f4!WOc7eo0akG1z^h>YL}kZ_`G z#d|X-S^M0Iws%v?3Ztmy2|?lBy4oF2i*dcZPQOz;&IJ;>v!qmt{st>A5{S)56TLxA

1w_e>t&q$pP9hzz%tF(PYimW@efOFv!&!JC zy3xs0F8}#`m%Q&pJR(Xh1MZ?b+x4ES>eTqoRtrW{BDfec`Z*+RkoXnsZrUQ2$P9A; zNe)70Qa3e;b^(4-QLnnV1>g{_edxhxVqeu*=d)AE3p~34z?gA6))LmE`Y-baP#}eu zNSa5yUWvSli%diXrN-)Sfhk;-1>F<~`(doR6(Nd$W`*_Ep!T82AJgr``$%Dd@m?b< zv@$$x5VQo!4`c;?6S^?SdY4eFNif5B+mdLOAS(I`Xko$T-S943lPvzOOZQPS49zzT zbzh3l$@6h4$~r+xV~L7wNjbHS7m*Katg{u#!>#p( zVvSIkzFY3GfVroRzFux8S`BClFCp}H6&--4xb~I)tT0krkZ>tmJ}1-QqrD0{nSGbx z#z+7q^Er{{3ph&ttf2gpGXhQzbgF)SXABeoB1(N#60Hlakb#nNpVeNDfRDP4g2G9< zw{WU??u=*V$Q&O9V2i3F7-{fMLTMW)%&icGlq}!GLJ@CUSPJeAb?^)RdXW& zfU1#ozkGt)>Fg|3OCyDlLf9^~7UGstKOTffa?d|t5ncy;AeBYOC51CCW-8avX;9yt z!ji(I6+G^^Cx`@>G|L0z;z)$bBQ;%Iocd}2K2AW0yzoes$H=o-%Z++JCGIh{Sqz|t zJLB2Y-U~0w)Jk|I_!KbRA{(e~qjVg?I3G$KJqv6Y_Dt}i6BuvSNc>=fdYd9A9-v3o zK(&UDXH;APU5Io_HuRWUiC);Hd}J_p=%s3!b!G&Kcn~jeNvQ`H8j(5@y*w2W-C8|7 zDJ>-YjJei=B^UPh&ksXm56^9K36!xYm9wO<0|Cwo!mej8XqmlL;n&IO>xKAy*+FXh z{CK$(jzoKC0}j$CTAY7nQlQ2Z7k_{X9QHnEQCQ>(*wlvVHTWVPxOfL?qE+c^yonN3 zwWXJ;4&DS-C-+5u=|sGbA{*kZlrW;@LL?6ZJC!UaPCBce9lD}0@kDRTw#;cIREJaR zjSXalE<`AhWl#w27H>2e2?~+`4x+VCPGN9Q8ER+fW;W%9& zmepA23o3ao)Leo}*s`8*=mIAx;c#B7jtuY zF?f*ycYtw06$~h;`rjC5s~YRkiauCzgbAUQtKes!DXBuMDp)`S2O}ZrqLa^+j05gN zjYhPD5+sWMYSfuYjw~vSqDCkowe1L%C7Dtv#mR9&3XWX2E-Tar$=*_>D(NM9_M*@N z9UfXNFasTu$=QXm#ua-OG^M;;p}NI^pPU3mOpq8BI; zxi`TqIa~VO?kzGRE@o^(%*9OD(9&H@2I0NzCY2aK#5{Gl(dn%g0)np^6j73_^HOF7 zDC5iUni8tqBxE8We`T(OnzA7V45q`!T5{o`xgC(M1&O%{Fe>38=^$XAF%9s1tlmv6 zS0Zgg9}R>hjyNHy$J0mJ{LBjRgNOmW- zI8Yl2K+EI|WQFCN$TY$&!duO>!ju!K$_dI5atNiz2CAlR68I#iCPm~m(XkC<{p~0@ z?!ejv(W+awx!o=~HEtB^+FrSHKu0^&7^zTM<07fgE>b%c%HJe-iePii4^9ARWM^ zVpZtm`(eG~@?t0%_5M`FjU*n^3@*p#p4V^IE4XY zBH&``awDxhLbknTxn7BYOJ$9Gl8@C^w0$iG^<^Xf31%6NExruAi?fxG3M3(b57job zT#0S!OjuN5)miFqpL;)%Cq}k_jJ?XZIOc*8kW=FIiJNn=Vw@BoYP6wU6o^CcdQl*J zM*qXR@sso~H6)Z%>(vXNgxK$Ciq9gw%2%W_s<3v|sB?S^eZ`~|!U0QDu@YA$C4jrA zi5gK~HF)asLxxEY2=A_(MJk5nbop4l?b;sylo8Sb<;I~h982<#4oZAma;{c5*8(_!E*@-2Vpo!txGM-U zh6-*P`+~`_*SzbtsAJVwe>-Z5m%T=301y@$M5Yxi1;_{lMo}rONSlwsK~XIsx0h}w zQAa3Iw}Rp@)@Ve_H$|lruZ1i^vdMW)SMd&k%b@&!Qw7ClgqNt|ND@7#`8LSIu zC8uF25~zuc@L(=gC5C9Vb&PCWF>dLIj0;Sy`agf+4R#oE}UV9(=!q$RDW`GP8Z5{n%R)$Cg`XnAB zzzMJ%aFBRP5URPw3Fc$}Sk%d}aN&=U-|o#xKym6^XNt(I!2wjlwdv~?F>u|x{;Opd!K zAG5(-OuZ;pB4xel1X5+b?!p{l?uHIhXox6zMcss|{m`D0SHfTMHmjf+Z^eh|1#26; zr}R&HX{2FQpVkWi55*^1G1|S)Fp}FAOM26pTt<6R!l4%zFxGu3hTTtO4guoBqYpL7#x(nVsk1)l->O z$kVc2sK7&K1lA}^0UUv%LkLa4Q{ZZZvCd18th%Kph(Dy%aidJjI3U}ZyBg5SAT==~-S4K9ig5nBP+%d?Mk z3vw4&XxXZ4D^hu$UTT8lv+XmjLa6v`OZJZ%smu_n%1AtOvwBem;ar1x)04r%*)uCl zSW(qo7S5Jgk-BDkIo*)x$Z7(sTPd12 z^eEH-95o7Z9ID&Yra2iQg?E$L5|TZT5yo<_!hu{rP+S!{HNb4w+k9q)S^=@0Je@v} z74+p%s}zXtYO0=jMH!Ap@&Avv)XM1Sm9 zi6V;uKg?Q~LtRQ-l1LX(5wW*Q&d$~XtvIe;$bO>pm07V$Mdj4FPfLY51qPBzh81Xi zYjM#jU^r3r`Kwb2s$(bFRQ=Sk=;fX@o(UjNXq^^YAx;xfhlA8*R;*K*jb6$dac03p zwxJ7Zj8&R_{Rs=Sl57)wTCG{tgU%Y>$he2eC6ur8X zmCQd-E}przy;Mi&@gax{Jn$#`hx-c{go0zEPnXO;_5FJalhk@4+9oHERw@0urClLN zn_PuHSjPmbQsjKPS!yd(`f+M$H;@&*G>#$6dQzAjL(MZ08=+B4?yDGT9&{}tC&V#v zfWoyJPe)FOYdVLa?yth2r^1?^a{IWVSfoB}O;4*Vxe46licfD-%V1Pom~=%|{;<8d zSaGBH^0YmL!~_(=Jq76vI87Wks)$rZWXS*<>Ana_3sgB-+NA`wGeV$B=R2}MWh-{T z`Hx{NFb-~&SkFaoR6>KQE(%)0pw7}Rg#O<#MgSKLyaBYY#M>Gb1F|#!ZVvl`2Zhj0l|FVZ2vRKn|>$ z?MaI@wt|?5d4xI4-d2y3bjBV z#RK9F%R3d3JW`a~j;hFu)Y`VG%|tpHl0%`4`AqGLP%#f`Z6+&6`8=J-)o7h`7me?O zTAL}kg|XV}XiL3GSwXgiG(({dtCc5{zjjWCf9}c(D^C`)&y+_*MuxTK7WQZbd4Fzk z0LOgcfyf86`Dab}&Y_KET=o9?ek-%_e9=}QRn zEXoXoz{YATO0QME7qp~8oKN9sT{;q{<1HXnvnAtom8IjU6mkrI7>5bjcxovsf(Lob zr-?^a+r*5G;8B>eOTimtMc7h>Z=81?4_&~#{+k+B(`xd@y*by%^jmFbM^Iv4Nh8EHhSZKiYJroh zMWDl0z@yHrz-0u*4R}!kpMgIqeHMr?+js+0WLU{**>TUtM4uJQ?wAl7U4 zBt9(I8CCg`0i_41C?&Kk=MPwu1~cs)YUCi=`dH(ZYWX5A1KHqJUPKHw`boDJQS^@7 z7f~q*&@=HJG6P(Ru|lMUhH8MVn$4b&)rGjDu!Ui`Q7-P4`vG-rTrY0sxH>;2VB8D) zJZ7;72UXu4anV(MN$TvBdF8eN01so`tw;qYkg>c+KuK*0UL)F`L|Or{)lA9lqkQk} zcWQn;YXT}7#yVTENmY9pIW<~lsma4=W_N#xn%!5IZ61--- zE4T@BkmqPI0IcD#c}Iw$tXM!|q|qS|Pyzqk=y+VkwEcy45L;nQf~%ZYmO}hX753ep?qrwc%P23ENoblabz2h!_Bv5zSejyUaah zjDO%bEBz+TC*8InzypmKCDy90OinxSsE>8Fq6!P2Iqk&SY6%xSyi6m=hYL;$tmjyM z18pI`K8~-&?S=9i14Agh5hNSL(LhARYW28n)%(m;0Z|VkzoiL!Fl***UKl7eq0*G{77?I5WbysGNFZ?37uNzM@Rdz|z)W zm%09+&NHbiuyndI8%dkCg*+fIw&w(Ez}F-TR5S{_kCv$q)k{%^XDtGQGQXwJHl0_DjTkS(8$~({bUV_SkN?3 zgMdJHirifXLd~ut<~2%5%$!s|Pz2-f#1x@8V2A4^N4ws4E#a)s93Dwa@M^7#q|T($ zsN!t+5GYStn>Maz`$^^45O1YUU;ZSFU41nl>ug3ORiQpeMH!$bs00(9(1F6NzW_kmV~(xC*5ZPj`&q_Lz@? z*pnAClHLRI4P%`bg6rNx{%0%EODIL)2YfMm6gulgQ&wy(s!z#5S*;&+acX=Eunu7ec$7$exkTEKA^}2_F;Jk7b+;l_u-MB| zxI`FiB$WyPj_Tu6A#7qOCJ~0dAqX8eloHdNK_BaEMOuMF&gC!&}hf z?h}=Ww<0sNKcA4^bRnz4)W$j+Q3+K(I~8gLvBsD1iu&9DM8F}n4PG}*Tn3K>vW+Av zt*#XE6^;ojA#PeVRNIhnQO*IYqz$U#RX`0pkVBbk1EACbRXMMkj%|Q1#wiGgejppb zkEj!dpHa}amjl4HI^l{H!zZ#~uD0OpRj+m^vf(pVY(MEhR+IX6USed;A|piHKq(hw zL8VPaMkF^=KNY-uZBWDD14cH`VUH@f3AuE6B%N$%iMTI>^;va9hbk`bd9A}te<*EMdnl{IfE;{#sqTF zJAijbho|fsOS=HQRNC=8vU}B;M^0mG#*!oq9=AkQxTyy-qmOdCh%+8TC~@7U6xgYB zI73icy2a_W=w(Z(^;Z3+Sa~Mz3uH#czkuCBx?NP)(UmaPcnh3+_ww9d)SNZ7#;P@J zu6J))H#i%F3IvPi>@32=Qxmu{cGj>~P!L0)Lp_D53B^`Gd?z6{B{w4GP@xqmc^l|c z6c^!(m1IDpV__48VdG^8Aoet-=l`kT&dFQPst{Tpi)E^mLLe^%T2QL!_?5e#B-e;; z4b7hvq{K#$h6d{5p2aAb{8GHH4D5a?23-lg>BL!Jl+a*N>L;UKk=;#C7B`DSJ;%Z_ zBK)vILl__o1uaSGZ|S6S&c%}hPl=S`v6d=HjnwPVayP+RN!(5n^P>e5T5G)Q`rYp3 zoALo`VT&_IOCS&%)!EG59If*!HvLlcQx7ar^^=UcI%`&N6*wYij{W5vc!Fb|?{g%p z#dLej*$Q>cDY#hVfLt%YG$lsd6VPpc*B)o(h;&rgu zE+({$JZ9n2eH8#jlpOg%-2fDkEI%K4RK)?$5MY!HsLCV3%!*-n-WA5?0@XtO+H~Mq zASyb8L_BO}v1$aT@Z}t!mjsRn`6Vx`X7J%-HBd){dl-L+TY?|{@J_7$M+Ba z_44xV;ql#vtMT>S?d9A1{r-`riF&$XJVKDj>e=o@7|>h|{WWL278*GXngmOrTZ6Q6CS z@88~D`t&dO^85Gqch?QOM4iwk#IJsV#YR}x1^>|B!`KL12>;&t9=|^ve|Y(eyAi)0 zeX~vhS)$;wRjD+kPhsxCo&sKz`Jg|C;5aosCAUB>gk+Cr2dPgW08~vFT~8@f@Uh0d z-tsC2B7Z#)Xd0Me-L7K&OD|r zst1u7a2|ZDn;E=!gZpwV>ro`E(_0{)CCQ4`^6z8m13%=1K1wTw#<0kU1f$h0BO6 z4C)bNMoboRq1ZQ!wVi&}R;m|3?W0!>7MLG@S^%TQPgc<5$flny0Pm?Nvp&|=f|d)g zyZ`&)qaB9b)x#xO4Df#RV-_E0P1A>fO8L!%Vm#%Yc-++mnSBE4YK#F~7S{z3R0QW`tdQHMX&~b8lD! zh7F?vxw+pz?05Iq@5XQM?%v+)4Xe^$A3p9cKO7%Eyx-q{JUr~*e0#XQy_tcbj^91r z-ZWW9g)OH!B%Rdz0Jj*P8RwoGiudBAIb{srYV7|b)6$GeLtWx|>wgukMa4YUwa8F!fojXWz_ZX2rSUgS}t zV*iRB*Cq!*vFe-`J+2n(#PRy^#~;T6d~Y~p$HVdM)!p3>739b%Ob5kE9G*mZRe=V5 zD#R~DktC+8``a|>tFd~a^yAslT|@r3GG8eQS)L#F3alZ=JYZ3IHv zSX&!DyCyGNKnvf?tHagh)zy!v#x+f!L=+UwC0sHwIJCw1_t6+2MwP^Etw@EUp*rjO zSlj70Y^oOxP`dVVigejl=`<~aSa5nM8f*b|2kgS-()&N6iU0``RfM6c!BTj73IkL+ zt#Y(#?2EJ|dg^VukWfVSlXeY{yXcK7Cn0K~V6Pf$Yf90s`a1`)VI`!oS??<4@=-*9 z#V8i8bsi+m>T=>LT0t$!$9kUrnWusdgUTir>yEAp-aBvLgwE0%GPiM18xo~ZkO!h;Z6;U`K8T1;>N;2r)pIZN7T2#>?yYn~ zcjkjp%U<*rGdJ7%lJ52w^9vC73xf68>SF$F({s)iLY4{N%665B7XU%PrIU~ak2|tW zq1CQjN7VX#tX?(dn(sQk_?C#HnPEwE`}$;2vkbp zJItWkT;=D`Zko(s;z*r@y#--_7o}=s!`~K}3*wF5cGUWLQNq}uX29daPvMxT&S-pn zeSa~XU%-ZKbmRrWoT{hcYXY2-d6b$b@DZ~=^;3-ccn@pKVy1Hfb&2v7HLLuV@Vx1g zrWy=<1)U?Wrjkk+r`r1xi}yi#9>%N|WkHEJL8XrkYDj(Hfugy(kg~a0e%t8AQ}fC( znN86J*KGx|3csAX9x0aazCTICTRBiQ_9dDUU0yz0NT$-Aqgyprf`K_-Tb;?b(m~hQ zpq9v^KU^P=NL~`w8G5~LIPf@IB4yJ(M-l_xl98AYpq18AQWGI_$|={4ES?>yN*nvq z%!n4lXUerbm#{Oq!;{t(a+$%7eQZ!GqU784)j`QO+`q=dPSZZ3E6@gY_mbEa!Y8!f zpfwPot%6`_zDm1TiWJY_1GEu+@Il0UByVs$b3wmIv?79;zJtgoJT0lx&6 zQZ>}&MHM1&6EG1;Nq3OLP}{?j7h?g_9e4MaZ-4lp5~`-PFqjb-(N&uC^{KidJl3rR zr9vb`o)r$iz+Y8l+1TI~K;8h9xC8fnJX}9qUOj%`8ECLYG!AHW+)1ik73@nDWzZXx zBoZNxTzx{4NQa#%xG0YysVhKWV+Tpa(RE+&^ATnbW3??zx_9^?tjhN= zHxcd!&svU?I@AlB(IpjuFAPPQ3mZ_ zdeLX`%OQCyBz7Ne?CNrNJVLemc)_!?Kfb+~h%kTyKqRtg0>Y$HN8}p_YLp@q%}nrY z)J2da9dYrNh(&5dpn25Ygt5j2bZeTfx`B*}f&iO0bSm$8sr^bc6cn~qa7HUTAM44c z7tfU;t`B5G!ClBNT&PpZ#ixd=TNg6V6!TMNoA6PEnlwoU3&)K?bCK}C3i?IFSE;6r z4Qs}9b@}~R1x^nRx5_bTM#PjrMX)1Cz!5+KOpsiqP>AFzKp$SW-GMl3_&{N3P($#M zFAsMgAD{zXaIvnMo>4mH@;Hi6A(9>l3->V*m}&7rJ-|t#d7%?^aCY!LeXOkwExog+ zu8jpO8vwTe;*(#v*YK8bTU?GzP)*Z&$PQBsYN*xEN&b!-P!0gw6pbdus zM_y?Z96D3^=C86p788Lb5 zYOL*E%3K98g>h9D9fK62L1_&Hcqm?lsRV|rvBv4QbXS07U$4SpnN#6ll5{!lN_xlb zZ?8ypD^;gLL}2WugD=PabnwAb!6}K}V4;he%Z0^d%;SfH!ecH*f(!l}*@v$$?j>xyNslrL6 z=^kneAmGS)&`X~6zzm!cdlR>?qh9dpY@W*kI0f%+RZR` z^oXd65#nixdrhPK^oU5l6DM)_xHtIckQB*V;}UKGbS@enLuITbe>PFVjVuef(kl3+ z{|H`SB{Hq%KtK~e#d0h*m-o9jhr1hHPHVA}XU(pa>aWx$LYh<*(gZdzgTggmOKr#{mVxuh`t;+d9fg~YUveJRtA&a};tdWDzH zw3@f9tFgAT&pOjJTR=o7I+rcy7u#>wO_wXD_(?iJ&LIk;GqUQEGJcXxU;#|-Un8|0 zLtv&c&4~zkezEzsky0MB9^@qw&QXOW@^Z|2ke4XQSsSZO;iLBsndX@iZ7!cJJ&i?Q zK4U$IO9I858da}#G2g|k(d)LDR@ zg;y0!ebe-G#MpwoTB!WuFya8l*n;6#)mXj5hBzvQQQhvr|^oD{!)n_$p%l=0}{IdLlG&@IXsW6uhd3QxCM zjwn2r*M&y5kJV-ju^cbnT(hQX()bcKWQtJ2)+4@y<#!QD+E8uC0P7LIiYA)JYtm>) zbOrllN$^;TIwqc32T4%TpCCmYpAd?3KyfEcsU8dOPA3E(OEtj(0*#M|A#l}Ly(XXE zTp8Dts@%n`A!F8jQ$^ew`B~u6ft4APYvgA^*6GH|Y&rXL{E#wHoa^bPBXVWHn<=b| ziFt-xk;8$0>nzOyKfdA50DS9iZ-jwvL>n*R&jz|xV~ta9>4nTg<-n`(nyPji-ExEt zf{jU-mjhPSk-xba@2+pHw6zig9AAHT`@bFD{I4^}qqG=+&7gLgO{bp7_-W`i3^g;>Y zNh4f>Kq+OT0_uooSwz^3{96cOQ?3~Gu0A&GN{|!=D&I^$ zHfH-G8YyoY&qU>=6L{BpDD+ssaXxp_69>DktN5C9vaGH`tE8Z8>$fqwhZoCBgfkKN z1jymPX@${nfuUxGq!}#njab$y3Z=G;|EKC!-gybr@w`~GV*`i6Sliu}Z-w;mG&lZx zGi%y)@>6iE=+~*TDuF)~R<9A%lRyOyFsJ1S`a@Z(kF~X;rI+}0C92R?ndQbch(cR9 z?;>Sz@a^)7>qhXvP!;g8wzJP#gQ%S9?zjhEAfjG0U0&9(_#`kZH;aaKRFmrRlB+_U zF(1U5zu{)V`SGzqO#v*2^Y(p}?ib~!87mRXYR$%{d{w0gxVj6Ll=LEhWmjBPrzfjO zuK8G7I|?=wIvLG?bhM%1B9H>462u$NMjHy|C|@It^*sCY$6RL=@zr(5Hv~YM2CZ?N4pA&E3lQac^F!lvCbMvd7UV#+#MLZX<+^PHGfOCdjjYEWnFjQNUc+%aB zCS<*TYa&0Fz?N8#{zBGV82Sm$yprfth>XO*sD>J?Y3ZQMf(q}eU|Ch?lWthn6A9q2 z)D_}^CHbahR|u(UJ-Y>T;UIb4@fL_#;PEXmWLE8L`g#fUqIC>N;>21k449QzYx#hj z&7P(@Fpid8zr7?hs9ZZ0gT-#V!HeCB!SZI07b?_kv+8&QOIAZ1QNY7z(V%2CNCx^? zeahM6y%PY;3psqRyr%0NaS({+8p-#z*9DeBN*8}Y2oLDH7l7ffBP^!=^58wH^9G;u zu`kn@==Syb0(_dRoH|d?-+blNKDmxOIzHCcmX`bK{k#1SR|c&gZc&oFMZVm-?KH$n z!U<{n3lgOgH7|n%EwnNSs=!cPgl7WVjhMqkL&8|w+53yw0u7Dg49W^2;vaUGch^6H z0ydd|x2j|-!i>uB6B?m|3&H}$$Fj1tak4@VmkRNbdj3UEdBQE}Hvee<#Cj$)Em6hY zdPCw;sG+X9pmkDP@cDz?a{rv&suV2{V8Tu6_@r6A=7wP1oR9cPlcqEwxsSD-e!~W0 zxXLc=j=M3vvfe%zmVUbfNx#2)xP+y>x|~ig-|QZD-yR;meSEk7_PGDCl|(U-DRq!! za)nGGa3U#b;=3)G0(JEzSG&b`9Eyf> z1sIB7!27}tfHs9nK~XtExkGwXdUYXSdU=tRf>vYye=$EgmS!OZu?-|hbAr&@vAyTu*uE+39}_do1!)E@5b<&XQ_58qZo+$?jO-5qC-K*P9c zcnG-kd3AN~P~f0a`RRxTCen=Yu|@-1h7D?E`&!-7RXUnM#7lP8XnDs*Z(NXp2jy!x zhOrEMtnHzG*0l`$Pp8%QQep|Ijs{BvE7cFR>366NEnOIw@k?#`cGXbhhVv()De8Xt zsm?Pbx*BYPfZV-7 z3=;W2E0jfZ!pDAEOQ5(Q(9?+gE4YINxuFOczncgMw<_fIsUc+<*0{RhRthLiDgT}< zsXVG@>fr%v+gyMkPyY6^$xK)q*VjcOTZfDp`j2|DNor z&;4hQ3(Ep!s4+Dj9hW&};uF;Gf;bV`zxuccgod=|V_%{v(Q)zFR!RQ&xI_ei-YtpV z#HUFOGTMCI#u98btgVw8Ce3G6L-dG0OKOnZ(E3nO#Mn=1hb3p435iebP$yJwWIC4$ zkdAUQN}j8ch=yt_hKOQgj$opydU5_Qt3lELP_*$ zsL_g+9wn)W*YhYRL5R8&5N(}sUo7ot8TOuE(iAUW&Zv7OA40|zmtMS(^#xD(cz>k^ zt^97Xp9nljfHzzLGPP3cw&UmTJsqIbA$>7{B!fom5RPW*-6=9Dy5xM;Cv|9uWR!jY z^3t(lBryA@QHMVBIFVj~*Vfp_D&}OlTS;9Mbci#4oYYESNZ3&0!SY?!i|5?Zj%Y!A zu3QnM3(gsGqLqp+XJfS?+P~Q?Yid{E>8rzq8uXOs;5?h@96jYmj+{DR)8Qd>co9*Q z{2^eFbZ{0hkWs;fHXy-8l?#Jj`&e5;aso;~$HVvg%Qv5n)XX&;b$l`0e%uqBVm&Z` zIRyF}lgPpiB1y>n#kjNz;%zn7cKW`r3Q*sV(~hXG5h~#VP%g}i8NKrvH%}wbS(3`sm65aT>U(6Tn+XghesYNI4wvFdrLo`a*=%UkW6#Ve)}a8Ezj@ zCXGHtM^!}nqW)5nV3|Rc=N(g-B29uY%prg>`>1|^+!UlXAF3yx z@O9gHiSzAimflQ)o^dxXQ98sgdWjil9I1gza;x?~%|d)~)2;SN0%Ng9DAs@)_bw<4 zW-Lcmcn&2GwFItqIMwhF);+PYdes<>63mW z(FG*ckt||uK#)cCl8$)#hB{Vb^)Vm7)pOS5_5P+33M4#gR9>Lb1hF{6NO;?1l+^=- zs#upi^de0nS*amf3y^B4tsQyG0B_Ca)b+0Q9DYq+k0g)>X#Bw%R_R&kGtzCH{+4tg zytzlOkv5T+AD4^E+w0riyWMfR-gQq~DP2xO0GgLr(J5jdE0nh}Q2o|OmqV8Jv7Xms z|Nfqim*d0k`rb$v(zUsye5v8#jtdvnI+^4lQazhBtpF^COyPq8Bx<7pZsKEYC*Rud zzI(qvBF=gs4}RP~UZM*A!OAdy`@@fSfBN+w-o9PLjn7UxTPy+LHS!SZMKYhyGITP?aH{VeW;#%!U6HG|5eM$ zKMg_Yt9gn0Lv#6t?+!eJpNbHVfff|Tk)H*H79l}F92XSYND3J^2qb`fsBz+LJ*UtQ z;Wdb!ls#o+Y+-FAJg0_}Qr%nU_?kZ;$NdLXjX6J(^5AjMUECaT6B4aC4Mx6?T`A*| z@V+<)E6bH?u;*puBdUkMBDI$xf7!GNr(zMGvS~3E1yg`fgxE=SoH0(QN4Mmm4KyB|aOSfjn&35-TqNDs3?RRb@o=b$y5nn~bL zc5dL+ayLbu7HU6nmyksc}=S~%d&)$cTLHBb~pgb}x1()@fbeOT@aJ6WT=|lBO zBz(k{uGZ!iYiekZk>hG<#yx>l>K1%Sh)oq-BMOaX3`rX(k^VcZ8;gzT!$2`cP?g<5yd{~ewW20YdQV%>sAge_4r+dYU!3bOI4#chj1BR3%ECdtmZZTfF3%kI^K(PZm zuwDff6%p`%o%`7w&tCP}rs(hcf4|Qc*geL5KX;vTo$Fjj3)$iMfd)Z4dK|ZmnxydX z3HV0iYv~r`$`UxlE&N=AD4iof>>8w05RZGfg}^GX0G<#piXRY&+> zrDJ0_NWn%!xDt6Cv`1L~NNliH7_Y%#|5ILd39SO}Y+6 z$9$~UUhESvJRnexjgN?o zL@E+qNx0NBB}ZVEpsFiJX+z|Xa4cl*FrFpM5-W9+W;TA!ECJ>8uQ9B^+cCr}AzS7* znI%Z%L+lJ`Wl|y6mcMBzx^OYCcXSQ46cked2RX~p zrDdWZx(wqB&Iqx>{YB>=9(ks+h;TTF4wp-~W?_6?J!+(O_bkffu?CVx%Ax+TDb zYtwZU25fA)t&uf@5`O|+=ik;4Pk`DaiUd(ClYd!L29&x<)A3(3sccoe@xKyOLy%nm zG)PS74@@e9jxSyiijvt7ghDZC!IK#Ftv$$+pmieL$&aVm*a4`<;R_Z3k$VV6MY>e> zB(7z3s<`MN=q0Sq4XK*pC>7lPU<*^&giG<~Cq<%64wb9IZG%`l2M_&`tVlpoV!4+b z-iXK&SijKz0aY7P_o#G5NgoUaf?z06EldofQljuH2})S2@(WdMnjMr4)-Iq!jf-f9 zf!JCJC$I2aL}p?^X%^BQbP)4gL}p^C+p={Wnrjg?AZ!FLcHXr}<_Se^xFWOsYmv+o z&@#BT>>a^>uSEcxR9jA9S@9hYW~Hc0b#S(pE$X#{`vd%ktQ2)ywhzWvri7>j`VUsH zwIJ+(%MRpWMB+#vkQplK3LXGTsE!!8fUGz;-JgG=g2(sHTf8PX5E$-|g@F!tTy+uua`hTkBtANZtiGE;0&TOI*&do9j4OVn01X}bP{Sc zWve1C{~a?$MLg1vHT}ba1v*aCf1)CAj6kKZ(KCum<_D4opxT4%N3AF>hYYFOG*k8S zRmSZb1Cne0aZKUB{y1RAe=>II|wtfuM!Rt6*h| zmsqDHanjjf^H6oJUBZ<*HcW_{&=qNwnyt8O6IDIo6{RpwdvY@fjx}o@A7&$CI&hpvbV99 zLhsbJUmbgJ_N0P(rwFMd7aOh>DlCAzgKSbBoBp6`!_ zcZoAOW&l$V1QeJW1S&z51Nh0|JH(n+iyd;cms$2JNlh>-MctOILxC)o*xy%%v;u5b zTc}+N-sWV_B6^8DQe@UJOGGUU2#bmYfB_@2Vi?R4QMdh%4T1;eavfC{Q3Cs$^fe)Y zL8(KeNsGchT!AIQLFj0cNdKZTzO_`v(9zE(`OV7sK)5`VBrr+37x*Hua{!fKuK=B9 zrn;8;Eah7xNz)zz;ULoB^DtX~=0;R{SR=Y2)OQ~8PKJ}Fs3V`fYmoM=z=8bz1B}Ca z0u`#@0fp{`Mc5JT2dDr^ez6< z!L@1jF*fW^$YBONeboR!{1lAV@Gb(vn}Di8)`{i+@#GLGTm~DFYtyywhMb@v#oLJG z-U@eA6p`W-6>wP*)bXniL2*sARYhkd_Q%+=104fSt~@={f=(pT3r7 zXG@8_5Gm0&qmdz$bs>FrZ1X#O@~pi@@qXli;uR!+~u;{uE^dgbWd$AFf5K;`!l! zj#l*zOwf4{yj8_@RIL}v_rnb$RT;FZ{tNC{!lASknhvd6TZ}{n#e0|`2=gI>4x2<( zYm0sYQ|J%SjlVM#Fm44U+xSD_m{j1DApaKu8~#v8M--$o)~4&BrQRb`n*z z|DvH~Z4lM_hw|)(Ew=Ani@|`-^tu-L(LY|Q$6pTq1=!dLIx4&yUOV`gC2NJT!v6>} z>ANJ$04T`oH7i=dz^n;4oE@WQEYP3@#=BNj#5Zfy%@Y_26eO|`G5B#oJ{N;aXwx$; zC@>T>s|9^1XAL4gC^ABYMO=}gCg-$pTx9|NRuZvN={isrtn?TNp`Re{K$KTDL?PKjm7Py09Z4|Z+@|W44miuzZNKrnM!gLhm#fT2?q}`BqCw3Qnnt0oXWLQ`u8nP~#|R4Kqg}H&!U%H};uC zO0@167N=;1L^cB@LdO-*H`q6r4OC;Xf-)3{7H~K@@ibauu8y z*LmPB96&^KRdOC$oDuA0b(5y!zXooBKlk^*Eo(!hDedW(ertLi)#Vo49a@tF?@7(? z)jBdeF6SW86PM$oS$tgX529JD^n$COKgmFGaxs*pArwfXO6$lJQn5l& zWEQIa$QJ?fjw>1$BPcW&5!4yXn8~GX(`*>rpg##DpcPeF^`*L5O2n?i1N|&yp!qmh z3ffbkCLrtskDTHiQq)=i4v&k2(#$a!cve6kSV|YOwbX6dI|O?aKy-Cs6M@5BB7~YC z zVHuH4MH#cO^&n{kg>4a{(|b=r9q9`&$weV_B)|9cm>(F_x9O zEn7zcN{7AJ9jnOlTM;?XKL9}p%?Nji+aom@EN@z2n&OAR;BYvAqr@sjWOwnytagZ% zjjdSb>x(qOV3CI$+La<`?n8!0gh&&JD?($!D4+p=N_HgWiWmaL>(mtu;cY;69d%VG zU~W)FLjw*gGEA)SCk2-lKS_-$^4VG;Oz42vJ}+EOLPcDIW>x*7Rq{7wA90-NpP^>| zL&0o=E)*`82+ABJH-T^!sy~WG2(2_?G-#++JPtTZ6owiIJY5L(83Y3vxYCf;sIPUU zh5GQ;nnAFSj$PMsXi=IA0VV9pV8Jt(RZBuR8_7Y)IAhdqwd=&UuKr+HinFf1B?e+~ zz)}=gLgX815(8OiO07x)0&mvDKo;>)w`sQh&uuFm=Ye zSPDH@?I>|!FZs<~*IGgZCxQK_H+P+1k(s}QSoAQPDl%(7d8#8WsE5YCCioTz|H=P3 z3xrvMEVbXt*1=ZD?$Ao@$3oAdfyhto2lof|ET@mC>)H{x0nq$i<~?#-#a=GZJ^}^? z2?%=!2N47eVMRd&TVNl9kce8kcnF|_!RTPcf=Mmg0om8keWGi^8mgy=`4bfg2}WF- zt^+b)AF~M|FbB}8Q0IA24jC5`Xm6o33N&-!O4z~rf-oe+?*Ujrr+{nHbp6*X4|rgI z4-c_2Lbfd?Q-y}4DbD!kOckk>RE&!e$O8Ir zWsBO3mO}Ol;0I)D*{jQT@Ng1JOu@}=gJ>l#%eJN=*3oM!qeXk=oow|pjfekNMdVcFR}u5B6PBm_x?={hqF%9%86Kw z79h`|b{3&`e@;1(cX_Ia8a57khNGD{A`}2HHcH3IDp17AWT={c(=PhK9?B|;1IQF& zt%4HneBD%583Q_foR|S$6o-<>K(yo9G~>X!`P~s7-h`;E8I)ob3LEmR8KT0%)k5eB zQW;u=Ojb)r-IlFG;pr67lsH)HsZ&I`Q?~G{ReDQ2og$){LjR0w%ieJy3op>umn*DB zHHxt71rrG=ksenkTwltCWDHq^4lQv^JDA}nhA=9iR#-!r9%9I<+y3_k!sB1FzM`}) zr_D*kQqY-jN0JFZ8DU^^aBaGd!w?T2*6RbVFg(grM{9`oXW_!hC>VR(I$Hed<$>b7hh%6BmcA04#F_~cQ9q)A~^C7^+e zoTWBX>2*YYu}QS^JAVl|LKyGwAx9Dim87G7hS#qORcoW&l+T9dJe z)*Vk`JVyJfl2raVYpvLLbA()NR>%n1T9&Qbw>Z zz>S5_0x}AbA0==o6V8qMTLn4tK``j|S=cafFB0hdb0C-@rnTQn=Bv47wxuGlDf2vEw2AV+A@kQ*$6g9$Jx9 zDG)Y3O^-VI#cX3$lV)~)U5TJP z1!NKbW}R`&R90B$2FJ51pAzuYZ#+i%|A|Lr=sSb|3^zw0E|sWu6l5jOO__gFnt`nqM)|TB*9wpiZQ3pO}h%v;iA)g3kBiK{C zMZ=<|>NZ`w&Rx~|lK7>zVqdwJk3WdZfG>qyKk)n^83kq<3-NFiaS61WK^0(Wci`;+ z<|T4G3zotOVQtwu1k@OXPPmn|7%W{|Yr*9Y$BwFvazGnsYOwGSH)J6<>;plC9Qo}-cjY$QHs94XWmC~g7|kcnI( znh2MY!j^}cKIw^Ll-Tvy zE?isoj)XB|edR%*^24b>4iyXQ3Ie47LuvV_)G8rP0nZhwV{j_KMuWDWTF4StxvjcM zGmsxR6}o9qp)T!TG^H5BtecrwKXCC^wa~NP=wMchJ+Y6lY=kHtsFP)3iBp+#jrf)n zTyoGUA>YE}jxzWZ^-a?Ih3*3wmFnrZl6C>K+*Rt^{Ds8j?y1@|2i9+Z4@e~->oz77MCR%$4}fSj zE_$%6aheKCf?!AEM#L|pWFf3W2)?Rw@qizYyJ2UEy=-Wno?#OKJ`&IwU@>}51Y-Z7 zXW7|MO;`ljcz*KG7J`r*s7guf?U;-jSDVXE}0RyPl0?DTBO!gH4Q_y+^9>oHM zAY&lu(-!6xY)5rd_71}unRz}6nU}9Oap$bTokO~g&_M9yQG*34`<&J}EGR(jb`p}e zHzSo^lwu=j2-8C>G*HdBw*SGA@bFr$v+z2gp-o-KIR|Zf6eR?DNVpX+L#{1b2g2dC zUNY|yD!dj?S6C#RD&Ii-733fYfs?!^EGvWqs0;$i9wCqhOB!N}6t_h_8P}v4#;=)H zc7KuakMb0pU8kRE^+QQ_9nUiUnM^p3)K)5Zo%6sFK247Avi~`noNBds%4ObiSG9-$ ztvErs5X${Vay7PuI;IIa3VFN8{icd$U0p^OFFE~wxF{v7Zs@HZGePTi(!-wnb}vgWOWCjzk>p;7P) z!m?PoTBk@WZ=INNAZBNahEdj?z_CtT#4awumV!)jin(Mh)wnbKc zGZAv~74o4XKX*|u0#$A@k_8h51S(~Lau0<(K;-VH)ayyGFq;huJyC8b!fa3=K=KJn z%Bq=Y>bCz6p9IgE&{E&TBA}X-oKOL5*FU7b$tWOTo`_(D*@~nhYeDZH*_v3!WGs*o zMCzN+Z>EZrr0J1$TQc=cb(^lUX~?#eSczQ~2%Q5v@)cN%(33zw5fnS37MPLxCc^Jr z#FuCVI8-qXDtSoio9ZS_-+#>xvi*yR1623e0L_IT%A5-n$Ei?+7`Ia*-G z_*8YRtVbR%+MujQgU0b+BB&+KazTYFK2@EWJ?h*6y23On$bTw#0Pe7Z*qszt$dzEy zP&R`18hH}1={O2Evwx_vyd{`Ma3vrd1ppBi7J>j$xZYBAn`RH;iT^5TLL2qq&4;?$ zv=00ap0prK53qGW5XPRaXkIc0N}&?vo@@K_LjiUmv`ctHv9_hMvnUJ4%?fWQ)*x2f zsN1r2D1=3+Yxpoz4jQ0BHY>k|4-s_0uvTSr@N4)m8(rP@KQ|8kO;7Fw5D=Cn#4q6k zKLVB@*Q6Q553U*RtTK!q#hf31uK*Y5rx7qHjNu@51f4F#*N8m`EZPF4W7Iwn73+g` z7iyeRCLw-8!Uxx;8N|=`xgqyFsZ}c7tSs#XV`3DjL_!&A-vH78tr9T_wRAqSQ4rg+ zgl}PE2Aq>rwmC9|u6;M?&fpwrH)W8Ug0NWv$q7Fa914z96j^|yjFFI00$~Y`ZOihs z05~J1gE`TNXR4YsF!EBVk3 z->@_Yy3w)s|A~oVg~2el)|2PL2KT-yZqsa&pC46*ZFlHZg7!fvOTfW(DUkubDRi+pItcO6!LmTq0KW-M7F^i~;j2t8 zs=}gf({&7nEg;G&#}#X(r}!zzx40~9!O&(YPCO1Lk&WO6aBaW2_gP&H1W`Dr5UFPR zIfcYCf}{~{4JuE?VryLyM>F3W9{;?4J-Q1hV z0s~2uT?u*efJQ(v0Q@ zmj*YCW`?aKDfki{?gP@J8_BHk^cNAG!yjO!q(NU)fZ6%pHD6msd}WF5i_(a&EJ zxI})2x4|p0g2pMJHq3!Y(BZHF5RHO$5-`>{-a+CYZ{N>63(7OClA;ne9eiTa)ZWlqpC)m$Bh{7l` zZxINR?79~rsN$iN`FpZ}iilJ!%p&4SUZqxN4=x7G<4imLY1Hfdsz!dWDT}lSDw}GlOOSC7G6sqaAw#agW znmu@Dpxbe6y7u0n(}Zec1^|7LNN|$aickc=UQpG<8rhkU4Z&h$f*&wUQSB2)3zf^m zTwq}$hZCyF4onOLS;Smh_Kt%ygQ&JqfUANTLe{oI++_4Di}W+nPl4e`afq;V3bcVh z16c}NKqeI6q#7@*P1k|wxSX^cd_=(E3L;HbPl_C62ok}aGe`MWHKhnbLy&9Jwfj1+ zCsm3gMWNl~a(`D(x#k;zErAFYLUL3~il>zva!v9dwz@i~t2d20RW;-=#I{s@0s6-`P@i*kkD3Ypw z)DhRz@6=m`7lw070?QKT7P4G`pXv~GgeeNZ1(zq5mzF8YZVx2TAm3S%!Co&X-(5W$@;!NUj4&xAYqw00XRWY5G2kt<~wtzGBCO1u9& z-&vT>%fG|S=Av;0sQ`!gx75M%zpOz0%)-3+8LFIF@weP%D!%y9`qH2m@8&PBhm>wWi z=^shm{04s^!-fnG3#kjp?1HlZrjrHi3#48%Jv&|wC;9d8C@gJYk-(uKc~!kqICY!? ztZ2X}fJ2m>Q1gP135WUKS7hf zZ?m(gkJf`{_BIS;eN`d&S(#eh*DEViN6=BCl4kfjHlHeNtQX_n;fEUgI_?qPW&KSk zQ3>!$$lBm34FZ*8F;fEQxU^I3d_ZCFG|b?A!!EHy`Zn4$8wES!Cjkc?yZ@h+1~!%* zpmja{3c?m~AVCj_1rD`b3*^wM+p@R&AcxTX>|h%#bjR^!If(d%$dyW_lD|=qMNdL- z`pHxpWa|O56(e|T4Sfy3*7#~Ch+LyMA97|WVH`AAiZbJr1HM7!1}TpXtPLbAOA&A& z!W;yvxi(!V;SY_1cV}}QRf}CnfYmtQ{(}Y=;CEv70dmR~u?ww@fC&fH$lfMcXdCkv z5@C9jg^UC`IMw{yj);DoQ~}y_odUzQG01C(c01Sw3oRgi?}-fxVRL%~+O#`Qg{hHC z#F{jH{{ceqOLsMC`;l6_y`Jn5gd2!QO*tys;09$nSgD$_xAPE;L|!4(r-X8@_z#}2 zJF4mpwH+32<{L>!gjKg`dY?U~KRtSc3-mOT2-ic^A>^$yLyL! zf3RtFH1Vud{J6hj;;Cym>1X0m8tCtdvkIPHG z2x&pKwj!H>LPZCFCF6RrMh+kO1x$7oD~i*GM@41`2I25#$)ikmRo$l9``K}8sASC? z^u6?$3@K_TsziKwtBA>vs@t-+`8y~%GtKBK<&AFrkp|3u|M}WUyATrx6f(DZ|qf~5Z zD^~cr%KeFCr3mm5dH4~RiUHW zt_0Xe7NJ_A$-}K~cer0xzSC2Uz5=r_`knnUxOJAl9bqqSo=^ z(1QAqH%z*McKQZ~Slk(v8f|ffpr{njer2Y)x-EOVPfFBCg0|KdsYD?PU}7p|UjC(R zkTS2K4<8l>bR2B-^x|0f7TA(XY@YY<$ug(l32V#VhY#8TN%>3lg`BvcEq)K8_}Y*Y z6)44h#&l_(r0A=)NR?^eFk5ET&3sYVfL!JG`d#znZc@V z`_0|wS*%ueVnXQ=x)Id$0k%rYh#-9u^NZCS1!7F7DkQDoDM+m$rUuKF6&+QzXuAFb zM?p8671cpFv+-}>eaK%h#Af}?Qvd9JfIn%kCos#mRr&XWLAI)4C#)@9ZxEi*SI?5! zUVztZQJq&ippu=X(zR}91db8X1^-LWr_V}65o3=9)cH@J86N63%|`jT#jNAg=k0MQ zQP@%N8v$COfHbtIP|)5Ai6X>GXM8sssxL(GbgD%L;Ws<*jwuRClK<+q-`snxhtDm2 z7^lHP={x**7UQ(RyF=cTcJ?H*EmTdK&i~+y;_O$0Nk0gq!7$QyEoWN;wOn`_f_|FQNFM5KQ>cP)P)V8W~Y$TD&p3ZN6#AWhZZpf6YN?mMEw~#Si)5E5$i&)Mc2p(O2jN zAgPHf-z>P|^t&dsyAI15`65Da98V*c@x`ogBO@(>YUX7Hju~G}-Ik4AXUa+*lr17C zod$$}FP%nu(MVw;4O1ipAbw4C&WJ&VZ4c2It}Pq;PH;U+pa?6(^H&UGSD`2mJSL|4 znDwzUc9ptK*RBifTQq-DGxhVA`*ND37D%2ILw+>KhpI|C4Ho&e=dzszB}5vaE=BtZozRzK)g^{Kvnh zpJrumNOeY>=0Bil8MaTLdPpt^Fb^SL$%@`K9Yn zGlp-WT`tngB+C)%3!u23HOiq_=@Pcp(!e0D88nu#_bE{eCA+fbZCYe@v&E17?!T&}Yndb(^m1$Dl!w zpp}op*PY4eLX^lxP?{OV(KYdO3Twg51%w6#7c3;ww4fj(DPEC*0`YmSP1ivfwuh0u zgNs?}peHm)$0!g{D#(g;&&weN$DJfBSd*sjKe!TfHK6{Y04;%s20&9k6wEhlp189X zdVBgBP&T+55VSy0lO?ibWtINXm=`cyocsS*6f3msKVWGTxvf;RAP5i!N7$}fG*O5!4H%2oL=h+rcy3^c2$ir=$jWlC31MX+29O*aquK;k40 zRExk3B)K>c7$v$GxX7=9%oK)*qa08IsVJyz5SWRS{fL#uB!vv3;rwq5lIM_c{zt%UE9bO=9-DF5m<%}CJYpMY2blchNxe#1z3Q^NID3SB)a|6v7mi_)qvK8UGO zE+vIgSevdRG4Pfo{t;PJhR~;QqESI}ZRQ}OPpR9ox%)(3Lk5+do|inMNI``Pe(ZlP zd9o$J;pEz~xBC#J!YyfIYp17Eg7gu}M}-0)B(}6JNeDoxn>3yOHDeM1qQ8g6*&0Op z|J7;+S=9F2T|NE0y%8`OELX~e1Vky@A7p2h7(Hre00=@M0TyIqewPjK6Og5m2LP)8 zq#CYGGgH6mxXRLhAK5VfARqAey=7v#x7^=Tco3{44p1H&oUTY%04D?IEd&ghbTwTb z;5Ks!3xJeh5^!z5X%KhlCns&Jn~Axi($9FK`!TnHM*23z@4An-y0@AWI{#aCK+FLdS)s6_TddexyGtMf zKK*7qWZi{0YcYN7u_;f; zMh!G-L10uQ3pHLjfKu1q3!GQq-wjNm2uZ58u($zK0q7AL1s~3ay+~zO)hHy-wa30e z&`VDg(x6d5xmfHg_wtd0^kNJAS>Tw3o(u9qA;}8#(VA*u;fPYPXXqRPmj*=~kj0LI z(g0^wO`2)=HS1COH|h6ML+JOom3}dp`=9gIg9Mm`1UATV(y*Mg$oD*!lR-eFK`Sk5 z@obsQlZ0gj78}2KHb4K-T1weElqgRNk#X=HK(@8T`KDQJ=;yC`?5GGdILOyc<|X%* zxmc)-VTGUAO)2y6^Ys-A4S{Mali(YeWJuG5Ap!+UN+bYYf-Qm~Oc(lPrLEsA8}}46_x6tH7{$ z00d;d!CeIT4;lpU;DK}!Kna#VarY4+qSJ!4X~y+S6O(mZ->>li>rRA}QxVK9-vDy&?n zKQmx0nOq*yR9Ty@N0nhqhq0KfC=*r4 zn&$I$S+Xq90=O>3FyVEd6DqlAPuc+WQS0FuHIZ32q zMAH1+MfiVzIMx1gSB1Mm?k1A^y7@!xE0DR_DD(lwwKuMu06Fn)L_-yUo}xfcxyVHk z=!&=YmmlF1(B(c zs6m#ojz<<{b!z&|U_v3U7`x0KOfZxofaD+mFftJlIzlcjK}V1qQOOb7!t)1E#kFac z9iH`1@Y!tZHgBtNU84je;!`8EVF8{ySUX@`Ag+%Bhk%k;o37m!SU?)vE(^25Ln)X0 z62i+46JbchUP(>EX7=Ul3W>7#R-j`bcn1%r+xT>T!tIVjY7owxv`#1GF zDS~N;+nYuFyp_T`P_8r+xAF$*qXpa{H@F-fs94; zj5#O$bSKGl`t$r}o*i?r4(`%vW}&(73$5KbZnasL&J%Ntv1#9|pz*3Pjry#TnvUwN zn6N<{zFO6+bJX4{xf0p zp_XSFyYEOE6EdUf(K_MInT@)iS#x00yK!UoEH}M*LF`u3<6!x-E|U19j84}BFTC+7 z5NsM}`SIF@T@4a?eR>u5vTx?wd--!&ot$k_rTUXnlkb&UDlL5H_8w{6lc}#SJbsZ; zVs60=i6sUa-{_pL;6R@%C3idQjf(b?Rj_h&?l7!FN^qX2@J|j^7p|#Vtx&@vXG^{J z-#)vKWOx3l#is9n(!r_Ao#ku#tSLIiXXUtn;~i)JyZFedYqEQ}7K~bwK4qoVj;JlC zd+a(sx8Ku-r|TXIDc$1ii?Pn$XEt^o_u}2F?Gi&y9j6pIs+*ZB+isqO~_`%(H6VaB=19=^Fdw@qC3 z?)k({)^EqB#?9;RbNAt{C9ZqxWo&kt7MeS`Pt6&|uxC^Y0pF2$KDYicidgQ;U&ce1E_Im!a$^!Lofj-6X|bM9i^N#O&B z*-7TxLFDM??Pex!>5J74@{^gkr^2@xU<36FbR4%*C<6kUd&-nZ(c?PaLLT6%RQLz_ zDa{bV-Kzw8^ptsZO z_#E|Lpm)*lAfCeg2HrvMR=tybPwnh|c-;Z0F#9xoZ;8)QkB9nY$$Cfcrk~VbX?*lO z{mVKw`W_Azy^haOzk|Ml@WVbq{hRb@)Squ-XTfF^?fmxFRo`Xl&wtvu5#u+d?6>K& z5aqJ$pLKQ75LAm@v$|S#bnN8ZA?xbuCie(x8z}SswH2lUG&L*C&PHg3sgs1*8vNc0 zqfyhy=qt_A?^>cVQg?CeL={5B~{JyPvu;0FF z_2%2_>~YiI<6fb^q52M=_?`cK_j4fA#aoLsdH-39-MY8!*{!SgTKuJTh%g7UFutur z1jPi_p_SlCN0XvniENMSyB4*kEbH`X)*`Bk1Czu*_Q2UP!r#fV9Puf1o~X|=~`h8m1GO>C%oo7!%+REceLlm#jk!|?a|BAJ$K6a0_+Mr4nqND7R00o)VFYyub% zDlfw#Uy#yOP8I^Jsa~C3yyUKd^zAS!IGgzM^oiX#+p4?58@7!4S9>`H%KZ_a z1rbHun|)7R)L8Ons*h;lXKZHtJ;P2-_qVeCW?iwrrap6i@C1J9PGD!CyBLi4+sNn< z(u-M+lvaL9A9d_QHDpb{heet{G45XbiKGjV94jZwy*Agaqjq(VmqHHnvbD5Z?RL)j zd&WMl`$W>+GKZr{fw__%Z7w{$+}Wl}8*z{LZ6HQjIs2b^W^Ck`W6U#?knfC~b&^JD z=b#xm^-(o)9n4&oq8R{4GjVg7e;c{N!!wZkkql@cD`q0l zTQE9rnMZ&bY#K6D=mF3+5m~{?VeLSsB=l?a9PkD_3pvK8+E?@nhzoMIlzwpMkp-r3 zZRU&m3>b$Pd%&j+h??AYJ=k}&S?D9^<22TN)`4ZYWq&>!Dqlq7DED_h=bO;9(4kl6 z9~V~2)&9$_9LJl^8W8e4chHxdmnSbNyI$TbxJc%gH~wRdURQa$u&TrHThiUb>xVd; zU$^8|dN1+k+ICKz-tRj5;oSjut5pq`C$5chum3b5!7gL@#p9Q~z23eJJ$&Y~<%0H6 z35((`rX`Q)o0RzQePFq#POejLg$%6KyYu^~;i(ycZqHK>)C+r3?|s~^xKO{Eb#~vq zA^U6_dD44V>RZce&(~n5d-yPM0V}e~m9Pdb`hm@I^+rMh3 znCUlj4n0xPzxwo2GtO6@wfxYdw60cPYNgEBJZ*i)%^lZIdot}o$0zL`Og>TUrc+cz zVz08|m__!h=8r0#HtgMqr>_@X$$e#0{%uu*T0Nh7wemmrON@Nb^nR|8DOXMYIXe5s zSi6~a>qC__o<5pd%sNdp(s$PJ%(nX%Ce2D(e|kfu#MVzno;sY?wo!o*xiShQ*Rrf- zcKV7((dCV#!#*rp)}qnik3*kV&UY%`lc;q)FTR);(R0l1W`j4}Y8K$KeD$bSAIr_U zIIK&?)R89|je4IPcg3#g{aitLpBGB0^}PS1dTl)(?jIRBa%G}X+_77EqRM;}Kfm8P zLlpS3sMP4%senTEr|(&{e&1+ErbX=2m}`AceV7{iDrT_5s*_XaFY6UoWkK!*Pp9@s z%A3*QWv%zQ-%aS3%R<)Aq-bQLK_8pm%Q4_|<07LPJ|6I4`;3;$j|J3Qc=vMOf_?>u zdJ@L3j z$?jL}6MIRgH0t`LO3C(R{ib|+y=>r(-4?;4);w=@+M-F^yUgpoEedW-NfNhzJ?+%@ zjtTX>2>QR~oKRmMxT}7;an!}3xto(r5ZtJ$dlbexq>t^?Kf)@vk!HOKBKOO?o3^_y z?XZ2&(f*E)-@Iuyu5Yr@x-qBL3>jL~qRPI5M&Jb$^gTM(sPpK(0Un978@vw*Yusy! zd(!NvS68<>?H)b3!{>@~)=%0}tjpQ!dC#9Xb^dJ8E=Ij8I5-@!-ZjgkTb{)ZInAou z6)!cfe17jd>C1amuUmM{%SGtCyYGb zjVw~Qx%rA3O^x01?2L*E${RoVN=88GiCt`K#ul}xUN@%iio)?Zcdm}ko7rjp+%4nE zC2!lk{oVF}9G-DU2Q>&SUD~P5oEJ$o-CH}kS1?S>eqZTa9Qo!+KRbiVeZ zO-G-adGg?n^r3r1KHmMk8)in>=DD>neOdWIkK9hL^v%6tw{Do~Y%Ji@_YB0KA z&ZORps=P4To|&2X#jgLvd>y=fHs4ux(7b$~+&Sw#3OCD_5Z&<1g%fFaYS(msWt=?7 z1q1BfQAhg!}R3-JX67o0__O+2WY6 zrEy=kEpm$qpDCSqdtvX4T9qF`@-<=eSFbL6t#a1yHO71N**()& z+z-kfbK=y=kN2l}C7QXmudrfOwG$V^cIWizkXdei%gd*3UN>@c%{%DXqr7G7+m}61 zWrw48?1htgk_Y8(I=yK8hixTt43h2Hk*9fhj(;O#2IPqC-7@X)Ytw&oly2RidF(RB z=V4Z+IZDUoFmL|KV{*Y5m&a~CDGB9g-u<#9G3WN1IV(=BGBWl2+g!a0ZK<-k+LMrs zIOR^q5#{UW@Yy_i?Zr_Jw=&~0&sl|CiOf;jwaqQd5O=e;aaAtlXnZ>HNv64N?8g~q zCJxhkgilYrSg6I+2JuHfE_#rA`_0=8x0RahTzdPtfCg^)dc`CbiD@#T={r}iS5J~V zG`;Lh@al{cZ+jc=7L{h-PG*gUhma;G`vPfC*Zy_IP-G3k@d z^%Y?|@0g5fIfPJ=SwpSSu5e@x}Zr zTfQyVY*+1!X~U}YnM;jl(b#nq8k3gcr`z-sy|p~O^*!F|i4QIPI=Wcp7E>;u(MoA) zv&U%c*N4LzPmlf_Hey-Ps>fb@+A!2+PtdSEc`8@Wd~fw&+zGo8t!Ad43ae(@V`7du zO&<8IO3xXVkXkhVCE3S(y;U3 z*iRpD=j-U1O~a?3GHWw$kK}W;HDTA<#%_^6%b(W#dQvxqi>G_uKufE%YNF_K^A67u zJ(-#5ns6)h#^&f@ZIZ(5C%hi|I<51$u%cr(pUSU{%lY*|XirJ((`1hiu3t7u%2lYJ z-qq-4Xqob#zCOyVTQ|%(WKQ%w=dTaLmM;#wcJqFXO3!noA0HTfG1uzZj_qb0Up_D5 z_1;-6znVLR%se?|@o2LW>D8R0<3ul$zr0K=2}RPk*@wM;8+P*Zs8?e$ zmy9#Ync?T%Db_#X>&vu+0(-u;zH$F^`m@*g_|8i+2d5@|iG5M|_}90lmch!=ch)_K z9uQ>GAU)rbv1We7&EiXWP7N&MZ#*yLQ0w$nmUY~lJy>U2vg4|EMI$ZuMMze5I-93_ zh3e9AQ7*tsl$J5b@$tRXryl^Fw`u3PU?BzVu}*CFN^9qtV{;d8%E9`jITd=Yc6(7qe*Y|XO_lF`v? zs%N^inc_C!sQrw|=8cAY&bfbIT60l*%SDHqhpwvsA)h4A{ubqy)mb?{e67iV#z&{^ zs%Z4Q|M`!OBi9@bc)UTpta84iMxSIi7M82G0SE@+H zN3Zkc25oE;oNX=8T$mdq&^PLzN?+ROlaCuU|Eqg@hQ&k_&5+M1{^?}Fl&5vZ20m!z z-|1Sc@7|W}Z~%B>ksK&`_TiwQ+m`l=Q(h+r6`gajnEz;b)6M4tJZn3wj~ms% zxaWxXj?Uwsua8wUez*}-D@^=TW8=oCe z&?#3Hx0}Yhrk#&#C5d@`a?-F)Mxh1kbSUH5-d;3o_0sLZhqirsI(=2rl|r%I``@)V zQ2brrN)9EXmV`din(ja*7&}D zz3)@a=*;qcQp&G*`y`_26U%$6{CC_Klg~3_(8sM~M=crVT4UeZT}_uSZr=L+qsaM* ztyBKR^zI}@&0hDuzNz!P z^ubXlf}7@f@uAkC>j7yC?D9}mnmX;h`L_6O+jmv(eroIYNtCB@ zx>x1q9mk(MG-`8`=zZ-k-;?)kU2NOo?vaU&PY)hrJnY(pluAt$Pa32jZ}$4Qhuh|+ z`!>%XJs~zkS})jdwpm28#U9uk7$06Znd|*8{htJ$78S3-ADS} z@1DQ;fF4doDaeT|=ciMIDig;|%7BTh4&ENAVK{WhPY=hW|3g@J&vZM~R z=2mW8rip(`k9>*lQybR~NndgE{eY?~8@}H7I{k3-jkP=PknNj(Vsg)SOS}*FJUK3Y z_pYe*0e%C-WvdtQmrXCbXlMBrV|RQ`Dwr_N=J~mm>8sy=8nowClZacrSDFkgd}QFJ ze@>azI^}ZX(aMRL7RT#Ni64=pRp;jwUpXwAk#VYpMUvOX!VyQLmi5z=B^o3zJ~PeP ztr+B&p7J*i5M2M;G5@(jDXs%PNd z3Cl%W57$nZ_onf3-y18NG>_OA6IafoQ@75YOqV5YU1D})b?%mTZMIyJ88#2#6;QZm!D|@cJ zoHoS#%AGf#?mj&>Vb6&|U7PqdbY8kQWof@gJ36KYNE#I@Gw{UuZlRaDcYALCeCxnb zL7O&D=~!;|>bTugQzEU)6o?-2c4v>GX;u8pR)6eg<#%zPywbXiMz2DYV^2wQ&f0l) zO^O&)gf@-pdHI_yi&c4HzU{jM4u@N|bsur6-K>xfFXl09^&pbA6;E|XW)?T##hU~Jg}wqtbXxz>Ua6#?{;kVlYslJ zTYo4~_fr1R;`K96tl0R(txd;as~UX{xcl$qcMbYQnf6)oWm3-zZvqpxm%s5Q!tVX8 z{*~Teu+FeP?cvqhC&%)Gg+)!o+pWUuwXg__-DMhny+;v?Z38QLdG0=IdVN>E8Y`}~ zOq=lD^W~cpv*%n|8MU^%-Ek{LIY-uLm!VNbzYFUFiNn z|9WOeWM%iYj)^qBD61oVH_7M7`FVXBMm%&6kuB`|tdxnHyrI{&kc)S&rC;cO&unD_ ziD%7SABy*H`O4d~bX%ttb4w=_4IZ55#^h?s~t00(cZY+ zz?QZ*!+Or`vt?jRkvWZqoa`Sn>dS_Q^CHf)YOGY2x7%u~%s=FPCFR)j)!T2V*=1eF z`d!xF>@~8|(ExLY)8V5f!Cuu(wp%JQukC7lXugH(#-cBuPx-JXbmfi|Z)InX_|{G4 zFY0|XPr*x~8o6qPU-$1l)41KC`@LL?m?!vnSbaRZB-iWo12>GKU25MiKEpg$>vj*7 z(&sOpr#jv}-zxagtya5=>GL^N zG7qdV&GWwESfgsiX5L;`ap{^iiM}1*uAO|V=0x+HwpCJIZL8|N&rP!T#+Km@M-uCW zJ)1Dn|46xO2Wv-7INojQ*{~V8XLcI9azdR2Z*5;SE|luxa_eN0e>JybcPv`Jt5mmh zu6ka{6z%FG$bdB_PocU}>W7`4|_j}yjWVPsV zE#GTh!pF=t>Du@H;FyOod7bVbv#B9{Hu>}68@-dB_O29WRm&-6{@c})JU?8p{t)j} z-Rn%y{Kd1KucyzQ?Y<_;yK6-K0)axEq82HsWSa`eyrv>`i)!Y z^Prf8=NmNYw|&ir5>qS0l^;0nQ`h7SFR%N9B8PXbe7DiPzGi3duAOHc)N*g>d;fIr zSts}Gk&9bf430H!_VR#}<4~nl(o4tvAM?#zSU=-Pxl?Oe^sm<1>Tz0?&0P;iDNK6a z_8$7QNVwyA(>%9tJQ#lekyrI0(?73ITXp0_-+*C_`^43kUaPnxzTy0Rd%f=3Y|SWG zu~C$1C27kWM_$f(-_yTgm%?KXA33X-cgnO!;<6-<)X0BLZeKfad}r5biFqePyH48v zIDE;1F7njqrQ1yZSy}QxrRl3vdnKP*-sf(Pud7^_^&FS8N#m5v0z*TF9Dg$@U!ew- zTFtKT*tGDt0_{>8JbscqbJN7Yz<2vI+TGhw@owUt?Hf<5DkOU3m#@*ymiBXF8amCo zbxb*S^@Zfqs~5TTkBWFb`CLTeqBf@kZC3UXt+{zonRNVN`{I2*x183l?!XRZQ#+WS z+T1Q;{|mQ96JJlQ)cwQG_vQR1SlAiQ9yC6%>#TkCN-S}VDt5oL@#C?VJFclewdVQ3 z^M_ySIM$@o;)Y>m>vodOdC(&C;O;S3raVlS+B}E~DAl^9Q#W=TnFF zELCgd(7^>FqJn26CTy(MaOZ+!lk?`g(=E5rmdQ4wf@TENo4x0e{loWh6Z4s0uRZKj zC!aS*236ZqAz!IGmC76x9slS067w6CpPkQbxolzD6kFp%>q`D(o4TNHa!7FUxWjSc z2ey}5ZE$YJR%JX`S5nu=LxHinZ7UD|$GiM;uSM;_`hkiHm2Xm=4(q%?d!h2 zd10?P%?dAhwQ_M%TO4@ADLvSW18+Vbnu67%jkDM@ZIyS<`U^7*!bGcWb*ckRH$ z3TsORwJdw5d9gm5EtVOV|8T8P!`Q|(*NmLy{qN_e=_T!*r`?^LwqaOejdzWktu=Yo zE`DN;D|%c$U(7h^T@U%5KKa%)H+^5Tu)>qgV-w5vExe#Gxc@SJUp&q(I) z{Q2uH-_9Rx(>|m8i3&FP(_@{24!C5lzr5&D|AEWTUt8aF`Lky=$8LNboin=R)mwM% z=fsbW?fmgrh@JhexvBCi&&-m_I?r0zyjTsZ=M`$rZhUt_*O1D$zFaD?W!Kv>GOx6h ztIe-yFJ05$G7kq5Njcizms;RT8U{F?yoj_)zvtpW7A&Ix4z8l8xb{9;X@{H|~PWLSjUg*`MbH{+%yORp9Fd0xndU#@7 z+y$GJ!~2`;IMcPMTfM|~1IvuQchI@iqogS{8;$FFb4`0s(Ids@OqWTij%Jm_M-G+r zdr@!wK9}nKAFkY6^l8PVJ6Cw8u8ch}zQ(3`tuFX0m%J9Nbvk7W0i@sxdy7+{x_AkWu z7xjH;zUtuW&5IM(E=u__KIKDb-P0>t`S83$-g1rN3)ULjVU6e4sHI7i)NoMyXN`q8yV^5M-)-hBitXI zZ9A-J#jD%yb-BKBPW2rNGcEhAP)@kpBcV%pap&cun}3n-xL0w5(OM6WR|lV!E4#YW z=N0J%QieFB)?4>ErGRnUVw+!QgqHJ9aBY9&!0Nya5q(R4X*%KdF8f6tt|shAzh|6h zWY=4jV?GW}9$R?vK@5;jpvl@C$IwQVNrtp$o13Nq~Sgcc%D*LJy_bhp! zUX?qmYR{HDRTiuke&zb|Da#f+jdI&@)1rB5J(E2(-^||~Y18#l&s6{D4-4=0sM@sV z)^gFVm-`)4JdBCmnnVowcnAqLW5?;bry+Rw@hteT6fM!n;@UI zD^FPEu9mlKC#R|-_aAKeMD+Yxvk5P@_g;IU?0)AmeWx97)1yLCy{1*zL8oO1UXuccYTGr0!057nCz-O_7-!YaX4Ivi@$=AzxAH z3Fn6;b?iNF{i@d&ERNkcFs7yWUdQ=;cU4*FYp!rBT-m;8u44W7_j8`vdhGqNNwXuD zPw&|+>O<{k!D~~KrzKsTV%z#fa=fD6uudNn4u3p0>g@Uz-CopxmK2;a^v=6byMWxr z=cag%h>G+oUcq%pM9JMDex`372Zmc+>RRf_+Xf#Z(%!z@XA!5cnl!OR_wn5y0Ogrc zvge~&`;X?H8<~H9)iUo(M>HBX=H}W`sl&TTyT-X!Zk)R8k?Wo0XX_G&ugv%2 z$SJ=Xmj<^jx}jQ!0`pgd6k1kg--)6TGv-CKsrImT`PBTU&u^Q%_;mSB)=mxTMmM)>8-|IP^cvM_1S^4c4c+6J9Dv!O_I$sKW*h9o^69%>#x>1=~7^@dfC zYq`Du;RcUZUY^u&sVH{S>EdbqKNT_Q_sr>D=lM%7ej0ya_0@zI-P=8y(r47|u>03H zJju6hM%t<;qsI1}YBZvF!I`TzwOD&8es@1*-<~Tk1x~a(F<-XyNv+q8AyX{8({8P| z?Cshz#bcc$HEO}_sA5en7qXJJC|oX0*6Ut{HlmlNg;rRHcoiR#H>`R4vjb+HmMjlV zbxwZ0w9|p4i?g@%Ox#u=d9eAJivL=4EgG`;ocyG}YkI{~v*SWvPue-Q)XWi%8-_hi zoLH&A&Uvo_JzQ4I+OzwG`#)PsId-erE4=TA$`w}ST)3!Ap&BikgiewiD)zYhf{+dO zj}Pcj_Of-K%ccd)O!FV|Tk-r+m5EEwz3OXJe^}uMhmCjS?=`CHG{<%ML*f=#JK1k{ zfA%bAg#Wa1Ziho(4Jojs&%Rkrdb#?Su5o`=%BAXSOwLO0Tp3Yy`liP37S9a)@}bY5 zVX4L3_byZpFE?bM(SVyXCY^aA-g~>}nK8aQs7uJ}7q7zJ)yA*yPisQ(?F5WB0VWGq?P) zgpICcWW=r{>&~w{JfznOPKv#=qR@pcrNjQIu)bo+wIwz#x)l@B zb4JXv9JBpm29_%Q`G}<1yWYWl+uq!lXdT|Vc-4PTM4W8-Irsd{duj$0db-iKk?GmI z&6_uFv(?^8xypFAcvYdc970p{e&4``%byI^DFzw1nx~T8ZB@uj156vH$b< zttVfl#!Vend&$zTD;ku%C^O!!o0;Xk3V{ZnjL&LS3Y7?%JB2;#1k4#y;UoE-qnz6J2O2mxjlI^ z>eiTf9mQW#cR4IOZhj`{@v43=ifn8+yLR-hH|M-Z7AtDB*XH)Uv|%eEK3;4*B`UDb z=L-GP=DNI5IQGB4GRAG^(sBDg-9Iz3TbZy-&n^D{kF|RW5-kd|1f8~R!(KfTPVIo(0f{5Cg|xn%uBK zn;9%!;$PrN`x6!DWL40c*(~&gGZ-Jyp~PBc-(>^yyx5aur)h3LujZgF>?LLU+fWEr zI0HbNG65t0P)x>*Lbwd}H|wO_HV{(6ro^IMoLW-1`NvsI6TfsQnn zfB+F(m2f3-N?|_EgNZcTFHl5=>kCi`eZk0R}M6dW#V!a!4>rVgXWXcLo!N1jF{}Qf-A{J zx;i1I;8{PEwRbS$uEb2!?CK0oo!TBZ zMcVJpt8~vdMiQ~f+npPt;vuAcHVv#%7}+nRUKEN>mt~Y#y%1)dDgsJw&6Uuy{00?# z)∨#I|DUZy1?>V;HvWM5PRX3iXt9C$Vmbylu|t)=+niwf3$>XR<~92Kw{dAl|db z%?z+oE3(;Rh=!GHvuxMbk&m)yct^`+c1%o-mQE}&=1F4U(?;6-O4zWS(`y=;I#rqj zbm@?F-{&aEo0Qbi%EoaBv4etwG4I3`0|(+)jGoF#bbaEA40SS*jL=7i;=uTkN<8u30w<+JL!uE7Kr>bs7># zW`A5Xr3I6*7jQTmvmCSz4$d9oYh5~Cn06Z$?v@-t zz_Q!q>b*qiQL-7Fx7FC?ciQucx6}aFHK_WEFcBb!;OP7bUxX)D%4dyNo8B%2Ir;D6 zckKB>mf3T#c;2xjbTfC1+_6vfU`M))HkmoTOm7f`53TsRw|UAX31u@LnR%UW-usP% zey%u7o7PPq$9OU4$NwP6R-!m&NM<7!@Y9Gn!Jw8Ao~iD+z)%m__e&vye(Vol-kdHm zfo1#h$;;t*{)~0E!K-9PP7AC-FqBXvuBm(xhw3`v(oPbZsPftA7rQss`*GAG6L!n8 z6gzJ@)0%y-%Xx9ztK1bWXwXhS*S_&A%BhGt^pSM=N-cVAh1=km5)(v%WO?zphvTHX z24@j%@lxx^te|qSa4?8%o~=kQxX5#n?cAEPw-rMXr6h=VhW7CL`cEV_s~p>T{w*aVlyY+nlD-RWp(Gp0Vz?ha)i zh+ife>f%b3G0CC0WYtu&-dzXvM4|kR2eN3Bjh4pMT>jH`?SZa60|8)*T0NKcP4af5 z0wT3<#}Qpu&s5usa8(1qO^ughls^8<5!&lDWKlR{Qu2m})@2t7vxWF{bjJ=ajwMIJ z>;W9gzN_{=hu5qvss40dO5wuk${bbsA|j(CN!mqVElN-oWY^;!k0T%>*v?WH zP`{ifan7dq%RFJZMRmJPV>*U3QU+D)XfOIw`6$XlOZBj3K1`0&{_ui<2K>I8e$#^( z&E8e~DI|^uIxj1HijMp!Da=5cIXa#I1MhWzPUK-_RR7t}g9)Rzr2_BEv8Oz|eOWv> zP94ziv>VMOiad6%Ku775z!|bdcYGo55ZFw4S&=r%KX)ixJ5K`xqT$*p%^-EQN;q_% z-6dlUCI|D1dJiLJ6X%JSMNIA~NIpXf02g?J<>u|=Uf|S5qUjP=%e|C(G6xbhFi^pd zyQx~64#Fq1O}lZ-bTc9>Dq$k(s9>#>cl?IH6YC_Y<_MfqiF(&Tiw(6xlnV5el+ep#6VGYA zuu4y*iv%bkrbpkd_Mf_iwe{4#*2`gK9`;IJY_PTqvpL=knUjmT-Y(?RQiB9Bqs2IR zT=*H*fWMc(UpoO}+BC)vapK&2rQ_kbX=hbqp4#mjC6V^ffAh2Dx^GqDHJE+0#ObQ-df zJggBtwv+45_*UU{yO))FP>w28E?ovC-}UMc8Lu_lai2i~*1 zk*tFEZwv(~$2QL#gKA@(Fm!Reb^7|zXjP6<)6$9{m3y)!vfQ9!W2k^fJtTovS^LK~ zPkMEzC1g@wFD_hyk5yvKVwk=eyL91--^7Las?)7EawFv1v_HL$qiDJyD%O{3U6=$@ zB`ut1K(0lu7l9hB}(>2{2ZxL zPXQvuptU)DpK4H6AC|KxciGYVW;vJ(wH{Y48WOJXKIF9J&ScDzttbTVgO)7VDwk4G zirN@&UxVqxZ7|!0IYgBC59&1n|COMMzhHRK7MKf3j$`BfL!^)A+yd=aC>G{S zuL7yn-kgX^Q7}l<=&``N7BQ!2BP%PvD>$=yj9>FTQcNoI4ZRG15{VyPL)vE;=Gq~V zC4^Zs(Rs%Zoe5tGsS=6y;+mqz{?RJJv&n?Er{dat2dg`oSfD49r9<{vA92usk%cI& zJJ4^0Zs(pu;QPekEq3L~(rNulYM8$-N^26prv41Zx*nNn%5Q^Paz0~j3T>104Z|Xl z^+&~9utC%M+&xA6`miiRxj?X|GB{0jvw1rwuh)fsv}a@KuqnC9%A1CUVH<+yYO(jU zTCqjEHMUe_pYPfvNm($WYo%b7tWBgX5XXhui6{eXy!IiLXfL9YA(BkWDEGgFK_oxo-efn)Naj zE1E0IaXkTc3zqb$)#IdhyW93b&QL63h|jo&Ll>mk!u;(6%W|+3xyU&thm&mRSX8>B zv4WqwXi9uMz9LARUz-xxS29=IUSxbzmrbaz;f$4-SwP)D78X~?wnxZPx3LLsnRWAf z6~xA=eT1^l1?Z@fqN+jAi@5Kj>ef>ACq%=^*dyayc+1TF>6xaX0MmS9MR@XY+-j4_)Ur%<5|15OHV(W^>Ie4y<*GU4)z z7RGiry-$=2MYAI4aQX7l0#mpuj2#pN#@y=)Jne$NxPHE32qm57o@WeWdup=`h{-4_ zIE`>>!E!H8x~?gv2s6GRDU8BJUOjP_6!HQWG!-Y-%s&l$eYMSL!bzv_1c)#p9X@mW z^#I=`H586&Q^kOd3T_7qL?N)qaHq9~t>>$>9*>jOUiV?@Ym7fbQSh37@0pO0Os$R^ z3(Huz(Y$q6#8K40uNm_3Lf=6K;JWaum)ZPLt%dNRH-4bR%b=AxZv9v)FVbpm1NU_Y zv;`p88??hu_`HiSaI_vfO!B#$ml$lyaf}F~(4_dBc!Vn!@@wC^Kn!S+SkVe&cq}f! zcYBEnouSX1kfNS%Bsn?CcvOivoJkojF@Qp2v)5lH+$aRIo<*>Xw;qte{%Csu1-IZl0vv&<+-1L4>NL_!dD(b2O6b7yf2?nP{_r?r2wG~HA z48VgDuh4=-I~_}VavIB~@3lg@g0XhIE9mfZG|hWHl$%u@H;X+at#4Yxlk%Q6kVQK6 zZ7C26KUaDCa{RjpiABg?HxJmEpPItX)zI}1=TY4@zfp*7%y-XPNt>%rz3YWAC&knV zP_ipIYQu0)9H?253UAG91y|7C1o;b)!iErPczBkR1Ow%Y)t|ydA+NkynWY#o|_&>4eUCYY5@1NveqR^4O2S*2oGvs$6I0lc0b}j*l^d(@ar;`lk;JLKOmwSLJZ}j9?;a zjkHQgz|TSR7W>pJ*u$7vR1`ByB$(UjMuL^kG;Ae$`gw`zQ)vO7R3iS9!(ED*Sw(?_ zq#GS3Mqv}`w(={H#i)#}-1v}+*KZE-W8ca0^0k*uhn}>LEEATGje*k{Wb|?2I~sGv zfRZTH60ndEZ?BkR8A83Zs%Fxhs z3xd$F7OY8U=7q7FysL1yqHLbOe)+@tOTEb6F^uAhYEVoO7VOIQH-o9>e=kKnty zW5jo7%ja1T^P@fbjS}ieQc>a6R#?YLUNCZlK^>V&Pc1y1WawV7!`zIV=Y<`M1EX~b z%NAX?f~(yh$(z!J`+QTzY!j zreFY>J`?*!Btv#3lXD#0A0w?wwaaaD_MXWkGF@Ek<$MmaOaG*RRY3Kh(#~Pdh`KZb zQZVKbbD|TZd_M-|%F&5VEOMsYqwcR2_id>UR3&Q}OFriW?XIb6^+L`#2!U=z(njsp zkhJ7C3QO)Q+&%*}v>7)CVmB%-40DcP9Z7J-HRWYn?i2dc+oZkT#%U!snSKTdFvy#V zZE?WIrzT2^*jI8Vv`b!)D(-SI$h1tbq!sV4R0J}btvGLcS?TM`L2>0X<+~cKo@Xif zxuG-DaMnO8e%2a?BsCkxw&Ru2)hIRe$x=%f?t`jz*-GSL zk)Pd7(Lz!|m#C2j9~8G<%X(ud+#cjiKse4Z=8S>i2)7{`|DyPHeFoXN~q_0c-OrNl6x#}kgvV^@;OoC zM?z;D<;BUwDHkj_BtHc_^c8kqdtDK+PldzH(=7c6K<<~}ymvy>f?vhaGx ztN552LTD?|=3pZ>tF(%)KV@%YY&(8b^R~Z`ePg&-e=lO@lzRz{Xhk=1m$r>7-s|>- z;QamZs-|qhhs9(1;TZLXbL4Si~o zkZJS+NRrOMO4^oj`0kof9BUiw{*D=;3TR?HT>``=)uV{Nz^_1oz zd>)~TfkLI=rItsc=Y?@09qC6kZYNi$D@FB9VjaXvVk}APzf(AV7n%HpWpGKP*`h%^ z?e)!b3$&Afq$4MI-P*Q|AFNbcTcUI9jI{l}U5%7&ykFU2jM)4<&weIhwx1-ouD*Z^ zrv&3*>vyw2u7OhBVuPIhf+2w2s;+5L8dTdBiS|!Yf+exv*naXzcu?@3AgtFcH(CP0 z^;~%dK>PD%r|Ja?o3>z!(o_;fI^$0g(IaIRsYXpcokW#)qo{;! z`>+^$0y1>#j0%eWMEV%U>`p{y4a5?Gy-A2~N@MBBdL7Y5B!gSe{IYUnxPjny z?(y@bIW^?np&Yg7(i-x@bVDUIlqf7lC*=Cqz8w$jet4+Nmy~K*7EC>|jiQYbgh8Fu zpwvE1;0z+p(u>g7JJLq_%@zC;-<)ks8y{tA%ifbYTIUr0g!N3T%kW66N>>tdHHa$f z1i2R$#*R)#`d^fTDx}jVizoa&c(#Oh@9&?h?$v0AnBQ)E7D0F<6FD}gyrk!eM{;Ad z=lr$z?3vg}w~#wQEvTl_&_?qwBNy5;8Y8u+bwG`9c-r((GiB$p>CXxy7~EE#&OH+v zK=z)In1}|ClfO9q8EPMyXvYvWT~OG2@1WG%=|i1;nQzvX{^4g2RnXIO;RSlz-sEA= zZclCzH5J0RCi2ChE&Z{we84aAlDP6dybQe`^I>F;kDqc8Sy>&~3yjV~FTQDE5R>Kz z7NjEvZ)MzfpFWUMMyG#a^&vKbV|vN#?Rq*`|e+2NF*av zdNyKL*%+?{#jvC0RznlIWJU!m+I_*c)A3L0@!anYo!Tlv?Ida){riH1>OF?{wjHtW zhM!fl@o9ymofd$gIru6Gloshy9h-n8N8m%dl&CubpygySopochG zU6guz^FhR}MS<+J<;~%Zq{Y)OcAXaK%z}B=n^q_POatU!{ngnn3vZ9^9sw{@_){$T|Dk%&~v@fr(b-R9*iIPKdNO^HFXi%5O=xdQ&ymbbmQ8 zqxP6q@^#H?0!S_@Rz~OV%#*3|d6JC5(Hk9t33pl0yPiYf@T%A8G z0vV+3sH3X^6oQZy^}?9^)OO(i!Bzi5-mTQiUd4|%HMt`+Y-BDm!obiWDX97LNKq_Q>h?qo z9{$*YjMe$pa%CIqlBrwSmlc!+)HlZ_Yee)ml60o!%mkWqwsGS!L!o;Olg?d+G?JJ%9{0xnzYozs-6PEP5_|oQSY00_ zi!j7SGjf)uGJvMMF9)hUnOdxoPvDwk1YmW6espKT9AZLcf;hocGm5e}9tdYFqc6??;fS<9S#E7?#<@rnirro-(~UA{YYwOt)&9{Tp)9-3?~ zHXNYlg4|03`i}mt0Bf)s@dsgX;(@IQX6cXTrmdGOaj}x;d80)4fSSsFh@p3m)c}T3 zq2m}R$>gJZ2QNXBt<8N3mE5{rzoxU#tt^a@RKodn%L>v2(D+d?vAi^)zV5S)0H*!* zDZQ#6Pc1dK#?c5}(v_1Pv`SIpNmgn-KW|Jd?6O86#O;r~NtBw>MA9-iH(fIn_a)ud zy}uF2P)oHYhfdX+3y$3t;J}$Q5!Bv17t8*0aG@2QUSwLlkA7VAd>T6g_YQpxcy9L= ztwEox=1lyd9J{)4+gpGtOff>K5_=;2^-ZMF>hI*h%c@7LhCnMdJK3W_XmLR(5QXCG z(&vhHm*xkHPjh_SJqQQ?Ii^sg+vqdhsAQtRgV4ZSDaYCWu+-z(V*bNioY1``^&Ru!*sn6OnH>ua+r3@OP54`U@m)YEb= z6~l|N=U6+tpc;lwXeybTFkCNB<128#GnfUW+9z7i4%FK;K?_(%IF|^29@`PUBi*O- zlvs(kdIo9*BlHt09VU$}U)?9xIg^XK?V?QPKFpFhzR=Yeqt&Ky(I`4|uZBG2;#%Ff zd4*M?+%npBoR#2cWVAey{HIyecI39)Ro6a5k7+~epeZ9(1S$vmg`oocsy`fSZl%^ z)aaX%%*tffd2dCHmfQP6MhC;=UZ~#KE9+f#z{*jdS(u~2JwRY7HkwbvO>&MnXjd=& z5$u*2M{{I@j`h^}E^~w&AW?zBF-K2-;@zIMD+zxbICt)T;W&^9!><)^W8jw$Y?Q-Hu%A)6EiMy z_knJ{4BOqQu`*3#TfZ(O4>Yts8gyduMwNI33TEY<&lI##xGo0`p0(!4o3YKk^stf* zESD_A9cSK3Xf|2|+kPvgenB*;m3Lc}t1}Y8X$$eAJTYqOH zKy{K&x{8#5Pt}scbl#=SGa_D2u+<7v<-tbtd|vT427zMWZdu8QWd!vpOoO5`D9f6{ zwW*mb7Kt-6Iy28?Gknr~CmSu-i4Aaf=O>Mba#08w1hNB+y)ts}7Qs*=)ZCK~@o`-j9z zE#66k&2Gm;YLj59ETvnJL6>)TCuBaW5X-0U<(x*eDdq<46IJxcP914Wr=%B!1qt(s z5#C}*2G7-=B>R1rDa0U_B)wYW3Mo*ltFa+stgKbUPSbVe9phIrS zr#UoK30-UoEJW@cPN$2IvaV%c)guR7PXytZj@k>RW@&b{fJ}Og^}L4LRa3Wl9u-l) z3;Bf`bmq2?LtbOA0qq zo8cSNpThKbA>-%#G5qczojkBrzD`o?THqp3XsnA^S}lS)&X}b-*>(4f3gtW8IBjUQ z=rNzFDY{JVmOFkzT{{x=B*i)M-r%p^YF~CXZ)&?fbOmhZ^VKij+w5cX%gTmyL`zPK zp%_zF%9%@HAW3_qe`amxdGBGvsIdY4;ER&x%vhie@&3tIpnB-o#NRs632f+eL_>V? z0>cqkY=62mF|zM0)`X+cFDsRJIL$h3Lt@z#lD5L~)J5p-s)Yji3jIC{vdcpXd!L^Tr;q(KhgiT%pq)$K)sRQ868KL}li|S2 zSW2^SO*WLUd5R&?>5D&qHAkeMDS)Zu3_SQ1(^+B*_-%{uPG%K+0pO%(dG4Jmj`k+f=jb~=El$neKkQ0wmGsMG~jLSw1=F2ee#Qwa#` zIjjtztY&MNA~|SVteh1Wy(h`M1J$&Se3tRm?{d{UJwm1&eb7t3l<;V-66B4vm-;vK z%4r7Qmv6m4QTkX++tVvHzg-USk`sDTLGTU1dhZ5O>n~ZYo0w{fGrGQhtxl^`4**a| zB2DdKZ-RvE-5_sj0g<+KkhdXW7~7Gzcm6%NU~X$YU48X-T3h(`yPX%@&d;7+MM<(w zvfJIU60EKYD&r7`{Xb3BEW%E+cMPGh_(!0+#Ww=FUTe-mmO1I=`jP=uqSW&a5jn% zY$#U!+=hI{3bHCZU#Ys~GkXJ|<0u#{AQ@aK5Lvnxe!TC3kt0RzhTrBs;ISz;+S0A*l{#HfIlvXYCtM&bDbI8tTDp}fz0?QT) z2+*;u{=4t)y51R3m<(kNEy}4R6w9~&OaCFZ_BgorqhVUNv^k`WZ=CSO_~ylBet^e= z{um_W(uBMN&IuD!p;@I<4=K_|Eh;cLR;x=G>-Ak7Z#qkNrb-OOrZ@2tXF`vv=%$Me z9nFCg4ev1KF-hXAZ?rd{o)gtVf!mk_RS(je2{<-CHH5ok z4df+m(+(WmtGb_H%282A-z1IkqC3m}-Rl)?Pr6&6o#O)hb#@}2bU&Joxe7WR3_d(dA8rjq zh0uudevhitxA^wg@#2mIpVXmblwMTRd|}&*W%MY(b$wzkF%!SEPddU0y}GD@)-x#5 zA>*HF<0lCeF>@VPT7kZe2l)!I5Dh*haFR(i2#W{F3kO4vZB`F-k6QsxTEXa`(`P@9 zphiA0I&dZg1awPyjJ>>#Lf^xRKL15N?eI#x@v12FBpJN_y~%c;cpk{bb8K-(PyJpf zOZCOKSwl614iaS%>h(e4)8qrZ%EW_Xd8BC+BfZ@EF7L?S2pJU)l4_G2-W8=anJjH> z3!20LY%lU}tcB`dBGELA8jIvLLIh?##=b3m7!d85W22^dUWUEMK&3e{G4C5#Z8&7@ zuz36yI{?)y0kOU*syJMZ&P!By-FYS!;#y5UN=&x#WV%nQC3B7~W<>296GLk92EnFA znp68NGVukpTax`8)K&`^mF=3C{sG+W`;={=JCmkol4c${Z@JlP66n2Qqu#H0{i5?A zU!>e>s-oDzXAC4%>)ZQlvqmw0%YH|mmwS>0K&S!2o;OT}Mo&~8lS{iTkFU@rjAIwXTH8l+mmmB#g zQByYfM1A4#)egxwP%vFT6wY{cgYVPG90Mhl^Q2ClM9&~jr48xpbDc&#XYzrk+WM1W zq5)d4q8X_$vZ$}C=Cyc80X>+mjC%Bep~9eHE=g#2O3$cKWgvJ7INSkB%0JFxSkduT z+v?C2QYx{U+a(}lwMd+Aj%fQlub}xf5CSuPEta~@L<`@n{M%{K+EB%NkrKe6I z5<`?R>;qCxp*+@n-};MTxGM@~KFJEs89xi|u@OR4WyXU7?MvW%P1Be!tg{M@R;;(s zv?0u4Szl;o28klJDj&mJE(G_2`Qvx z)j|w~4cje#9VI{`bz8b=81RwExrY!{6>@Q> zeO7Y@R1%k#1)TIm8%Q`Ut^!dQF|RcBVNN`)lti4T0%>Q-QRH+5F(eW~SH?M}qx??2jl0(#tgyn^QuDn60Kt7wM8To8F z?@C3I0?K6$8uR`&cEh8?li5olt;z7+dqJ@*lBUn?P`Vro@j^}vnx92da0unOVASw~ zrYgh{c_wNwiHgk=M*2-VRlJ5A`Q-&lfZ{DOnN|(am%}}ZKP2(Pv$ zg8aJUW@27-V`GXQeTiYTx()u&?BjCJmHZI4zxnHsjp!SX1_(-k4rwWujm}2pdbCwwB}(@g)r6 zh9oU=;a&w@9>TQ#bzFNRBJ#@s+Ymq{JCq#fTpQKE#{g$T5Rt;tYnUNcMG#Sgz{jE5 z*eI89R#o8a?#&E+;|uNKqG-DkJMixSMGuX~#Gn1eZm?9IQ%XQA zs3ZyScneV#hxa`4!WZaJ7AtG+x_+WIcOmXYE-2#fkk1Xk@+C}nw zaE3tJ+RmIfsh-9!(DOBP-dh7jqhNV5p9@6Y5X7tu)xgamTqAz`cgIuaPyWZ{T6|7v zmyvqj;AtEs@oe$jH!u@skv*MA!YGU7fih>iVv_#?K&g}FSqK4HnS#jx?By#BQ^Y1J zW>wfYV4h}wj5oEv%xQ&_-8zXO-W*qeZ+J1?5CO`?UxvE5*nbm=!VJ>1;!h{AZw`c^x`I5ty5D?s*cW@c-M=*-p?O6*Yl%@!osh*meJl;)`{ea+>zMo z+Gr&GD2_VV_{!oCwO<``Y?hVLR_9kWM&60th>gd?tx>weaFLGS9TUGzJRDvqevzm^ zWrEewHk2>%J>xe=+~{IBpAVV5ANIM39OtUS?Dmw<+uZK0z&rN075~0ATP2g-=3yAh zt@?mu8w`CP2Vr)(Oic|8LF|iinH*2vL1<%6Q z$q17J%R&HAfzi11WOXqsHcgQQH3(VkS7a!1Y6TmxPBZg$3krG=tw8+z1;9Ie&ls$B z4(8pBZ(_cWo|9Iki8cs}({?4&8ylpnxsdGU|=^>L8%Wv6en5 z(r_KH&#c!WM~~08#c+Q}=;6Uovb$$}uoVT?!!puipmx+CyJT{>LhbTLd*)y87XC{t zJ}_M~X^j#WN<}IAwUK6>#F1k@kN-Uy9Ka2(`rg3kYHyStu?2Qtmk>BKTx?thm#mRI-iEv-q!lgB0u32 zVw3MQej+n`AZgD@@!qi<^x&!YE?4Q3z5oR)@!QV#SV1iX^KD_*u}vx}6*o+ubz{;n zMRgyGmtu;z2CT`g?CAS8HUvF^dz{Yd&kP)Xv()f=nSOoIdyJ+_hrHRa6o4>-%KBJ* ztJG=Eq{CcjbpI(fPP!3{I&L*K(Ab0Frc8=?@o6iR;9dSWKIV^M$99hiQ>|aZ4-H-; zXE+MAg=&@%U`(bOXU89|*au>Q>_3g!+YMnrK$adT7OWg?H%B&P-L>tod)eJs-40Ryb%D;Xrd@9~Yr+X- z8r^Cwm}SgEra0gL3UKL{1_khJ96A&1m-hq;;)IyS^LS!Ys_+T;6jZUWb1d~p z*uE929C`WAouW5K2bb4rM1Gs_md@Pt7UsEKX^@k>GfeH#`7}%hq;{(whpyv|4j-EB z+pQr66d)7<-d0@<>gOr6iBczr$nv80`l|HAzbBn5>3tQ*0MUM^#m8&+E08R_dFE47 zma}eK%TaliXE@Y+h3xN<8!E}V^gsdvmqb(t8WyZ*w_ByA*7rWff?EdJXlJEpD!@k` zsiyH%8`DtJRb71iV`Uth6Cs*;h>BZa1Q0Dj%hwT+}@LY4ri z6k$n*I1Z>5@jq7uC`#l{WhdCDKZB}>)Sei92gu%F>y+rg+g_8{jCZ>~tiLxg8$#-c zZEtFy$h%4gC!X7^i)ahd!kejBi?z%onF%>Klmjh_ZTO(gP1 z3LddZH)FTfd}Qto@hA{!Ve7<#uYhpsphb3-SNceN7mx_Vu_}F0ra%@-ra~-_$6xDK z^BvF$gG@0pixTMMSoU}qUt^NCF0_Wf9@lQxI^!$zU*Y@Q;HG{T}e2mcJ{JI7>jb6MOMp7^Acm)|is)4(x?LxWH>bsuJkO3BXCQf4^D4$=>u5PNi$ zX*#=JfM<$e*?^k`-~bf5Me93uD?Dqq2}dKRfl5yl_3}OpSKv6l5UQIg0?asmoxz}d za$25PHfA;|T!q`PihOLM-gqUDxa8&PeEL$D2PH!QT1*&JMVBf3&e^4IxE9%tVzR0Z-i}nEGAp&DwGx;eG!dKk z`q&zf8o6ALSUNXT30`}?jZC0xhN<#|AB8kz@rHL_3H3cEG<$+>QuDr`i`GoaoZ&D? zQ3zfzXDc?^jK=%4yX}RaLrkBWW{V9%(cwvED^dqnrMMz# z7qI9*8=_gj?_r~iYG}cDT2v1d@tZMa_Qt1m^#LCEhNYp|>3J|wO!QJ$bbyo8CGHzX zp#n5Kf;P3Mo#d{cYl*ozWI;tjVK}uuM%;tFLBwjm(2gbYtQ^&5v`wU%hv2v*CUUWJ zw7gO#e?oHGeRPi2RjB-G#hZ(D~EKU_$5JcFa1WK2=Ry@ zU+_Irhnne5^R`pGX(RMHBfcL+iyud+rMT9aL+=QyT`g9H>iY2fvyh2Y?iSAruOd*d z#b4nq%{3#=M>(v2xoqEYu8z3jHy%}bKXI7uBq$JyhQ6dyT?#>bL9Q3WyZA0`T1V?_ zxKko>Oj$0Lv2*-`flwDo_fo;{g?I(2?xh#sK;~o&Q=)o$*jvHcEmv69R)XwbVu=*a z`VmTa5!Kki-q^ zCHUVc?E9&-D+X__eFbc^Z>tO0F3n=`kzOi8-B@vdC<-T2?n93>`jt!9#Jt_)=bN3v z77-=B@Nv>>6eMQ8D-fEINWq+}L5t8(+G#t1a$(%Xu)}AD2nGAUl!1{m>1~!x|U9U?a)1u6UB*DY}arwVd(A#EeyD^J%u439h^HShDZ)m+P;yh@1z) z^JhVYwZ{Xis_BHoU#d#15{XS7ll|=Gijxt>wJlN=qY<-vym7?(|`&6c2K@kT&4r+7FRen+>t{y(E(rZgTGk^0Gh7ERrbD8~?^dUe?n||nmo__s*kpYmkn)GIC;avN&ZQ_5} zvpU*2*?(Ux_>dO;07<0!x&Jrf^#3rU|AU)n{MR+l@;}Y#e+3r*6Y2jig`SP$e^KcF zWBK0{`v1V^|6de(_Ww}m{}tRb{A-%xVB`2-6ne)0r1$^3LjQki0s8+#qW>Rl7yn7@ z|DQqa|7;rgZxX%oRO}W#Lf0pi^EMt@Ye_;gU^KwUOQI0I`6m3G&@7g@^^gPH`)B0P z2-+DTBGXB+{y_Hr@-V=W`{4Xo@~Nq0Q2F!(89KTn)~>wP(4`QV%v2 z2kuP57?15W*Y8!`B#M^zcjlwhr62cMn_rARJS#?3zqRCLI-$-Qo4i)n-r~ZF$|8%= zd-Y+{i{nw?0|3+bB5Zj_iBlvC2R5Enqa2(iDw_kdpkUxbH$|GhJE1Y%Ec zB02%NL9JJ-A4Y2@zX{-J5Tkn*K8rXKJ0^ETqHh5;hYnx|=tJnsPuncCEII+0N$?Wc ziRwY@sa45NyGnShnJr5Lpi9N7whXTpds-|C5)=D|`@r?W^~CLAcXzmQ*_48X!{UC& z`x2K0A!pnQw+?v}IbEs?(i78z>kZ}v<-~QeA0J?Jikjv*b^qSeKwl2gfNVsXMVWre zK4cI(6ZtEmj=s$E?0$BCRnPbw<8NxtCaH zmgrse9#32uu;FQJ6+u&PU#eePPg~wLT_a+fyW5kCY#BR`6ApYQ;80`6^#CKRFV~0X zOY_)p5%tM=g`MG{dpFqIm*MW>Fy_4t`R(%$n#e~_*adC}EamXoR=I2}l;ZOPWK<>v z{NHW5|EjBCVf}Ag?!PJ`{y&-b|175XFXo+*Z2?P+Q5YPd?**J=e6uN(S6evJHnEvr! zqtg?keUSQwdi!6q{2XHZ{_TZXkPOA?+1(*TYlHJ?@V8H5e^DWosMOrteU8D&sVQjs z`!_i-0ysw?t#zRC9|Zs;7f_9@jV*ucoOeLO`1%%>brgMrYXiG;5P-~#fN9^gc@WK@ zYJfz2tNNav94dd&=$AYd0znN-!anu{3P42yt$1Q%iQmjOuO*Ue92)*#O=xH~Y%c%N z#f^>JxQq>)M8@A7Lq|rlMoXW@A=Cg? z0)QqAge&vcG%-w#+@!)n*BbyjQB_+!0Ag%-aQ~E)9Q6LF8R)xrF~WX7FG40N;DLYG zyO;!~mJ=7itzMPC^fMI!A0FQn{`o_j@E-2STkglNVB0RsP#gsS@`&SgQ{&CxJ#U5vKr^5nLAJgZg0Tx1u+EI%aRS2cK{z&fz}*3-U0dg( z{H@WOnzo$q8e67kQ)Bb_METKdZBGy0aP-4K`fL57{m1_=*4{Bjw=U}1EZeqioU*G< z*|u%lwr#sk*|u%lw$b(c`0}Rnb$7ax{x{}Y*}3;l)=YNxJ+CpG18|OdFo~8S_jgSH z#o?NQeHn&lA(S`o@&(_1>NGTM;U~KJrq%)eFpqAyWs5!NuOm;cIx-}3$wn2<@z&llp&yA@aKGE3m6 z6GP1xUv+fp`2^AQYvqm^l=ItOOwAX20r5Wi2&5d~{UZMBfc1~Feb6T7^v}CX;Nk~- zW*}iI6T~4LlW;dAr%$|Bt+10$8cIX2FZoMAd%Dx=-1EH_jHv7s=9K@>)!2l}-FE$W z%GeF1gt@`V$K)zWw8PC~n2cew^gL}`{@>bbs+l^VRGbG|G73p>ZCKj(98L?z*m^-z z2`$Bxa!-3OZ=HOd!IWuZl-kZ4WTPxWg>*}KJiIzHXJ{?_In>oop1--Rl)6j{h#MAc z;qPlc-`z4#kC7=ua5hZ-&^wE(H2V0l*tP0mcq|C?EymTjy zX6o92x2>-BPnIR+&kYqixve7_^~&I0BCfJJ(H2UANBLhw-9p2>jUM7j@GpY^J5>?T zCAEKmK#^EI68uEW7a&awYQUx)z*&WMUxY~Vu)$-$gy=J3m%iG`o!cR!NZ;VJyy@z? zJd7bY!r2|AVuys-`osdD%PyYJkCI~jQ?&-gAdO_)nin3#!O;Z;Ac^SiHN>=FgwSB~ zpSPVL-l9b^5>TU{dj60osKQ{31GRC)d>v-J?DP3yve8vKKJU0B7T88wtZ%NMHV=dL zUGnU9#h}N}TCB#XI3$pXdfjNtAl%LdDhegYo5!&eKe_kKyp8I{uJ}Z{#byYC?^bso zBYI-M|_#3l-qG{*Dl&oxZ1l-SW(Azo<< z%1K_gC2cH<1%BEz2th_=&Ivu8JoSzzg#YF7rm z&>@v$>=JM;Wpo#hTT4rNEQU~r1nkG^X=BwqjF!)0Zj-$O8-^*qykmE-^0;ttDeWl^ z6Co?IEk@gDPvTHY~Ky1X*&n)XW)MM^29WoL0h#;ErPdg#k=(c2WSI#3!dQ{ijACp87jKe{JEsJr)NFNXNRYM0(ogo@txd;7p69VA>aX7 zKxGJm?%{R7lWd@CJ3qdoUS4S!#oy!UNaLBuNu7;hDTe0H$~a5vIlj%3XFZC_NvsQq zU zGM_KCIEkT2p+N4+Q4;?ba_3FuJcq#2kZ!I^YZ+`t;=5itkZV~{%$mr{YZ;=^L&d-N zz_FakPA&u$V$*q{uThJy!tgAK4l#-_<~1+DB&F?`~7wC!TRc|U>`-M7l2uzyIK zL;$@~8*KIWi=BHggukOQj?eX*VkZ;tB4FPztb-|x!|>|qy%aN^AU;gw z!UeBD<*DquT8%t(XS~3&r=#oy!!gbU8dWdS4}w7iE8>`(u^8smyaj)6SjK`mGgM4C zSZ(5^{7nD7ps&Ynfb>tuBw&3eaW0m%&Zw699sTyWH{bSmbm}->I*#eR0?!;Ma|M zWv0>!PBrPug$vVZuOhh0gu=u#vDUF-Z0Ol;TtQ7A!ElF?V7j(C<4Aw-64cJehD_5i zU@$U9Ir=|Te`cd?8p|Q@vP|b`JB=PCw=wLZ&Pm%iTEp&d!ZWhJe)CRRWmX!#yZ@DG zD>cZAq1k>OEWAbv?)|>Rj)~pyzRL`G5*?9Eb{Up_wBJPjJwuEH7otfUB1TI?lPhf_ zz*O6VHUHx@gDX7b?jCog(<&weMgB(jCgh|V-8j=}A;6IIhvk_KIAg@im&5AyVp^lB zsgopi2HnUGSF6?e(NYZO3_`0P3kZtI^(5%d?qnx=yUzWCPCfU#%0A~MM4f?iO?C#x zF?JxkAsnK#K{I`j*w`%cE$OPSUo2d%78Th*i(Yf_h^uB@*;#t$>)2{uI_9Y19RAXy z{B}A_S{y?OcT#~QJkRgGZQl$VB$jyr`Fys*%%mgAv3!~>U)BFL0TX+r59Oe)m=y{8 zt8Nv*{vr>6V=FbuPGnI7d@4k{D1uRRINxG&*>U<$Os4`8`n(C{lvL0JWDUhIePLU^ z0tNLMk*aw65tNy`)Bv&K=Hg1^XbanzgVShoHyLz6G3xTVxP)3({@Cfr6`sZuKTEPi z&)f1xw3V%#F|x^z1h>Q=H%`>1ffy-ATY(wwQ}4bYJ^DxER?Ky=PgH?}nfs4tJm{Zm zmi`Li3rLaASkehPsLNKb`zq%axd2Qb#;`7BhU?(ScIVtgIG@f5{)0&zFL4{ii--x; zpNz4R>Ea(`4ni2}O!v1?>`OFEh83)P|4$pcGiVQ&M>ac~_PpX#XOFkDEzNNQL?skVue%AbaiPv55vx3-!>)E+g697UIlsN%PXq5gOP2q8iyL zI^ro~r;Hutt#S$Q(a$93&!?@s`-aD>i79%JAV6SOYGBd~Xvxhe#uP0qcTbC$g%=)T z=2xk97p`x&*Hf2CElMtbX=n?w@&~OpL%G0bCx&Wklt{VM+CEBZf(E#v#Qg>=k4@&ev{fv&t5WE*A_H5p_~ zHPQ|U*4Q941B;!>4nelD=zq^)QGl^Xvj6H8=n14gVxj6+z&A@^-D>LDj=}!`q@C4K zwFin?H^h2jyrUkWT&tH>gGBf1#Y`=f;N)F9q~BT>0t_{aOU*+ocW8)!&eL*aTV>{& zv4&%|r~R1}l|8X65k8IgnOn$BPNenSFDLm8w(_O#My2B(MEgfa%&#d~amsb{uS{#_ zV8(?h=TZhh7r(D&8y7sJD{mnwXR}S;rPuIjf<20zj@v}Mb$S18Y}!W9i!YR^J|$z%j9EUm5hRk=73OPNBD|@F z9Y$wS=2=kMeoG<8VQ&YH<)H)ARf5@Hv#4g03mGHd9dT~5+2-M=`zRQ|WXwBirLT#s z)Zg)v(WfftPyCZqvaxXmXcsbf6I)@7Y5s+@QP=YXC?@33$;HQ8F$euv`;ZMlZyRbP zb#vo?3e+{>M`03so#^CkzJn+e2(8il*!#w-rhpzP5#i8UQY~W5rvVVWWfL_B3|ySP z5ppgBs+!?2l(i`~)g?Q99QC4BFC7{#n#X9yMF#J>dP@z|0z2R7VG(OR073*zr3k zr0@K8s2sj5S1exfy2#8pS4h@*M>cW3g<~xufgQA|>UDF8UKVDX2>&#*mghkEn&+@V*BY=r1?Y!7qsqw#5*Nq4$&gg~ zQ+TOQ3fylGbvTbq-UP^`(wP72eMw{?M3MORP7Wo>%fIBBw-PMuT@XiKY|FE5cUjNxduN&U@-FbRC?3 zQ6iPZ;s%O05ki zjfwe}TPV4A#mZqPwdfAwR4H0Jd0&W646H7M^~6Y3ue5p%RWnhp#g%MRV!jJ?ooaIY zb_z_{Y(>rZGF;OwxJ*+v9n*nh@}N+uHooh$^%LuK*x$pHcpEw~F2AFMku>On3Afv{&7rY2&eA5P@lZPp8$ze!128pJy@N&H?)M+;&?@2t z#!Q?%xC`fnLEn znitfQGW@;>p_g0rO;|TjpJ9K|YF(yL;Eif6qXh~j5b(VyJGBB15`M~=oU&tIvk!(M zf&NbS)z#W;ZO`oOcrQCG)*nLOfHtIo>==nAS5!?W^$dJlA@2W8X*qhlzOa+O;!3TX zYPhjhe>NDSad*zQur57}t}M1P-)rMt2+oY*892746C)HAdw)#t1N^2>QPbijr$NC) zbF!Q8Md*$qu1WiDZxhD@hF_?S`Ga*dEOo*i9#g`5DJ-PHw^?4@PwY~DnU?hs>Dgv{ ziycn=yvmK70if#=n2VDneGC`I){vZo#g_C2!2bc*J=4zdQ~kPBm1uvuL+8dAVUNI>08jTe*%CQ@L|#%)?k)}(&@785Tb$v1K|dL0 zc@VY>F!#{7IE#mJXM5Icpoqqj+YhNbb!P=8GveZ`TO*?@+i+Z#VAMCY8N7Ey002Ilo4fpemlSID=cehmeGVj6dC4EhJ);>XS2R{n4|CV0n$@rR8uGb z-II1QCoIe9?4W-&)on6Ly-BZ!AR}%k(p-t~+Hf8g#3Yx$$jk!qJ6rQRPc57~fm+u} zjJbg-G-63OtJ96ODQak+z_6}B{HhF!JFg2h0(J!UNJcb^>2<&K^-XQLig%6Q@kbyj z6dtC7`=*JpQnG5~f>mUUw^oHs4r0uXe6d+*rCQh>e>6C0DyIB^C@o2HYgUu}WiH_B*;v?^ce7|_10)f` zOSVJ3QVQoG=F{iJ&^Vr4vY(HNz1Fhn4_lWF~6!-g+S36tK&;O4R~$H_mv%k5pVJnP7JFX)}K|*%+1{b zIydA=%?>tcsz{%GQHc;djTQ?ESV>Bze5O0*dIHtd2h*MQB2rJyOthQ*x;oP7>ZQ0MCEVZjbOfMmAB=SXMd8cr zd#0$He^6tB2-*ySUB`=ZwN{vfBj}VQa!J?p!j(W5Duxaa#?lW0#_h=SYArgYgAU_V z-O{Y@k#431CEhadG5~$*q&3GF`=oo?Y*{jwg$iQG@{TPpgqosGa$Nd-hUXAv9mglc zY0}~;fxozJ2|AWQdmcYLJ?pB|X)^b)4==#Z&ukK+uyVYKTs5DS~<5&m_=F{17jjD3lM6`MV#R4Ci6!mBT)!kXo z{C(^UTXGG{R8geYrEj;RjtFX55m&h70y9beBqBuJ0GxSLumBl$H>-WIGI4E^+O)_e zU5s`0T^Xuq2^Q|O(S>?CZ=B~g_Tya7xDb0MW1;IZu+~nDI;^0BSdhn%6V~N*8R?Ua z(0v~pBw9W<`9?G&`&ZLLn;`3MZaYMy8YTrfz-JV+G0BwEOxjBg({o=m&&(~Vy;#@B z%v#i1)34NSO@H9Xm)6|9A+}W_wYgZ(<#7Wv`{h+GRC#qxN}LMiThE{rqS1xwyC51w?<4sRYdkZO z5%gL)M7k$sek8OdSLp&`(l33$S+qA!=fvPp81r|YsF?JOFvM8V*HT8&^MDrqB1K12 zVN@@>*`KPGcA^Q`nW1^ovUH1{T4py+>Q(WwP8hJ|gdDlkFCC^4H~IHX;#j<(79uNh zwyWQcFjIA#F@w+^9a<&K8)$op`%=+#sWHxJ7#w=voVTU9H$!D0Rv*zlcq9a)3BxFJ?HAM zNq1 zT+mTsbSXPQ;ZHvlp^ru zjWqAnD1)0>Szo9~Bb;r@4gI7WS|6ar;K~4XZz+g&N=z2XyGv%!@{d$J9T~@^QGwZj zD$O6B-B$M`^na7&urje4JWYp0T@5ltqN;BXF2#uIQth;nPpJ4&Z&6 zU=1c`YTf)^MgPw5mW#L_xRn?Ogfjg}{yfBh$5T(OgeYM-N?9@UwTv}6;%qw&5wY>J?-&;lWEagI#?}i>RwdXlfumnDA1kKbP zfew!!cb!iDWU>Ko+qNl$YLf=rD=xM&JQ6vl^tPl>lZNEc=At!TQ^9dOZ#cjAf;`cf zk8l!4&ZMte$Ebx>EQ3V%>3eUUxfIoDLJ8Nb^Sm~lb*fK`M4Z3-X1KfMahXg=iA=Y&zJ3@o8 zONru!>d)6$tW~@sp2p)=+ADa)5yZ3hoouc>1xPOrZb*L+Qdx zkWxVPHPjIK_l{r)7bQ~^{*wlJtury|b*LhAQ6ihhAIBPGT^bC!iFfqrO9U-zwY4l% zSdQj3e;w2y%+U!No_ExxM%y;on91Xg_9TUj>fNV{t9CA$)xsV*#FneMPlV(9^N9sq z6kYuQZZ`k{NitO6C^`8W!3Pi4xq4E8Pzd1szD1?dqr#tC(oRdqEXu8gBM+#XWk1r2{Z)OzxQTM^KuVl8n z?czKATMD+6F?!L>Ix~@6ru7(N%(d!~s^arS2|BxS+$5ewEBC$3sY0tzqy*N~g4*X= z>52>pI5*CNw@2J?qKl-(Ct5sZ>5Yr9=k9aCm@Cuh?t^uvr9ChB&$Cg@tpnz z-viR{W0}X`yLHScKC}A>rn_1P+ZMGpz$F!6H;LPb+*`RV<@oMvc)Z>nGoJzqrOcc! z>2ndzR6aZj`8~{Sn`Y&!tzM9ob5E> z=uzm_Rc)`_m=CcTAY$U*&YiH=?hTlAhCzpQ22_*gViZmVI+Fp~w9ZnDI*i)o@gnd&pX1a=NPo6~fQ=&S!p+y?*a(~sI$4v#*I z8G}AxqoCC1n8OOGNgJvv*=x!joLq6VdLB2uzuiDGnJ!yg^)we`Q+!rVAwH2amX4Sa zadTA~)AeD(cj#h7I{mF!I7yTEPTp?XZ<6K+t(@YzV5+1-ohYl|Ax0Fi(c>mXT%_t8 zXXZ;vh9l@`(&a_4ab~sIJr{ny^;ftZd(|ON6XTy+%nzN7*1n{du?eh68Bx!iZny>p zXYXmXiazLq-ZA7qAzHOXXZur~pt`G9Zv`04xOay?9bA{&F*Ivwjvy^av^*K##$PO5 z;YbbcofCN%rLuAa|Eg{N=msf7HvloXskz(985|bv(0^6GqrAQ=7*E~J;hB1=ZW~3`f`DVNwAD( zqZM)-tY%#QaE_zTEll0A`x%_eHIgCW641B85V@B2=^M=CV2^ET8>+KmIRRwxm3F%A zCV*Hj^=H_z=IvN^PFlTA^bKRSiK+Hr$4omcTK1yR^K`u{36*?oe0__dpmg%RK#wP?olJHPdQ|$n7%iFP%VmDe$SxF-6Z-d z11g2E&$neTVb7q5H!?r>P}A*bFN{~(_0al7%8rL%{mI+IAU=*~HRjpKb>e};D3tr` z7g~-B96H`BC-QUk#v#-noRP`zS}~{Unn_VRtG3M&Ei4XSmPsbS9=j(D`FF{^?;b zFBCs9m_kHmc|&ZykR^e7BwDE3Urao1MsoKDg3hEMD%POfR#xL|FaY^LU&Cnl@LZY1 zoELXkx*pL}!@vTWAl#R6C>#&oNi+poE@e|-y1(fSwSAtXDACU@O@E}!sPfc5evNk; zr=WSjF+mDpT%bjSoM`BPPlP4nZEccrj!#wIGN$-!@4{t^!uT8RLOnRzDREQtsi>j&mO&6uCE^HyOb)Shg3CzR1KVh1!6A0L|c85ih9c(HhG9poNG zML1f~yRSU5$3a%#Rd*dB z^~oEdr=ijGSGqwv!qjQ(Z}g&c8GT$2&rEVv4 zQ*qY(TeFn?wR|NTT%6la?HA~V&_i-D*5i|DM@u~Ke$*H?(cVM&k{Tr;^qLNlXpr5= z9PBq8L`=Z-{a<-LUNDFsvtOszJFHO${bGxMgKPq>cpC)X^_txuGDUN~_P9hZ7*}ka zOy#@TnRpbB)tt52_|Zz`uDB_~GJP1K&RdIAR@o(osYf5z1u>(g;$En9rxuvV>Yi|^ zF>GI}ORtpOkI7|twfG(!jVbzPh$AL-hPvRvc7r@7@+JP>$m|7VL)~Cy@!bZWwPHOj zeTL#KhdcV&8Yrc(kwbhh$;g(7goCa`v*_E^)WyB(%y*qHlJ&aw& z1>y}8H=^%LR~G(-+2K;0t!v&YWdQ$G3C4-jO;B zK0G{!>rwi%Oi(S5#xGmag8AP%rn2F5?$=Ud@&u&+Z6+=;Or(0lZa=O89}8^=+{zdh z9@ZVBAL>FS=-qKw0{{H{wLEq1yu1}_Q+FREUwTN75cO+W+q&dgoRN8l|J;~3aeJ9- zPr5A2J!f<2d546=7m5#EOiQ;5h=8+~@~33bC0{_QPEF10Qm^Wsey3@I?$neF;(4lv zpf_`MgHkRq4*Ppr%&>1saS=f?^)J(-v15G#lO_Mj+Pq4jKqYT04$mB|7m4sS4|Y=xL>- zzHar|HG7*5I_9?=9wTix4;A1e(JgxVgp4bb%XD7#)wJRyR^jZ;Ih)K%YBl!T6N*y_|hb#hFBb%cn3jk*8uB z5x4N=QSdCaBrgC2%j8!$d1yK=XF5B!CR2f%VFb&@AJWa?CFPoj!;yS56_fMSBzdBBqVR|xft`8+urNzgjs33*c9Tt}d2%Y64TRhsU z57Vde(x!}{VAbFh^YE&fEtB{^t9gUL!ALd2^@6n3y8I#rXwkZJf)m|~Ldr{`XqF}p zUBH=8Q>tu?01Dj^oP&4`f>g^E!`*d^Y`Rd~^{8JB9&lp?#~=XICN1@l{^ zVK5v5lyDTM8uQ4CiUs-_z^XV3>(9dbSjpWf_++9#Vz{*~S0!b_tA)7Ex7J#RD84*V zqW;|NbV}78%ZU-?$(1+LbXA$J_p6Z=-FEEC!FteP^}^W0lSF9abcTc_*>CxY^dFQ=kJV>roA^5;@- zflnCAL@XFUR$nWKZAMx?Y6V8hL-ZEYls5CHs;85xKnb1qnz!AN`aPNHx0xGpqxMN* zDhN3q;a0+UB!-XF)ce7z>a4>uc%&NIf>qJ_z<|=>s)B6a=KcEL9xA)g?*{zJD8< z8{_{Hn}n|nX!n+&;P-r9b(wA!Ev-~e*yL3`Hwe!+6I`BW&K@HSrjIq~K<=JW}bMCxWN183_ha*(zToZ(wk)!5pf=if(I{g>5D3lLz8*8K-l7ZFT-nT(2HJV zZTc7t117^Kz43}OPvhBqC~DgZFIEoO*efjb+-4(SIIL>TX(n$Bi+iPZaEvW10TFXR za;~W${tvj%CE0tcj9GUpC&+3N9i}h2bNGfys7e)YXduJZ&se{nS7AK&=jeSp$4@xSLV5h7GXRu^K z5cLMLPaBs;*VL@7F|O|xd1oO4GRf-{U6;T~7WSumNd=$H^*DRlXh z5JM&#H!uVWoIHS5QBV(Y3oO2JLD^q-3C9DUm zh}Ws$0M{jYE8&DAEXuYQ&-<*6VJdwxCCQW07L;5>_ooFbn7+Oowori5n zymh>o>)q_%Hiz1ZyvWo=4aKQALKuEvqOOEwCCZC;An+6-neHfCatS%MQHJuod*1+X-AmX90aS0*Uz^ zAT0(9HG(R&k1yOQd1U(v*8MwcZ%3EEIs&+S>!ar*{t*?6pMJ~XUL=RHJbr1WCH|gu zhlM>R;PP}FZ*-g%2A%ku{>}T~%n8HzBr&TT>iuWzvwUTMrcc3gy;t@kJ$11H09GJz z!}bnpH)QkL2jyH^(yAV{$EFLPb6kA{vtZOaBZdJOd^5_z%YCsf+IU}A!VYq$X5yt} zN&C!~yWKfDCpZ7Q8N#d()2`51iUTid%~x>X}1 zUN(>Rg4oDgY!X)V=2c&e)oBq}22Vi1$V;8`a;vBJrk#6#lRVPRz{WE#LcoEr5oL}SWR4iQfm)Ls zXEW`ojPAirq_r&SpkqAVB-*h#m_{a;uh>ueWL`2U-ud#~^9jdV#9HjAc)H8lwH^WE zGxwNmAj9ykjNvg|%S+d*q=iw#-+)Ni;@NI9_LZbxbPHa6uWmGH{eBS48S^1KnLmU^ zxX!wof6^%kyp z%5WZRhU{qcC|_8z+crqta=H%Ub5nv50hZC|W<;?i%(g+SH4dU`IcGjjmOE0x*(hJA1 z>(BwhlAOz*k}d5`0}U{j&IT_Q!5Aptd$Mb$`#q`&TfRkQxR2#w$~b2={mUtZ7?p(_ z!DHEGC7RcNTlFahpx3U!{5c%hrnz1K9PAuDBAL#8^_Efz$wF=2ZleB9>Sd~w97rKa z5O*6n0RM}!!@c&+YKU7WX4T5}<&1SYGUNGzY(I|oe16$Mef;oC$$a!pUNwTk@kBXG zFjwHyMrc4sC+QBU#n2ikza^>Gd-{lMc?sc2LU)(a@6aD}CFlG~etL1?JTxkQ0A>y6 z0$F6}5O)h*c^0dCy@4&FFtwZVj1bEPZ`)rX+E!}km6Ik0gU#(4&-9ISrrVp7AYK&} zum^C?|EK{!fUnRojL|U9fMdiGsnMb!kL9WMQ@1 zF0mo|;KmnESmaxm+b$pB{|Y|)4~3XJq&v;*<6N8}Z4^!1jMnd@>Wz{+eAR6J!NO*S&8D4;aUEv5oeXrY{ z&~pvzA{?vK%=7AVAd?r5PY0aPr7IsAMbGidi_kjFd+T<3YnKx*%wdBYZMMdX555nZ znklWz=HrEf=HmlS<6%r{NPDI3$NnGu@~-m2J>e^>YiD~8k48$Ud>O^T_ujtj@dnDlLB z6ZgC4-FztSm){*JRLxl*&;QgTG(Ms=QZ{7eOXqK&6&K^5%$#T1!Bq@E;RXRerF_7u z!>IKfh}gwPQ(-+V+!xjp8^Di0!`|ZC5!>T`dCodg8-HSZ5Sc+pjmNl2p%IT3{o<62 zW#|ifY`8_(W^_}sNu_ftr(_&Z@>)DA^y}-o4u=whlKGzgo}BQR-Gx{-KG+Lbr?t$g z)D8+4$Vn=K%3cqC?eu>{n#HCpGg&ohxudeIp$~ymrXdk9z}{dJ={`iC6wFE!ao@zB zzpyF0e-?m@UNhhtR4#tsgCZfd?sdL0W!=^ECQ2LWGufku$HOjk6NpB#X^px4i{BQH zCyxiPBgANa#Nh!Ri7pcV| z$9;|w!x&Q)yzk(d1V7@c$fYyg-5q4lfyd)Fg`S=+w=R1)?=0*qZ}n#40XS+dYC8!S z!*BMPV{?&I?{A>kDe9B|tKR$Xi2c`lENra*(|P|j#Qz%{nVI3g1Ny(%PDTPI)?a%1 zf56Y*W@G+e<^K+Tb|@=tZp>dtg9L!PngfB9nrk6#ZfOVnkzu>Ou7;Em^xM4#{B_j@ zb%n>HLD-Zrz3M_&zwxQ*zPXNCh!S|Iv6wzDJ^o1_LUmfCITbPj*%HyAA%1?chm5Yq-#-Z2 ztTix>N^oug31A1nuLYFF4TR$b0O#Ve{X-MP2?;=xl~-csv@ZuPOKvGijSkqmgAWLk zVX+=HIY|?Xh``6k*Sq+@hv^5hy0(r4-2f#;5r+C4i5(3`02|_$tw;OSboZh! zE|5*r=lj$@J<$iLSLR;t<=4}rJ+J`ysfzT-Wd+3e^7MP>HBF!x&mqTGR{LOJ zt?X?9-d#UZpNX^l1b%Hz>Cn9>R_3q*o>g$rfwbR45oB~S%Xk3ZF08BeK&!q#pI>7L z$AFF$gVPW8Z3*E#7yu}2buIzlN8&Cj&m%=?7>mHWQ&9UyAhizl4?z5vpPc|cIzE0k zd{BVjE$CQsLxb--_umwh&~<+O=bx3Zt8ZT&r~tE9y2D@$pRBCJyScgmN&)R^ng04Y zZyV`gmz$Riiv8RCKb3|*@jn|aKRZ?L9VxFj60vw+^uY?%E-t>fWCJ~uAbxFQn0WxN zSEWBXRtC1#CvQD8KQ+ngyS`hKKV4Dp_#@B&pa6%*~L{ny~k9`bS5f4 z%~w5083>NGiP z2$BHcZ1BBX)iXQtbS?n1eS?rNb->nl=+xx5_|$sRn*7??c)+zev{hT*YJiCsFdZYl%u-d%lO3@d`lYZS zKZU|JX-Z*Fxq`1gFIM2*n}fg~c&m*L-xvNrxIs&3-Xv<;b)o@FD83|W0^yktqL`L|K%jKKIeT}1hNK%pgzk-7ISIqw>ex$t^XsKdB77E_Q z7-%j26W_lA>p$`KD>VKSU%!I$Kk>nKH&D^~-7Pz9BKPa}1DsjZD@uPq0F}mzROPp} z17Z;BZ1@e{zYh50-_!L7-j7xP39OH+{^ifWxa><%MV`rrP%`!m9~55y0jytvS3?59 zI_Mwwy9o6T-mk>*>EGl1@}Kzf@0t4g6>XpX49eBM1X9Tx_z;Rm-(ffm%6R{kC;ydC z;QdBv8J~oQFJG=s!R>xO99;Z8p}o5Hj*1<=Fxay<`nS20KdXfjsQmd>}^?w7=hdRo+hoCj= z-p$^ks~cygM(aQF|FPaMkXscD(ptxNN9N6-qd%w`oGG~gdb+hBTwqk3D>aPw<{3wN z&b|nArDm>x&va`i)2CfT*m|=@YkqiP>6AVgQ6WvfghVdasH8g)9uM0FwC3pMBaVOeKTp)qP7DgdLC}5f z+klU8xq$C6&V07D7!(4-DMa9VyCEflK@Pf~Sfyn(F;qNA#v@$n(rHuG;(R^0J-+#- z(&If!w_NB}>8v+sO4UL;js>i?dhLKQH^i8#oAXBw{!*wT87@b7izl-bfI|kk0O~ky zn6!t@5nF#T@{ad(Rj7YPzYzy2lUHO1R0B?!?$5cwqq$A{-H`-=>pFqYS$>%?W)fM3 zuPb(qu`HQTcLUO}>OA;f`RyAg@-9+n_fDlr*+3EEX`AF}cAYzZ=cmakL-r$x@;JED zOj=QYZA|@Ur$Ed$CnK-#%Mk=uPUPvIdrjPplRZ?>=4b%aO47@#NZHQ!%|^+Lf5v>G zR#x5Jn%%qw&-~||y>gfLj_b|H{WNd1 z+@L$xS6}i2X*vVjD0)}3I{N8qkYP4Eo>t0?=o;$Mryy1pH~^VuPw>@-ILvW`L@W}l zPo(`i%m{;SCo_%AUS03miVou4GQSpZKjk%o#icu*6W#I*#R0rcC?hcFg-C8YK3s~= zD;Dc`U%t$^vG!yM8HOvIyad{9+0JzUSbDcaJ%;G4num>24|$XI%hNl^@}Cf$LoQ4P zFq6Z~_)I4;rJX_N)^>qNUb?kfGT&;V3xFKnq?<8> z*IXgOv}%zAn?O+5=qNhJFbG}Or;0wMuJakV1XlzBWJ4syhGS(uf`s{osLKMxe%Xm$ z^#m181dkPkmTk@>)4Z|Hacj@eMo~#(uoLy*HWUdlDY}ql_hE!=S6R77*#Z(Ug#bY#AcoXF!z3 zuSS&bIr9EG1z^_Ey3yJ*Q$QVfLme*Uhflnk=2RMNo7v?bg)8gVTtCP}pMw=#`@^2_ z&&L2O_nOHQJ2&upV0!1ni}s&&H82XS(7pl8ZFx^qlSa|NcT!w@j6#}ja?N))?-9Me z0SkE@j4S+xvqd_(6%)3kzGGsIX)e%8OUPCaOt-mD!G1 zk&5uCuP7wG8n(T5w{x$7Mn|7WgPAEZ-d3cPd?DZ-Y)q#;&l@i?(jp=*UeYenJg&O-U|g`op4hOVmk$L>;f%(>&05|JekFwlMuRy>dUK zFp7Zjh8-pvJ<#%l!C$XQx!taHs~j842;NqJ5X5<#DuJr}RDGk6 z>qS3z3Z?y!>SOL9Lh~D=W_8OSWs62u`H`>$a)`D0iw?7f!C8I>A94oDAy@mC4R40R z>U2B|+G3yHp5+VWxLRxhYu->%MarWz5TW9+ zs!pBIGgQMtXOiV31F*J$owr3m9$R$SgJV@ye%6$G_J6J=y^dvg7lNC zocVVEQUFcxkUcpiwB6*oIXzjPe?IhQidhzM6jtPp*w>>4fwCw4)sog1t*}+ zBuTMhWtL-W5tWrUSPCRVsQ2q8H5wc=&e)EIE|+Ve;Ucv-G$HY|Z4p^KA{Z;J>UhqY zfe*KXrXS&!S;DU!)^*L|bKB1BB>9(af19&8%ObQ!Q(vvmwpGrDGm{B^*J2&WWzqOJ zfh{R-?B!vXUt=*i;jw^^99867T&3jV&`_HE9{==StvMG~KD5oDTzW zw(Y}V)Z?`R-FWB(Cnz^RStRw3p=<3tg_yAMOIX+KSfWDj*U3n=#U^#405{L zQABVmyne=+`bvMQwa*p`^+Df|@+Lc{L|~v((%KGv*YI9mW_-yzlmh=Wr-H{-j<-Ob*WYXP$B7%(PrE2$_5C?qEkLz6uZ%%8kjUPZb??&o zp#A*@eLisR21S1>p3{VXvMeufcf<&eqU1Ld16OWCgqn*HSRv;cmf2BitMjXa#}a0s z%Z{j4eXqvPpU*e?1@5pE^ztT4gzC2-S0}H}FS?gBX_=m-$F4N42>WbaAHKH^;>v?i zYby1JST8SV0nBKqEAfWT63tlAo`n@VKb(h$aJP){tzW=nKbI_8;}XvVLnGvFWTeD= zmxN6!cxXGx!D*D$ws~lKL5|1BYejIm$)O&uQC_8gGT~0>t{}BjC%OaYV@EhIi>Lc} zUEp=?0|J%Qt4H1BN;#`+QaED?c#1xmmsYFj^N@x#h5z`}3avAo1l-$o4UA6ba>OQ< zcJf-4pyh5TGg;bIO{;cIbDvZ(O&|4`8#31CAP|*$0b67;dlo`-R5KfdMerk4Ji-{u^X&79q5N0}_UYG~YGu*7&kocWewu1om z*WoVo&ZN<=ZFvjGSu!jbhdyCgelh`lV=GM_)j&!WSP1nkEGg6E0-PAi%pyt?{0=RE z&zw_f{nIuWvP!c^O`N*z**Dxfb-{U+!&r8t5I&rvBLB70l0K^Tb5H>rDWoX~KoOMz zPXA#6SrC@Qq3xk2S(lp(ntO1FkgFP;t*)0Y8JrKlN@_$bdSCVFJS`^AJ>P}cYW|J; zSahyDxUz3fQaa*2Gj{SAhXCy&4O})XWRxqZca?}p*Vw;-QtSG8JuMO!-;7!8 zw&t)o$N@48if3Xd4~o|T+^BlpM}Jt_fRgoFq1Eodj5t5;vCD@n-ZFYd9~4_nDOxwg z71{;!PnH9hCt=_ZDT_ZkLpe>kh$EnMMDu+>xEiUJq-*aCI+Cq`g{WI8ckLAO^}eL2 z#nsE5Gfw)B1gFK9m5$h$0;YlsX4$G~@h)QB1P{cs&?KMNE+y-_LPU?ou!@$h*$IWlJlaozbQY=N zBs}cSIpu*Y9V(CGlepU&?YM@MB$<_Zu6edw#4v5#lhRMPQz|(%*@#`9xsYjhAG`a) ze;XP!Dh9}g1VWEDiD#idAgfm1(@GB>r|`!Jjaty*cBl+*unjlidq_cS*A5yCe_km9 z2S7hl=!M~Ce5LG1e$KkBo9Qf_>T9Dx(RkJNyR^$`hw!U9W8JZqXZGjDveyyE8_7sR z(48*f4|!wGhYN{o7oeTNgns3t_2HX3oa5K}UE}p+lx%_csbe5Lze{;kU7eO{cR{@Z zYdP!T$}_{A;-gp0CO1ZbI3tThc_&GS0*ChoRjDBFu^sZLR!&+{MAoI;X(2lGW-=3W7QjP6-A}F$-NdZgh z;r1|ck>9($=yh#r0m^y=#>(nxsP`@)#@>pj5$hcer0hC*#Pl6zL7t759jd<=`3$W& zQ>6T8f8zNJ^Ap;po*V?T6VFp5O27JbtxuoC#2-`mV7r)BIQxT%#DBG87$7K>rewMR zzRXaC1vd%}%Cj$?f<`?EKlvy}K?Dh{KA7NA zq}%lZwN|7)#J@sDsBmoXF^2JW;mVCs=R_0>x;8{jFC@PGs9{cXx85rSoMvoMbqLA#$R-P^{~BdAY|yb9yOUL`k{4Si!j=_PW6Sbo2ys zO^4FYQqD_f&F{YP(F(@t5(j5-BW9b2Pvp~}r4XU(NL~JQm;#4(hi}4F{yUy4VG|@J0}O2kfRvZ&3_0wZys{`1L6R$KYdUwixC3<>r;fR-wv+odPYfoAHj8{q&sI40m-bkr zZ-dVIfz_I5YPJwD(uuwfNIft0k|k{7Z7{PHv^dQ;iM?70Ec55RL;4Fq|*C z!m^m_vf?vc7cN~8l77T<%&%Vv=4S3Zbj22M|Js8EG|lAa)y1x6Q{8zTDgzsHZEoR+gQLV@fP-^ zY68bbGA3YJUVwJiYmnZM(2)A{1|zaPV%b|78Ncqa`$iz}1UD1YbhhWT6Vv>U zp)~P{y3UM&M|9_uyg2k>nNoNCIXCk{E{jy^bvQ@62v1c7 zA3yDyRLmmpSoMU^YE8(R4c}YT;v$HY2=nDKS&G0~d2Z5{YtbIuFHhtSHzpL7_EL7a zm<@{7FSJ`T=Ci^g3KX4uiD`2g>TKInoB7O;rD-vlv_yh^#MK9Tw*px~c+eV_Mb`?eqbw_N z@?#~XO-ahnc-Xc>+q~ZzH&o{^FQG+7!@o9p@l4U`<_@y) zW5Q#&xRQm=^e-MwYpb}_*~S0vsm)-nd@SfaLt|F=GkZ}!t8S`PDw$*7xF$$E%Vvbd zFr-im6Hs!@U#-^hagZU$><$GqHZpy>uxCz`Gp878sG+9x;N&0|>Io%_kZfyK7`ZZj zwL#0^MS&7JJjc?C3upj|YZE8a-xy`TyOCS;c;Z>_{powlClbTR zn&`_#-oMeZDw}iR7X0#|Q>l``9pPId{^|}*tuhNv^g?NIv!Yh}l0yVS-|I^V{78OZ zus9QRBcUhw8srZ=+1S%yqcd>=O?Fi2zN+=K%oxeJ4EMQsT=0D>s=UmRHXjYYRIBdR zDRvuX7}2a0EyuZH+_Q(DSsC6xozk8-~SkTuFKzC&?BduLgjdT=c;#Z@RjdfZVX z&CGcCmFHuu^TMF{YZP(iOkB;LC3{94n}_V?p#0TV^t~r=2YY3`nwUq6-l=WAE^LEf z;ux`?%kh-usMa29A{K|fgfNoebj5T&Lh=bJxFWM1f*&EAtQ-|Yx`ush7K5;gW3Hhm z65qm#nmXfBypCw^rj(}*T1BVLQ`-l@30i}0HQ7v;^Q9r)D%o!BlJLQ%EGUb;>0@-Z z>z05j9Mvp;gKKm&yhrz_eoi#_m&OdB&KvIzQV_{8vK^lKYdC6xMW9BXs-ZDUr}+Q-*l z{;V9CWW@NMojw(CmGo+K4S{S61DPKMJi@N>rV{BxeLc^H43X~Umbd2G$SpLPB*5G| zFuu{Cr%y^wHu!@~88D-dJjpw=E^#cn0>rQ7#N{yDRLgW$I!?=Fe(B=!w5<>EnqG%y zM3dh^8h@2)PSUj4DV25DST_)RgbF_p4AZQjaG&Z~C2m9uCt2DISF;7*_v$Yu)i%N; zkf~4eh--2=z!<~+QCV|$cVYBnQo^t|9Lk~Ocd>-V>3O?)E zJSvrNbx)Z0w>c*fCqmQ~?r|Y7gWhA(u9ymwfLoCt3PN+Evo?5`;0^S_>L-j=q+c}{ zqzmbvd@-<_NiNgFPj5hmL_wrKqab~%=zf|w?GD@fc{JS_V>)pk{}kVx5nBDz-BDFh z_<}JYg$T-BvfcG=z4s^y@7qu7uB(^A;M%Ni4Z>XnSBOam?^81i1_=i!q+kOq8=00 z40G$ZO4mx0*E57w@;5srv%2JQCv6$g!iOWYbfXVvrl%M=%1nG+_p1`{QYthJ_gfwj z;WnVsGC;H0ig%iSapkt0X!a_BD29`@xj7JUu54YlO5BBEDGLx@$gHtPKt*|Bu4@t8 zCv_-)mLwt)R`QocY_A_$7oTVNsbpjo>y5R}n2XO!w#R%`bGikn^Yto4q}C&C49-YW z-rgf0sii}Cj@UkXHMfB(X1f9OA3TuJ4jKm#TCg-)PQ<$n9%E4~O|{y92UVv|T{sc;OiX?{s-r?u*MhbM$!-3*}AP z78`5(k3g)OtIfpB2{Q8d1nXANeOa_sEj<4mikpT}0$Rm5A}$2Pz!`?v411?oF7fIb zv~YleHw{`6`ZGP!c_mM8R~DEPaUp7|;-~Qn6wQt(Wfi3%v4&HcA63x(n#QPg77-e8 zkrT*+(1P&R7A>OPvsq8N%%FIjBQXxnPY7*v<5uD)-tDMN#GMAqb;6HQ>!V8&b5>V| zHZdvYVb0e#{=!?H=6tjFbhl-1#zRnypbtz1B?OD>xkq|}CZ|`|istllaqT)qLAie2 zHX|pr@EdCc^?a|F_o8aSXn+ZrQarAC7_sV&mUceEYFF?g9;Db?Bt?hlvw(cnJe=6T z-(}Lh)GGt}rusX*&x!v0A)ccjnt!y?5Ad#}$rVIBMG7}(NDd2epJPW-GsDL1@13<% zR-<;SN@f>#P_az^grae34IZ&)mJILPav}Mx#(-6GCl=CSA{Ie}8A4UiVbVf~hqb3k4N+Zpz)`w;{|#?~)oG zoFSGI#RU6K+Y9L{w5_7xMr$<4AvUb><0O?5g-I@hI z9M2@_x+!ZYK=`oSg*>Uzk#Z(8Rm`sm^X94V2!&>T?l`LK5j*8Ts^IacUglY_AR$D{ z1lK?xw=wZaQ*IvYGJjZ$u6RM_@tYz5ew!XrWwNq1b(wwhsdD`1wZ3OHgJ~AGdTe6c z!!klebYbgH8lq?=un?-E3SiG1=`-(_@x1!j5-ex13EN)_5^Ed=F$ zBqLfhuqtCGm}~d%8rZQgbm7V-iZI%kx~Kzb+VR~5YKhDNdg(1nL_-QI#)pd9_?0G2 z6g^vYt8%hpE`FZhL>>^tB*T81Hc~#cZ7Vjz1PmsXn)&X`G!_>`7*%$7I)D5%p>cSx zW5OW2pBO>_W0#FoY13z!)v%Ch2B&xgojiB|Ad&JNnFLsRTO+gT8$Moq)>tZ&U5V{v zeEhIMP3LVMRn*az9Ephztu0(NYVu$pTxw|M(EL*Qc($+BqJLjhfQPgAQ zoc!>7toedg$Xg({SJE4Yy^%XizwP%hf5}r>gR4gqIrdBNLhvJJqT@aB$2XyrWo&qk z%4ZX_(~dHLoht=$FDf}cZbm2!>_jo&!)^vEuh~(5Bf>C}L773kvQ{|$iIvC-#{M8v zPqR#u{^V4JS4XQJ39MjuzgqDY!Y9E)?A2C_fn0(dmUFJId#?LAZx)#AT^rT`RJG-J zSKeQ#1bU&~SI5UW?TM6OU)#)+hIJFq@u`(MYIf^}@h+`J9L18Mti34T-soR^rCCehspI2XV1)+W$8wg-jAgfp8Kjt_R z!5P2aGWCtDG-m#QptssqkLfTkbFZ^3$SC^|6xA0hg~wOT%{54HQiH6GBy-~q`-M4b zD8e3E8X0DECBaQ&R;o{Yl=^W`;&&*d*1iV2SBTNJM?afiC9KBNibV4#wWC&+)4BUP zSK86I6Pb=E;EU$07o4sh#dgfnsJp$*`JG2)@N~Ek?6S$#MxyQ+c`$JM6(5sTNVc_) zk^^XYtge)0cFfDaNvZ0IxvqUe^e?Fviteo|+;&`r+AHBN zi)pryHY@oAV|JFVxzG-q?3{+9j#Qr*HPCn=Kw)P&_y@B%jHWKnGh;(bFnwg+*ows- z{DmUWT2Qk8Tu&Z2Mf%vpofkMRDi7wK@xiO)d}YN#XT9y`4EAVl0#7`4%Zs4TH{pul z-Lf}&wGy-K&>P5`)h~({Jq6)7MttK{_{n~Px52+bUfDq)9kPG5kRm0mg%oEj+eoW5 z>plzhtms|P=$&pt5~pkr_^!?Eu|v(52YXOF%%MlxYMIt$!j-bbFLj= z;YvfN{G{t@O5ud)K+pjGn@0luvNjOkuX-eGpy_u#4!()ihsgdJMQ1mOnvHZtp#+Fa1)q zRAE2xRS0TwQ)qc<6mGZ<%Q28F4*La&{h* zq(qOyd(bCalQk+(@!os1)QhLaEHs|x(*dg7MOtE0o-qXt{#Z6)GJNcH z!@BZ`s^z*DueBg(_K=W=e>j}fd?6FpWrO39?K&_#7qPyrc9pdxfp;*!m-R5GOIYo# zq9=;P+a8pig+E+7n8~7ly&CwSyE8!n?`P;`KmS{QEo(ty&PxPSYBDkS8uXHfawLYZ zN0#td>JDXE-mTfb9FaB&j3`m$L+w=M&qMn!vXWV29OxkY*{rlbvDnj&*YxGQo8e0M zaqKiPDPeLqf~+NqS;E+Vx-HbU{c%>Lr^E{AT zYf}xAU^z(u982cB@*H&Gd!DR_qWNQbDQj?fP>Qp!?I0jXpVS1nQ})nAq9{F7p@hGJ z1#6Ggpxm0Z%rV&DmS+@{Z{?MD1mAg*0}==~Pl&ym4^TJwrw3}QqT$CU^zeMaHRGO6 zei?gWQ|}M&aqzUUIBdpPNv8&9r9<_;cD#KduK-J_3|J^%wAPv42NIo(N3(lit~Fm3 z*`?Y!dZ1gMvAeV6VwA2c*6d_`6tDachIUiAa{A_-PV^&7oDu={bwUv6X;MUSt@y#K zy|h_T&pNj5LLP+L+xZLfu<9L5F7d3d z4D~~~G2$~f?+#I(6DTR%H=r-A8}04IG*PxW_bZ3OL1*>5)%c1kNV zINDV=RNBkmN=$4-sw`3cf@Y}?=q=gmZ7VecjK)N$W61iQBP~7n1lULFQzcAo>FS1% zU(i3gfOZ6WmU!*8sJ-$Jf^;$d9-$EKwFtfz8=YR;TOP-x2fvfG@gYQFpBDc#Ta=ob zXGHGW&x{ZmSQi9`1z}MXUV73*GLu3djedJWOGN0&)0kU5p|+m7{Oy_9tt39AFW^Zd z?70ius1qGIciPN;<2-h-iF6^xFk@B!hU=21D6{+o9@35PhKLxTpFGqa!6s2$p(oV0 z)|E2{q$%gx%z+D16K;%~dx+YJlhUw#H9Ob)N}QZQ(YnfH+L^FCE({NdAuZPZf=zEz z#o0=MO^hy1z9K@jV0@v4nuG^OkZUsL3(e`Cc))jZn8~Ca(Ww-V$bh7N5ksiN!a-Ac z+_pHc+&WLXxy%(yMm`2pE@BL zz92mOb~41ct~B8UCO)N+3d1p$o)e?mm)2Oox=yMaVNoxk+}{WfZnN4e)$FLRsD!vN z2Rx&rdo-)?cC^z66;ar6Eba>uD9@jBJ8rZm)4wS(7-^EY*lgx}44WK@GTwjUqW562%b7Le$`Zzhue|-iNl%5Z)%wsU zvFtyd?#FLc!kT_k_*3w9>_81ynDu5%`6g#c#Hd(ONWMT}J_)4Z5{%*j%`?5SZ(tKz zNm|V%9j%VRBh5-kODP^)oWwSN7A$uzB=umK$H)}3%PC~Dfyr=&=AZeqg$X>}eDn*Joe|c&pnTQk_v#A*D})Fa zhh^SFu&oOdNv?0juzcz%-)1AN1T>&Yy?e4}gV||qWTlZR$xXear`lJd^^vw@TyYYYhwUp{C$nsKdn^Y!h`^FN~w{Yl(v(l~tjBgUr>j&QKw#qdl z`5jVSUz`XJZKt14bsl45RJwQuV(}^SUuLmuMtBB{B-uY`(|M*+$z;5 zRq>Z2cetpG$+<&(zh}6D!JMM!V|+O`-5-_?r~%VG)edP&$*~kLUV0FNC>08GR@V^T2Yfytu;P)Ke=WFG@F8N`1d)wHvEP=3=*2k2`>nXjr8lOXu z7AmxC^-84%s{=0kochUyF`s*d@J1icA&a*>lg)=)o{5AmX)LJ5YuAj^|lk_4HMQ3J9-u^^!tf>)R zVk{`{VS!1YGC1(doq!C@rgkhx-zlSY6KOiclIrtV@T65l07h2&pWO_?lJ4@`@hB$_u7Ykzg-91y+OZd$##yynQ!lYs%NpO5xjQ% z#L>!=;raH4jJ5psk;0BB@?Q`k{-84aK|*2x&;$O93c>hy+Bv3wFq0S={!@O$AIzeE z5_rm!HSCtzkq6EnQQSlgt72Zx)&rvNh)9p}1B_DbmJj$AyOpYmDOO~jjwXglJfI4p zW+ak9!_7yYSKfX*T^n40^*fbSG$%$Y&}qsqK;?)Lg=6}ZVIkme4sZ#40UOg{b7pad z(7Le32w?8>+ZU(2!Qp@esoe${{0MfPSDS}U4-u?FmjX{#{d%YOOaf`a0?miTkJ03a z;~aC))rJ$El&H6bl}ro4Kas&@2b6Iwk8A1AH1 zOK9V(*3zuCip@7)Y;42&$$;i5!SzET+7!+eEuY`6V8^b!iyB*esN|#b%jtLMGcc=* zuss`PO0G_K&LDU}7bb@YSl{En177UGPh&?Oy%!_T`+7fy6FKlh?3f1eqTo+=eCN>7 z7Jk-lb*72nd&8f1(Wop{-mxk1FSlpzXCx=KmmoM6>gAzh@r0A!*6w!h5O;HC_u#Zr z9A1&>IQUF%A<*up+9goTXF$Mf*hn>1q5vkH>Z@FZ>Hzn-5d6VHem2)|4to7+`GQL- z)?af#r-QO7Fa*|%NoBkgvCY7`M`u`{*_TDt;$npZ+fq;#W0?`VLNS5mv!R+&XuRT_ z?0gJ%n%n0IT0`qDgL0euT7f$9`ESnfT1YC9_Ym$H19uu(Pai>f^;>Oro2Pa*y?zl( z;IJEtCTvQT3M_W{ZhEGlKRbh;brLq52+XockWbh=TWd*L`V?W>N72MOiV=k9w*1`UcUM2b~L%~S}9Khcxm3u{7qMJ9RR zU?#A}$M-oEhGRn!7yykiZ&Y)u-bg)kG$bv-2@^D_sVQK_$4mQguvpgPuNbHt%_ zS1)Z5QRNn#Hq9FT23C$I!$G7o(T`ZZ5Cr6nf^F38m4mqs)FIYJTjvm!l;}PaaGr?# z^a4LRFL~AO`9>oUqPaphXWFhr*u+9$>F|tY#xO7#u{qyHd2A)rdQd1=N}#fJ`2+S1 zKxU2od(Ae+gBC?taP+2q5hv*sFVWyQ$Z_|H-gM51?}i8IO{zh{3pGut$y_$V#Crk1 zV+SHFZuhva8n=(=Frjo~;6~0cp!jM4^q_6?L|t^f2MMi}<|BnR1JfJ_%a7O3Q64>) zA1`AiL_zzMVCMsWVgp2&eVZv_N~I9CHY65w!44!-D{Yma%BIT9Sn!F za1(Rq`5fk+@UseYUsDS{NjITO6bvuA_$jCDjs=Q>pQ{X^v57PFoUVov897IBytoCT zAJ_(EbV`Yx2z5q~3119-5d8*lI1UJC=kE=I2J}WIW1s5i`96KRFed(n28qkK(9Vb&C`! z%3{0UlRa1(<5e(L7HgZWGrB#o^V|0ZUpS5|LEW&{)MAQ#6&Fdee-)GnKfs0Y*7&Y~ zrrC7C0IcIlgr8HOQfl!8A3$^ICbHCqvQs{`IQO;udGVC!vTlKJdK@C zLF`S!t(i3uB3nq*sPzx47WWasudQ}Td>+`M{w{x=+2cHf;J%8-Py$B$+UjN6xKCqNG5OhgDI?5 zqk5pte55No4J*QQ@&K`XlMX;`)=Af$p>lg*>rC(%>xZ{IHBKOYwy2j!<_I z{`Ig^ACezXnMG)#L$soB-fa>RltS6rz(HVn!HV*%tp?xKAW=pAxuHI)^p>zyp&5w>5DF zmaIlc159c8PJaCB3pFc%k%kF)i;RwmrM-fSwZXeIe>LUrLpcLGD+gOW1G_)l_FY=n z)IiVvT|{SKGueR6{6oY))cVe-$BM z1&t5=uR8y|H3H@tYwHvynE(KWX+|bFGBHI+<-r|_T?PfZ4}1tv z@eE2L<9pN602vWlrO8S<05h|a_0Rz{DBlO45d@ciCFMUDd^n8dH?7&U>5>u?(dz6Py2u9 zx7|-4KE-}m_*X*z&EVza{AaWOrts3UFth!a(gWUA_CF!^GW}&R|IIuv_-HRJ<5Uin&6> z`-$^l)ehd7y9wesa2}Y7K@Qe&@5vI99aUQ%!uFeWr;jxXA5v2`1{L@@`1$4+=>bs* zbxso+Rh!KivdGck4c^b{bZ-hB*I)Pj^}ul^z$^H>@2Xx2LqoCzsyTdFuZGB=WAVLJ z*s9PIjkYk7slj_b1a)`)s%1Oj%=o4QmPV;+7sD@zI@f0C%6NRF2|<-45Qz+mI9k(n zdf{){5J;X*-Ve&*Pf0=c0yADr(cHZQle*r8n0~1{uHXTQTr%IM+KI7#&8~5EmzbbS zU;qI!g4a!+&hqKww}<;*-4dss9Q7vV#H;#hudDLWB+GhIp9 z6)0Kl$JQ8-Hjg31?a-SyB=Ll@hzx3hZ2g8CE2snyl{nu z`GKEcFgH6`f>P%lJ13rUBy8Sj?694kcbf%TSU<`@63y@_?xy8OOgp`g-KT+gbaM~L zBGCSDRN_8fyx^?|gh(4uR8_}&+$}`;nn#cTp3VMrR*dAKpEFYz5&3)<;U{+um=#?#N5(t*sZ-`-H-!2;Sd|Owksp;CxQ)UiNa(WLYDEI z)Z1VL!JJwCnRpMPQIKt!%ea)OBPS;oWM5)<5b~7q>9JynjJ3S(r`d@Rf?I|3Zx|^L z(|6l6neDm@GEz{XbF9WPOqJhH>76M>2#4OQFzA zA@JTXR@KwMv5QGbP01bUiqOTBUKt@3S4EN7Et@$)ONBU;HKyQ>+EcDXCnaL|Zpn6o zE?cfdnY-Ytlv=<2v?o1tcBsr5R==6swLoD!o<(=x-t4cTzejZ*84Qhts3cGwIR>JXmf3UXio2%0a@uPXt38&qc8qHb`P?)p1 zWWc|L!M1gUFm3>@+}3>w;lVY1{EhCHIhB)i4#zklaSnEj?623VT}oB@f{@PJXuv7l z=_I_a8M*TOf{03R{B3Z2UqFJxX{4eOMA9wf0@r77QND{kR}}xxfgYbDVR?~Lsy}I{ z*OpPjsT0Y>bmfzqT8ZI5&N>o%-Ot1x zY*xZ%(0nUl$Qv-1TrZ~_U7E3s6N?Gbxf!wI1$I(=tm87d3cKn++;w~EtOh?jGZuw;N_e@G zU%zm+)fEebaj%U1O_m;*==Y{ICCN!kAcZdRVqMoW?gaU@D}38kl(nWMz`uiOSs1z_ zDrbif|6S-8SviGX z)VKsHhNhg7`H52*%OIoQSzb_aKkM93UbSAxN6wrY^_Ji;q}38P6^G`WtZ_3>!arRg z1%)po+%hO?-`CjT&H1!ky^^ht;IvH=imr~gmIvpPHhy8M?}hH-5@5mYGu8>AvI^9- zKMnki8glv8P(XH*kQ#zDOdD_1AE=DM*$R+rOFnNje!W1XLA9{P_VocVY%13<(*reV z?hiDX7V){0rY8~8!y9lG4?MH^_>;Vru=}PMdUmDS96Jc`u@D;sz1?L`L_f6r#!f9G zgzEe9e6M}EnG!InlIENC0(M)sqOW(ajkzsYpxCyXPtdOp3GU*Lt+RHzXox{Mtx=w< ztoZFF_Op2z%w_QmJwjI&I@|}p(O)`v$1CilM)+BWtxF}gY{}~{-G6O7w)lugCefr8 zU>XT!5UznSRYPKaLu!J-_zY`jxZ#%Zrj8~mlgkZ_kDN-w=GcAKw@mJ)-}ZI&^=K6P zf$Or`WixUd_QHQBD!qL4xwk&0=L#fNTg-9$JV#RA+*x2@wyq<*s;v6$ekt2nc}#tU zdKY&ZlI~NOwdO&SYi+5mXae<}EX)}k;l_ytx8?*yg(Q*-bi+0CTcX8o!XX}s<}8)6 zMysst`eDm={<%|FrbD4EMH#msk<0C1qO?kMn%@C!542Ab`|K4FIRH!-?8gr57rdhWn8bhKSOETB`T{zr|A9C4 z9zEwz$M*l=u)q87%+&Pw^nmvzwf9grKo9dTxB1;M2Kw*xOu)16*a3L<9}dnRH~j6< z>ev~)SC3DtD66O>twJebqHAEQV{c+*2@Kt1_gCm10RuZd1514!OZ)e0?eNv;0e|>m z8t-)qS_1txpu-J}@S|^FX>VfhLM`$S2kqUZr2~4lR*IG;e>iybfWI)opU&=kiT`Ze zKd{dKa2o%A0vGf^3AHk1+ zcLE!1pko2;oUS1wY1W62(DqDu_u%SmJjw*QcN@#kZM=*J%&Uxyg{lYlc)yl>0j(jE zoROh};d!Fjp44ghaSUk4LX_|J`u8O>X_<)5s9J8~ZL}*1 z`d;W-YJAbCEW$*{)HI^lqwErc?7Vr|c7RZws<($bnTYXKhl&MQjWEaf5&}_?&^rMO z!H8n1BE$Pt7-vvsv+?-)1_l1JOZQIU81h05#SPjh0f(J`VIF_<#=E94FfzTLe*gOC z(EK$s@V?K!_V{eTbsSoL9cvK-6C-1L;MxihchmkkHQVD;GXU2a-hs4{9X>FS)cc^l zo5lO6QPZ;m-eb4{Q92zn!yos(V{hOULOK>E=0J`4?@1s3ebxu+&L3w@%z;`!&-BOh z<$b1q|M5M_+rNKU{QEd>z;amM9|wOedi+x#|Dj2L)cKe6|6Y^+c)b0+Xh4q-pl4(F zr(TsSdsxdWqH!Oz=fl2=$r+nK-LzjAFH5QNsX?PWUmC{=o6a@VVa9!8_CX5+VNUfO zE)#bc00RM2|IXbHZ*MdptNu$86>C=Zh5tS(K|@Aa1HE9RYWx>Z(ivRUiRsF>dt%_b zz_-&hr}J+IPF!cr2dmGvAdnwwwLb_{nXj(XI%S+=eL!%OZ*`D=I4h5lMgiA;K*67L zzE2EdZr>tLB?>$Ea3NVkc2Z{y-Rpi$u z(?EGxm;nl1^8Q+;o?ITHUI934HqP|aTz4(TaGc7z(lzSAxUh?(A`k+m8S7_M zl2NnRANzB4Y4PURY8l#8fO{C?lQ>G$YQN6x8`JOey?*z2{q9(AR)(b^b?{UTJ+hGQ zAgA2Xd(_LO;#)H*n~$Om2i9b-Vj$=f&980hV@!oWNqOvFQV<4KPy%;rqmvk|=j>^+ zVFrJLqs}!#Cgpj3`m>?UU8`PhrJ;$v#&S=<)$*3YVoU6e8Z1oJ3v1CuOUfkAOvP-D z^QGWz>386s;M_?G2{+4c!(U~k*HcWA9%f4VMYLLP52$8cs}crt%czuRw&)(oZg2g9 zw-qk2MU`L(?%QxL<(G)V&AEnQ_|VdyRAqPo7+Z=cCIR#2Wsc=|rFBY)wADEpH$T2# z`Zn>4WIQDPQoF?LqWv8E`bm3J*Ev*E9~|AUR)Vun0EW>MTzuF_dC5){44KPr58C^Z zu%y2SUkVJ_#IMtYx{H8^959!VCiFzgauVM zQb!s+Hccn$!^J>bEH{+Z%#fWmIG48`cP^=nC+Q~4Xwfw*AIiY-)&L(DPIJ=wYp|5YVl-tPU6ln@DL;Y9FDPYf}CzX|MDjTt;CMl z*|_|G+vE<|vX|V*Kr3|ZbMxQvrL&1nCf&3&{rZAz0v}pxsX87r)G?$9=;jUG64qdUR+b;c3Lb`eHP*-VUX3@NJw~}gQaaA$TdDb7OXzHP1 zF-A>Us!XM&q}b70-&9uDG$`97qGFtMwf~)ZzijlTPNi@bC65;cLF`*7T6_xO;AgnL z6Y8BdBm#T@2qZ%a{;3cy34{aFqmX0*fMqTJ2MJ*AB-J>^EW5Bw;f{=fV^3+9)v@6* zq7|=dPfeG!OS+z#!)&lx8#O)mSd=vAuXzJ<+3KfL@6l)zN@mw z90^qG665r2v(G~1EY>oRc$3FTkM#0M7pK)XJbtrGTl+#rX`!nU3XFl1cNSX!s8Lue zXiEu#bKCuXG<3M$1jY=Sfw?B;WSo%|^0g)+G}G7~y_Q`iWZRIKgJySX*zNm|?K?Ws z`{lO%C(bc$iICCn#Ba2h=WEn7vAn&F-!ve7%{cyhCMfWuhye zdD7F(7LZ{lD{|z}ddO^o#tzm2vG@&2+XKErS4?B%4aG~t#zsAKX*1MWx7*?(lVi%< z_?$N?vp8GsOg7-gXRZx$TIA+Tjsi2%VpvN#WnQX&3VqnZAUHSMu3{(zQ4mmOh&3s- zMIswEuCF2fS&uojQ)eu(6xFup)fq*a?$Ppmj1acuE;WdVId#o1a~tCbdkmWL3+^!~+IL_B->#hH-pOtZeFd z#l)b0Kb)6c>-6gUkygxSqOY50p93<}R>?fox?wHY3@HFZo?Rs1A~Bx8mp2iR1v!DP ziK@oAhB^KxWE&;0Of`5GbmfvdRY*%SJ5t9szrl-yIx?`@XV&Fd#XYgwncX@worlH1 zkVb|=hPr#>jx*3g0&a-$CGc5*!SquG?%RGTyT*G@Q!>`iO)T`!{RwlU6dU(v@PW0M)xsvsU-L7 z3n!&&FdTh`@hhjWUdix;EiHG0Y{3tXlDcOzdy6MuPV<$ngW*6!^D|>I%7Q2reu=I% z3wl(iMyDs~$;yJZ)*f!E>`4rB$X>LVQ7^v24QO0c)?wc9!St=nc?7-bTMJSJ?~fX$ zlOHoSFoY-RX8?g!^fW|tBLETvLlNzg$jhLT_yuvc!t+Mqv;Esg(QFp1&llg|n&S~& zk4nmBWgX*NFrdjXsU(j~w2#C<22R*CvpVHjGsQSS?5fM`g=e>AEA|tPl0K}Lp^Iy| zF3cEIG(*d6dtN^uDzko*I!k3WUN*6fs|ewf)ckVT*lc+PXAICtUu9|@sgu74T+>4) zby~2(KVJ|$_!CZweJ5Xw?(jDoa)E6EXWx{ii&chc%8tv%C8wQY;&YCw2>y}3JN^6> zP+WR>1KPpjlL+}dh^()lvWe zN+!oEFkyA9%Zhu><)AjM=lt{4H&kPst4vhAy#k|;GScM%sgTb9+ri!40t9!5g9n1U zySuvwNN|TB0fGjX;O_2DaM$3$2@c=sbhQ8_Rs}o#Qv{JpOxfDkYGjouW{Yp;@B6QLv>11u>CTT-j)5o$ghhu%^rQtgWG2 zs>cK=VU-z>q(kzq)ze}OJes7uvY$CEz(&_6G|HW>3NrHiDXOXaJ;k_4@p)F?c%S&c zm~Z9Kp0AxDeg~B!kYQP7Yqz~R`Cfn^g;$m3NZ8r7L7j!4o|pdaj;V6;0-OG&(o3aT z83C)TX<0hKV$$#$#!&qx)Sl6f75{)`M#zr7N5hLCssgV|kDQm)u;x#QV&j%c3A)ytY+>jsD;t#c#Nv3f0EO6i26wi`swB*N+lZLg^%$Id zG)Z3Zu$56Z-hBR4RFTvA3$(bj7IYDq$rfm|>~l3;m^C=0j?MLhO%qm($=X|(j_K~L z*D3A;DM`;6k`cUKYrh7oH|qjC4R)^LP8~j(q-wCtt|>9DOU9@<)3M$+qH4ss;+nZJ z488@8LO>Mk^fgf#KQwKJ@G+T|_8$lVc}5FoP(Owod@g!ZD}0k00)>Vy1s2`M9_i4> zS)AD!E%xdaKdIES5oiF($b)qQ1S(8)wzK~SN85fZciB63)7HaK8j+j&$XLbx{^wBj zc=s7N&)m;(t{9RNo)7%!?(&OV!wlg3?(*`JVDj}w!^p((+m+@!8Q@Q$9(YOlS<(9S zdRVLm^nNv9dTpy7^a9_I#;7|SmGk(F`6QO1K1o#c6Ue7@C-wV4$7*0 z7`eI-EvkO^89VZkqi&~((24S-gDB_es(&KkO8_sbsQ6}gvErAeM7M$jk#m)UUzR2E}6bo6St&FU(1!OY8Z1}JFrDIaRcdC_h6*UzcBEnjvEa)j~d-we4 z>^%~pjfWilpaeQ_;*@}Jp*TnZa6jVZO`|@z9dn(K3b;HGq=e&)w6i8Su#zV9?6+ee#@2vl)f~!0$F*}OCYDUv9F?P0aA0A-5+Szn7ue=t9E)~V<@k_+gOCcm zzK|Q1@*8A#`Iz%k$qGlJ@jdOha}Po#)ur~^WJWMeis!p{(fLM-JwM`<5|#p`sJ;5Uk7$Kl~j2j${92q?<>_j zZ)ai-{!dK;$twx;VP-;<2pe{X1in^@;Qr)KFVx{Gbnq%DuGR1k8Z)hs@5+QcI9a;N zk0L5mnKDfJ!I*suRuxg`%xK;96nwn)l+bIe_coqN(<;d!&&EA8BK%RP(cVwRdl6#L zm#dX_Gb1OxBz_modyh3~(%D~)Ubcb-GTCk>9WZt&9vp_4)lAJe1Tz{Mi1Dr{DJ0fP z=}r~#aU%z&#$MvU6#oh;+CPxAtfb5`8RTL?%wcd!4r@#_V}v})&EbsUWRtMS?Hwpq z13RXX|6`~rS^>D04ezE(+%WAW?mnKzyrf)>IhYPI3u@&I&6P;?yQ5`Wm0=$5GCL@r zD3_=H#h(ePh-U_!dBM-WsB6!i61|ZgOwmqnL*_+>AMF|x^dzf<5!F5tAVexu2Aaqc zC>sM-JL%79m})8BkDk5NUJF})Is5^)bGKmfZ6CSBwnx)qUUywuZ5MGoDxNrK(xH8` zE;Y|&MLXQ9zWnsMy_QTSK7NTgV!$;FNk}aPbbvJg0U3?t&BQx;8p#8NyL|Z;e#rF5l#- zNHv_~5^k+y>#(En?x64^))*@H`2teOG`$UGzdE)mZGB3KY^Z;EHM{)W>Yh$&;;M{i zXdT@(8${zd9I3S}n~Wa4y{sSTT{c@-Bysjlxpw?H@%#QBB647oiy^KH)sy6vWS$nZ*V&wt z7CshS!UX)KQ{Wid!Jogl-tQGm4#>2@UemRv^HF4L8myL-=3~%zogNdrq@S0h*;{SG zy&U8Y$7Y*iM`L6lV|Qw9i0~UTF7M4`BpNpL#5QKk$m<)Lrnh;y;ArAiYwP0`O$NYx zfhu6BY`A!@6+GS-oKxM#(y3}t>cB^;+qmLg+EX84AEn567{o|@09h~kv={$~&MiCc zidImzJjcQ{g_nssT#Kv)%Qpa89NmiIIOpuu?KV*L1&V^-K83%^gWt*De?e0JNx1_m zZ$Mr4Z)BK$WC{C8L}dGCtIyY=Z+hMD)cF6StPM~p1v^n98FC4MoXKjGKKPL8XD*t!7?$@Co?c9Hm zwPF9qB*uS|wfP$C=Lvp4?Vo`3r%@&jPUc^Emi-s@{a5J@j(pbL_VSQnf#@eO`KU%#I!*d|^ju;xA5oF9Z0BjNtn@kpit>%MT(8*(uq`2b zBT-M{snfoPce~T#2M9BWvv@T;Gcd`6Tcn``{2>MmIe^ntR-hN`z5e)~bVE@X6KhJ< zymt(siserKsDZ<^wffTeD{Uox6qUzI+mNHHs;4$5{t#X$HO8%D;?g$OH$FA;7$ka` z>r$}K8NClFAk@jUb@xn(9^)Lg&xkCtZy#>poT$@P|32lyRT~w4S9cpJb zwrn|FtKNONGM+mVfcfeGT?+32PHVPqh^;ytD-T!FwyB< zFPYa=#N`AV7tjHtYUrVMyu{tPeOzYAU_;s?HKnlp`jt8A>Uja9@T@I|UdSU)VrHAOXS|zL1hO^1 zGe`Tmjx&;50D`g|G8cwPrgZz1n4Ip3r}{w5xjz6QfSL^wo3~+lJ6%F0>cf#C3qdNE zQ}+Ixl#3ZZRH##lCEMpQQFRk7Wj$fCGem;pdi;rj5nT7D4vB3kH;SED5qAI)uk`>W zL(jvfkL9bWTO`wc%&YIAM3geYnc7NGRdM3i)G&0b@_MI{Z#YMWC_6m}8uV<$pl~rO zi!k`)Kcr5FsiAXdMCk+w6FXt{mbV_}YO14`c0H{hHI&D8mpSM;Brnff<_epq@cbek zz8~#Yhh&+LAJX3yOhy20Go0{&&{I6<9p#m;Y7{l*r;Pl!_#nc?xCZkcV=Me~6^}L_ zotfTUu3V<;s|o@;ZlPCTPYp@G%3CjwA3ogB9);^j+pYva>!*H6tk`~yoLvw*8^9s{ z&icvP^Z2*X3C?6a)3~N{Os$K7y~RAEmU4s!z=9(m_;TW_f(qx&j>Ho3Bb@T5_<67% zpVu#LHe`2d%)TvhyONp+5=oHEa`e6(38@s1`|SE`l(-5u=i+9tM&Q192@rwZX>Q~5 z;OSW=YCEaM06ROCbh#KDsXpEe?zGZ@#bL|_t4;k=>ivdpG8&(z~jj z&(egMgST(BXUlwS7#S|;heDLuh;FubB<<$0n0EwsHWo~UV5W=L**nj5h@>+fqN1pG z=QFs^>I|ASlAV7ty?DA-G@TWQaE_C+!pEU{jsex&oz0}@)Q-L1neJFnj##In-`;|Y zSN`5d?jzQ)W;s`BBdXnb~S>R#%>_EFs}M-`N1hUtCKm|cu) zo(K!iL+sGVAX(I&XqjOx;U-Aa*o`J0h;SG_b7to8hq-7Se}Wx0x%st-iws5*&#>skiccE0jHRy5tB@@aJWuYeXwj89L}wxiC2JO@ zY4y;L3M5OYL8cn>zv;$SHh+yO`ij;HMDuNoU@$DNQOoKz2FJo+zv9Ae*`N-^si8PU z`&{MCEwYsO;+1>!iX#byjG(RG_?3Cn_(5{uv2_VnLu$Vwd#FrLim}n%0y)8% zPt9=-+8(x-p4?f;7}SYw(BwL&OpKpjTw+d8alkELu;b`QC5L0JDp{+7t#vJKC6vBP znN19v{rF};1?mdLrf*;;xoN}7WXc*=a7&Gy6kPUbJ10mWe+`56{VNRuJ$Kp%K^Mg&O3iHOX4$*$Eah1Mv{}enh;LQdpIj=rLUjie}H24Ut}lX2l+1$%YO@s znK^+L6`(ERAAMF|hkgV*|2I&~%n4+7_FtVe{|*%Y6ylGd_-o{UvD&{6#q59IlJlQf znJQRSzgd}liI9DN0L5SD^KF%X;fVbXikX28^qAOxy_NjNk>7iQzBhsWCs6$NFB5-k zWn%yBCh#2;f2G>5Pz+#a`gJiJKhgNN4f@lco8#L?{N@1yLb2ik0Noo1#Zmg?Dzw+j z9MEx+l9I_)jaG9|`h(@6B=r>p60h!`a>~I1j1;+2i9_~hr>{!}YTG%m(@%0yLoRF2-pgQw*TbE8top(UpguoPWCT5J4{C?H3Jl5uJVQsCe{_159?b~*8RwM`t{hO` z_53V+05X$S#(^i70KU|FD2nL1r|A*8SyZ-ko%a|MJ97{r>eGl|G)WftG?YP*iJD&Jf1Y;84ap;~VldaFm7 z?~r#zEWV-N9ld=_hk1Dw#~SI|%Yo}5FSY7Yet7@+(s-$_BF+L)ns_zh<4jK97HBpo z;*pT0TL{B&HM^=Sm@yL6UPBAvypK1|Q`ppP5u!H=Y#I@(&Cw11qY6vM6;Ink`W$Dm zw2HgQzr+mBf=BKX^SlC2W`qeL{32yV$Lh0>k^r$BKjyD5)kON`Bu%zl{HDx(a99iA zJRk}34jH4W(1u1oz5){`&KJ+v-qP~c)y7R(4BgY%iF!{X*@I5UX0!D8rmyuBGJq5# z)JE0*ZN_y) z@CT_Rf%bhfmOom47-TCHku3(tJmaAQ8?)D4XCWVY)f*-l6_cqRr>Wm&Ah^WWhXOVq z)cz6zLqt(MRN-MYp(1WL0wTl4NyF}ZY?7KW)}`cKe&hAlypvfD#6jyzjwP$~QUq01 zXx|~^7~pO>ho$xJ`*yaBE9Z#X^+iM4k_Be;aBP3heRFFNeY$qFy1J^4aG2Xr{o#3R zTFKM7q;{o|&jqHraML5Iv`Mp~s5T z)pl@m8v508n@VU5(tZ8y^`u|nyc;*ZhRuceD5BeG zPws{gaR`~xo(CI?EPX$6%syc_AB@jEiZa?F(Y0I6>ejq8_ysCq4w@YFf$9yxRS9F_ zUIVqiPZXBUbKPlwMr*I2z*nT!Uf?Vp?cGjcbr&J~FI#Gj7M^VDzeDJxbf0HK0US}; zb-u_drK_ir>TH;>NvL;ax^?4}GG zrfua#p+f>hwrTo#igIcpc9cxr4i6Qu`D)z|*wrVZt3R4AaDGOxVS~+JU#k(P7%v`g zR->TRw*=k8NfqkK8Ir7-fPs_Od1~zSfO(!7x6S<7!Nw;akkHg#W_C!l$9%+x<~-F0 z0h#;h+Qg6By3s6o&aCmG;_ORu6+qpoEH`eQdkghu9>!-^Fu~LD;Q~W(26LjHN-7Fz z<=|*TXM09ZiV!?&3B`HXshV!UDAi(fRyrXH$M%W=Vs}{%>$IkGesBR+^4_`XME5zi z)O(}qCBsP)9bGBjtlc;6X-f?2mSVNf5QHrli^{xSz5^UccXu|ADneO~(O97ci&z9% z18YJRIelP;;k)uVr)U9O?Vzx+d322-B0y>} ze6^?zR9Csf#%D|#vt!cS$1dw3xx-r)D|M@QJ7uRsg|m@lnT4a4Uz=x6QTmP09u$Y# zIFj9Ec#dowL@!+HK>{MJ4uJG1$|!Wn+!~9DPp2co*XWw%iXZ*~Owfq8_6K0W@r%Ut zKW0}Bpil>FHTsY2`gQ1E(A+N%HU9^1e3=12#OC;qZ1_Kg_#<}x8u?$W_D^Bvhl1|c zcg}zBt?GZou8hjRuq!z7s~;V)@AB_|n_U4wMV;f<%foND_gx<1cf$XFz^)vBpZWio zUB3qV`KSG!U4KDtCJv6@1YO|&g}=tGfNvY|8@pDkSt>4wVcs9=+FuUF6;y0`@4QP| zqVr<`hlFllHZbkhzJwL)ET9xO|k40Oqm6eaqc*Oh)YE;}F*M_P#5apy(hx3ph;lf-YJrqQb<|6+Udbp11J(rfiK$cb z`X@!!rkM(I9*HCcBPe7vCQj&JSNN<@BmU^-eu#2q)}3&rE64nCeiBo5mmv@BlcJy+ zZqMbmo46fkT0FV^mD?bF6e{7Do*A3C#VEuaEyTbhJvq<891XF?R1|k7mJL_UZ5g$@DY0fphoDj3Eh^p3zVgUO2l^b7Vq(~6qmblxb}51NLy16vw+UC^f) z0!5k$5~FM9>(KrqdVxp_VC962h!d2d}$If&w z#eH}x*3@S`Z~e2R_V&0*hTt7BSzIipDtt{U$|T~s^DAcD+u)K;%htN%`CEx=)aN8Z zphIw>@Tz1`a@7&CwDg!bRRmuRS^|Whg*${i(ShZCY$S@}MRg;6uq2|Gq^K$&V+cs8 zcvF6k@g|7rY=UU>Ts(U3G_gv6?{)p;#|)-Q17dCZPIy0(K?yo!2YA)Dl9PU*{6(a+ z!f)|Bg{76NU99>n6UuC&C8QVDbckR0m*~8%Dyt*8D>-0I_(FdVkZ5ix?KYj%O-guH0W#R35?0VBiepd#cd%F& zLc0|)mR-(8n7IQpcv?XsmP)Umreyf%!&<__`eV$nWm226(99%`z)?6O?uXcQUE@Iz zozH3#J?bF%249sHLXg6bSvEu3joc9l-@d7OGNLiq8iNsfBQmzq4(Qe5!78DBe@2^J zah#m5Ws(>$7_L1ADNq%_?np(w+ClVOt!eaaRiXy4V+^9)+S|dCb_AOzNAgFl!7zoF zvxOB7n0nepNvwqE2vuS+0ude03ZFKKP7~9~*_K4aGzTqj1Q{@2jFHdMJXHWhJIZ5S z9TvpW*uIR>=e1vXmar!%ZYst0Ekn70Jrwr} zz$b^%sH|c!upG>dLz3o)Xxa;gWO_JX1#BVIdin(*P=(WsYBzBMs*vdgK$sywd4Tdz<9}{dRfOZ52UOuV;e3 z&pLgTm};{<=1p1`SQRae`BH$(sq@S9!xM}surLM4Eb|sQbsgD5x^tWQy0Z>V!loDW zf>BNFoF7?FIc3qR!cy$gxfL%v_`nBM9&yWyn8;fu=m?#IkRNdM5-m?PlHLGjVFq|t zY*X1&@&g>^?NkhjpVO!Vh|>#ESyr#9nG0u35Q8o%1V=f>II08M7eh6D3XdD*X3ID2 z%;GbVIn=lv=y20t5BP7GsEhQ;>vQNmt^ zg}9Z+YnyNsRijBq+o+5;gGAX^!SddFk=`+{Q3e#-VliGKgu=|q3)cgEj;jLHlQ z@kbQ=HS)h$?cb+hz_%A#|07!iGsmx$ZgzfXYxp{!Z>#(ZC+_zOF7R6P(_^}SIw#*H z9{yVt3TsJ@HN;!QSeuc{b`g5!19}d3-}4YzirT;{@lQ7PhU6UHwrFJ z)O4K(pa-m6V{Atk1w@c(ww%+|iYE;gu`v_*$G=8ZKf1@{D8R^~fxWbGn zE;oiz9&QNmy{R^RK%@-RU9>nZyD7M>tzu*N#8E)rHU<5lK%#FIK)5te-Dhd&U$`nO zzhvoXdij8t#ybai5_M&dEF=%#iE>z1Dk9?P<7i{jL>RzyHv6>CWIPQ4f;qAtT zAvUhkr$U=C=E-@angsFAH%97r`{!l@cYtiFj5f1%(~+wk7Y`CTJQp_xVu5Mdd_ppF zbfd<$u!&Qjvv|obNd`_u!;-_ugLGgV5ro9`Dd{a4YRG zUJAa5(kPzRW$mYl)Le#!*~UdUl}Pjqkq_i|xUWVGe8Bb~^+01L?ir-)#Ev~&jmjVt zO7wO)6P?M;oNn?r>X=3h%v;P(bavv@8IrKZb*?`@)rnvvOUmKs<75X=9IzWnGb01 z?NgdhGqfx&q`M2yv6b`p3xqw6>hU8CqitPG(AWdt zI8vKj+uC`zShLHdQl;9Y-IcAG9)$+6r!xxc z31tZkm-$PXCekI3(9YqSc#NQ>D|3v>}X8G07I`qRDhhOLOZIyqU z{D8(#U}=G$f#*9R@$1l!cJ4pWKY=ETzq@aLtbcwD_VWb4pY|`n3N&*Aeid^$zY~c6 z?B`}?{I(I_k)N8l>jFD+%f9N6ytgbk$l>|0dQz8U(yYWs_Rk>X`Y3GTluxKcInQfd zpNK~q%Il!=x0=q8J`dm0Gzrud(QSLx+bFYx2@Cd+0wE8bYu3JvZek+jBbd_rMi{>8 zb=dgr_{%Hbgb7+Qc!v389b5=fakR#Dv?>zFFQTI^d|f*4UD_*mgXK{hUo8=eOLWkR zt0S7D#_)txQ2Ls)oH8`Eb+R!wV@GsSl9h4X&*Wq>hYe9Jqbl#1qC?S}Cw08u3_Kn2+sRK_Hz7 zZvyM0jLki{N7H+iXNHGcjuIqLb4>Yc4C8WJ^l8NeN4wv!Y#Gm>@PMn~5;L2aI1V@2E zDJJa~OEi^FL;wq_8`hHZAVA~nO&rp4&X&t=J+~(1llF8cX-;kl;*1wZ( zQEk|K$$qV~A0t~>Qb5@0OA0NH4{Xm!0(W}5d9N{oOnQ{afKs^PKUK!4!sD59$A)oe z+7Z?goqD*IQ&JTCsFb_M$%FP;ih7cDrchYmiwE((hSVwn2L$uyENS(HCCN~vVl8NE zquQ$S)cl$kVfn6OzFVzgaEos?RaETbV_Bs;t@oQudVu$jXk7ELBEM%4myaKE)*hPp zOO4mU`@zK`6+Fq#EDm*9NMy|381b7pcseH7QagWwlDBBx1-nKT72`|-*Q94fG~^Nr zLQ-kXeSyP`*J^w?A@RXmrFgbXKQhdQf}&*U{tccGK>?AbBNc@mCH>gj{T7VrPc&}a z`w5`^%*9Hb2zS$1Z{M|9DtDBpBp*>6WJAon>g(3tT7_sIMf2&V>hA!E8h3}IU?P!T zG*&y_E7PacpE+(+CX)M;+b1#18cyOH6KA~RcEsaj2VcQ0L(@^J5RU4`wXlFfPGKm= z&#ax41c+={(6#KGLRF_!zjo7Qz-*MfaP@gr398${qVu-t(>-`(ychR}W2~Fz2C{2i zKl;<6z`5AxSQ#*Z)61@AZciG6xBNxS>%<5lQa^*Lw3H)gGlDPS6ym*otxaOEc8u)Y z+#0=#w;XJ^~pSpIj3U%!MGVkLa+L8B@&5DhSJb zogJF91I?L%6AhFcJo8pAef2lqt&W6_i=k;Nt4wHJ^Gf)+S{o~~TX0QdhQo<^j20&M z*(ep8HJ@MXsaf>%@<_C$uXq`&-ghOXWZRl$5yLwCl{$6_QpaKO<0(Sn{^SI1FxSTaM7aVP6yRJsp zteJwcVHA&6EB_bQfpm+=CdeO^M`XeGKC&0oo*iQLxJSlzhe9z7($MHtfzhr|6s}8m zj^_eCg6uzGcw1xam9q~9D(W>Hk-Qpjr=rwc#ltqQ(@3Uw;mDA zvx8_iUAsT#oy763_mpByx=j!jA#&99n4E-FCd&**lr$+%PfQY5sG*it9Bvi5CHtdS->0?(Xuzx}l zq>sMtTlmQFMvxwtH@R^TQ8Z76A}y%Cz+69R>(>h+YB>iLCC%r!Pj+;p3)QjoN11m{ zZ;84GD@_n-Sd%W%37M=rCSOK!CvnV=ccBbW#VGXfweD&4gb~#`%grytTzcf%Dtbe2 z0Kf6v^;8hk;Wl;rxDh*k+l}}>XM`b}SicUMqL#LAR-52o{!V`R|A9P##pIYczj^ik zP0bM);t$F5JM90X)&3cZF#VW3|GxY2KO)caU&zz<$qzH{oZlsu|H2Xbz2*okm&nZc zEBgNeYd;VDcgd5P@$VD+ACl+K!Tymve@4fjMwvJm*?(KiUwlgSQ<1ZO&>WjpwSUUI zdtH{+MY&+a+`h07cXxbk)c;(tEa(8!%9_o=(3RGio37vRsB~qaLSK*;iBa5IK)q&? zP}OUsetX%r*_N>>&bc_nr^$CMhFKk<484+=) z-hd)%uhO@EXm9rm2^?#E9P8yvQu%i7U!H<&z?)G*Z;<-*SIx941ZarI($qP*P?>(- zYkdB+(~_cFJsx?BV(GkhPe!FW9&&lW(;GZXusG-4`j%~VD}MV+*Fi(BN?h6PjwFqe zB2HgdnKi{!hSgd@c0ZlATV+1gtFFfqUz(SBS_oLRJRzZ=GdP}DEWl^wJ$K#S!_@=Z zjPW@d2fjJ7z8G<$2DgC;+vz>WyBoVv49*il3w$K3h+?R64&1L@DksFF{AnYZJyrdq zPW_1!^`9xU$H=^kZqGQd{mSEurQRl;Bc?Q50_`w97+NZv2ctl?s4z(HbdlA#TfQ*}p)&7m>b^cH|7>AkNb^L6@5#rty#XSX zO6wRoJV5zd?cD~AonaIkBT4+L>A4FOEwP|*f12W4<7B+5AQwC6fjHCA7dQ}c5aAD~ zlpAXA1(1CdbzqH!#zhpKsoW`xRiMK>2N>UqiDaXRtD6~Hq~E^sx1wn?wa4QOQA!*^ zfspD`q+p_oIxZyD`cOcdd&5a&bJJU+#h!0iXM+H3gM()0;pvH>qx6!AXW;WuR%a%w z5ze3xq&#Loim+As!%{t#u2&+Nc^jBaB@t>gvcP%zIU8UN9X`R-h+&CDrV1|aQbxva zW4s&hl@J2x`Q$_@9g1MJXBUEshHv`jIKr6I6#9sDq}iit4XXL)8{|6f&Zb@6eurHn zGGpGL56C@*YT^{F?D?FjqdBJ%>B_J2GoOzgZ8ao)fJhM=Y6B?JL)M{~tk=&ufGjof zbO2Ou#HU}aDm*H1HMbS68XYEKs6g)!2T9!sgP2l%BG}Z^-o!3uGwIaWW-Mf4S*dHr(#4`oZTFR-@g_obl_EOEL(*Q;)E7fx2k7btK&>;vcJSOZ zeXq2g$*NXC`qT&mjWOeRppaAQB1?``Tz2UM;#q1;X%ZJ#4&`14^aon(_UbO?8sSdx zgaKvB^vhWi9?Mtvj#8ml{0W1C)@ht}M~++a=@^qW#o^^96*e0*>n@T!597jZ9~atD zw6cg3U`Iw8=-!wxJ>EXCkJir%m4&1mzts?x3#7X;+34 z44gE#>qknfv|>hieSY~)&pbw+E4i|&toK)gKs)f$))LkC;X(E(J*l8*+Q`~{)%3}z zGSN3K^*&)o3M;z})@Xzc{f=a*?NZk={S5+FP2?_!fyD-zHRygNE+-p<^T1bLq^t1D z=7y6)8`l|3jDbwKn^7+IUc=I!I=8+&(O%s_=|w@Kb9+jP1TrxZs+p7T>10Rk6_^c zNUj7{CS_*)W|I3ip%ED351{eq$p2!se+G@rj9;<)r;?_>wPayt{0g~$E8+U@L*tXv zVGOO)dQczaCqA1$%2WS5$DbDYn?U#T6u&L}Z*wHDLksiIR^8tpo$uK6cLe_LE11l{ z`oRA+1@q@%KQH>XYtqjk_|qs8JIk*nrkI(25+=WG(4U?Frf-t}_X;NP^^q7FkRx|Q z^D8#%cLoMAYe~zVL%VUT)DH5&f1FLK=6PX4>Ue#zKmtar<`2lo@2Fr+l1%5zlf4~W zg$tJvo)G(xVaM2z8NRg&i$Efvw+9Ih-Q^CiF4QiO2+xPW*N)YcYGN*SlKTof8wD#f(g(;BFqa-MiH^G4d2LejzDm+=>1=a+W|>UVpc7BVByid`CdM@Q zhviu~;4|&F`NN?6fn{hNE(G`?+-ztX2y2Y~6Ucn@*~%XC*v;3YS7WjQgtV~iO* z5(?M}ei5-X`8m{jjdD4DEW6od!quD)mvbF<zeshvije4#*_{cCTn0?)6pJ?1DN3a(q*LCSl_aJuIeZ~h4srh z@0t}H$+2052FqjGK|(QI5lMMNDdE{=?-FjeziL>FSy3huMjTGVf`#Z}u@-7U9&{zw z){-`YIl%VfO7Gk$(y`NqOgE5q!yT%i>`^f+0a*$R6@vD`9o#%65GTqpO(2FsKkmHi zg64I{a=tKBMx7H$QVb=jIesr(4dQF@m;j#9XIhRJ?L|oKj@LPF8{RP5;2GBH&6rVq z7G%YCbjz!P?C?`ODc)F45iSPm}$rIQh;byv48&@Wvbr)D*0QIx5w+PH&yw2u>?jpr2? zt$|5QLPX^+v~Lrx8MfVy!D!4>uGgBlWP7o# zKmo(`&?>!$jEgEW7nLU?c!ChW)&ESF!g)O|z+Lckd9wdj!Bk6pa<5?!OtpZfE^fFw z4p*j_HmyG zL}H9pUZKSzn#Z_>+sp31z|X-E975Izc8@H?Gp4*4C3$>Ou1nIU&l;v4!#mTTw=SFI z=Kly7!E{MCuadzOaz}S4CHnAkO%30pLT{8-_(6!hh4$HEts9-B89cernJYa{-&gkm830jRJWj~UUj>HoknC&Vo zALA>{uy+xZT6o-!L}S9A;qs6pTp>R=)k6@XV;W6fhEO+7iYCPu?##lerL{)eR4ET$ zJ!C`=nG{Z?))kTA4|SvCEK+RRr{!VK(v#FWLr_e7QL#`mUwdQp$RT6SaW*0;Ltds? ztMQ=W6UkGzX${LdYL?hrV7d{NiAUVPIylj-aktKdRQc%>bwd%7wi3}peM7f~*?faP z81}+68}{fI4o`IJ_J?6*IwjSt=0#p%3=BKX6@#(>M~lpCZHx+?3F9$@$pIo1hMWqu z=cCrU5Cf4|v#=E#xqoo?#s-a~1Akxc_B7v~j8S*+=8mO2t(7HOu&UQwe zE%BsvDoBw{NWJ`TErp8@d1P9iK4eImYx^an>auz0?|o_E<1M=qUh=e~@10FlKE?gA z%1~f&q(3r5))eb9-=JH;xTuLeundk7crU5C+cMEwBu0UtF0+&L{_0Ckp6lAUscC1mRTl_kL{o6@Fmx(9X?LmQCm;U58CW}7^t(KHV8XI z#2X@hrmI1ebxI2q52=G;;0s8yH5 zupF!0@gz*yW>=^`xecF!I7GYw_x^}mz9XxYpWGAo4$8F%8T$uh$@HBR@)y4G|1~)h z_(f#?x#imL@Zjsvw=e4N@ZkT5EP?GHn1OvwzNx1FRw5i2;*ZGkYvg~i+CSxrACe{0 zR}%hTktH?TFYkbS`hJuLXZ|{!Z;Sj3C+zoXBr_Xu|Nlv~{(0y}8~5iFDe7SBZ1;0v zvKNX!H8XHAH#Ald6$JK@A#!qX26h?%jw(3-dlJ~`I~dy-xwR6DNzxdP5;%ECU zFa*;-^$P?}^zS46AM2T4gZ;e3?>Y1rdF-QmcXu;SO;xxitY^2y{>?CaMy^qG!_jvKr}u4{;H46RyDq#j+B^LoK~v zfbl}5C{I4YQ3~?8WjRYFvInA%PreF65QdAjy41h=-KUW_*);N#<%oAz5 z&1iZTl9s%Tquws-+~r);)k->bVIJDHjYhtW!rW?D5_?+&nrsT+WWFtOhWM(VUqGa_A0T-;dprNU$42ulO_sNTyHV}s(e>dXd)|`>{A(n zBv42N3>&vs?6DH_7doa`_DTMD>l6)wZgxcpnS1Se?FhASfQ`XI6$lPqfQCH;A#_u> zb8ZLGq3o$|slJq+zh!7@;slW;qvM%yiA%XO53U=C7pEpB87V?0tr&5nBC3&ja02(5 z!`X`x+$K+r3w1%T4l^k~jF*U0fNauZ#Vy+a6b0#p{022b0>oTsXvF$Yij(JfuQSTv zw09InbMo7W?`|trRg|J6FQoDbWA^c2ybHQT3T%@XN>3H20m7+IY3shUNUVYHXNT5G zqF+9hsB*uc7jo&J3=kuc9b;K!5J_|5Fhh}4SFc>2-$l^Cb!YVsFfmNGk^$s0g(#%k zNakx-PMY3_ks(abN|dC%V3SZ-#C1$qP^x4v1S|d^5){Ps82RX`aZxQKPXKEVljn=a zaSm8TD*-%Z!%w`S2=_<%Xcw0ypIL#6;Sm`WjDKa-6CroC^L(qr?CSk?buwH)2`)__ zld6hp)b$DP7{W4s>7#-fr{tk$e#mGBJtMH+*Q*aZvoEqdBQ&A(CCuh2xnN(U;B?b* zEeJ0QoZcmp3apCG-Lrf?ZF+`PuNC7aVU8?pbBg9iGQ|8b`weElp@%t4B-Ap`NOe72 zK+vI1vaZy_$r!dlbS3f(Hn`(GT7Omug9YM|Q%k<2)RbsW5#UZ)kLvx>gjSp-#D<1) z$Lhw5mY8mG){*I*!=q|9RxWSb4Fsp82NXP{4L8w&Me_}&Gnk9mVBXOlCPm~>s@D*7 zQtEgUB;9$&iZ7>1Sp7QQ>ULXSLrgdmKi|c$$c}NWM%&c>Lf~-LTmtkzsN}z+l=Q*5 zR8vVbqZ1htgi(h+QwzPHzNeKe7b!N_-FG2ju#d&(aGIXUu@SwkSgoUf>}Fw0W>P%em&=MZ9I!GEqnAN7CWLbd}4r$%2*E zkpv8wO4DF$u<*Jz3yks=$UC|B=_RZlwP3J|ps+}hq4l$Mq%*Q__QlkyHn>w5p50ZD zhIe%F4Be@2u94RBGP5eJPgZz&G@{vW)Je*3)+wS&Q%R!CvXJhX!h|F%d8OuRvqSqT{gj*`WtLtT#k6OcS| z`M1N7wsmE+!AYGhU{|Y63!Yyz*b%5}=nR=`JzXl^bgX6`U1(|5NzA>EFAEOrpWlQl z@mM{;5dnuY-X3a{{FGP2b+t}ob&hpK8_{=UR5!K7=sUh`Iyx9-BM@(7-!LWToQhQy z&EZ+!RdosjvqGMHKX0Lw@zA~H4K<(vAUi8Ewn)G%D&UU#5R;@aj3#)c=0uOn+h z6G??423Ee!b1lqU+~XgMXvwv_{(sE91yo(jwk?VUNN{%x?ry;$xH|-Q2<{SGg9mqa z*Wm81!5xCTJNyMZd!O8Q_c`y}dtZC+x0cqjX0cYyvN30`K6;ITf-o4e83-^d~ zr-lVmtD*gg&kAGY)=Us8bA5#nH~K%1%5iq-|ym zcPQom5ts$gaSVXG+5ZzT3#j6sf!WuZ|2Mt;SK#Dted~ba*)Q$=zXWFKm>8J;I?q%5 zcgf}cXwR=+{u@i|KcZQ{W{UxUXn#$>U+(DFqQ4ute~o7UJPP)okSGA2GBEttHv#_~ zi@sL&M+1MuqOV}`rIenPf$1Ne`AuT+ucjN2ee92c_y;H2{1TC=0TmFDNhMqdoFkHW zg2|U+k`5j$S1}yO3qAl*3P6fpv|P5dOJj#c&E~P~v|Q-o-pFFtbGUmR z$y#|hG&mx%@-D)VX5kXM&uZ4-WzHIjB2OZ(f4w#7LK@LgxB^F_JvqTemvnIxdXE<- z4boG&UqIl(ifp>F(sx!M4zCj}!zOF)6s7#^@3?m$3 zIz-0^?RW#2by~h1?4SCXi|$i0P#iJLLnT9s)0J5T(W%~^^##@fak%SB1)r#i$hgQc zYx`4U{M5e^s_S}GZJQ2^Mf)ATN4u%hh&%OGCfcH`3jin)iu*9@%;1mo*pQ%NI!W~k z6QX;qaw-$Py-8L%Pp>G~yYXIO0#B5C=NqUEP7+R)ib%t>uF8Pw&Z53pG1P7hO%Zm3A9kmS~m8 z+*xlRy={s25?no@=qH8CKD61IC5yW)1NlDKJv1}af7FY2$ai~#Mnif5@-&K|$9XD> zm`mkF2RmjJc<%(SPn7N>kx?lsvRBKCVsk`F9_URYo2MZ|zz-$QWXkv;Fb&tUtnArn zXVLAd1Z;t&E~b-W=P-2kHoh5j{9&(Cwd&g^CgUfg+=LI$D5?aQ$-TRXDPY{Ae&2QH z$rK1#fay)TO=a?%-Puu%vAgL*6ioK!5NV8 z*F>4(`PY-FWpGngx5^cQx|wHNc48BpXVF>B|0KV5=68yA&(Tpjpq?qn5EopC z>H=eVp0AjZe4DO%PGV{YC^SZ~sL|mjLRzf$9W%$(EwK7*-LeA`qiV^H^B43dvuqQJ zy7o}aq7L?ujlBf(+~u5xIIPM6v@lm*ICf=zG)a=xd1Sx*&qb-sU!TL@(ua3xo`ls@ z=8p_ZaC z;Fw0LB@rTZeF?gRX1M7@vr zqMp|}DC85};>%#{ENBg9&2jM%8u40*5YV5$8(+RNLjN($uPf#ua|!+QW1=x3;clEv zGaL%d_@FZrbplJ(tNEus30=`hxkt2OmaR^OurYZN#^TM_L)T@__IqZo>wr8~t03!{ z_gzUo9jWPPjf@~V$9*nTZN+GEzO-Kj(i|_(9}`hjV{|f%_1u|Ad<9R{zCp9O>+1+B z@Da^f{-C7iaP=ez5;dxlo4ub7tzGQp=8^x@0zC0Nl>HQF^EhtTaXXF!^L(-vrQkw! zNHHT`9d0ZCp(HJcKWqUMJXySCY;6-3qM@V1+(R;?tOHfW_;L&v4K|~HO9lU+b!wfp zi&wxSAZe2BDhh>-saqzs8hdUKTIX3QI2KD+9fpj=2@lDj3$0xZ-HX1XSJ`2n-9z-N zdl=60BOBeUtI7&r?VHr1yG)GZ%k~?biNN>893sZ$tw)F8lH+EH6=@0P6$^zcm4b_k zasC>lM3@e#IZ5-?&PMkfV#_Q6y`8pUTRWRd^-q2<6-o`(CcR>u#v5J@sI4% zN=_QOub6f$5g!wt(f87_?c+`t;@T~;Js#7a)C7f6Ikua9EIEQg=2B7Xl)oWJ@qtKn zmf*KuK0%Prd@B7rH2Wvl^MBXrPY*bk!NB;R6@afre}_{3-=NtSF)IT=pYT6GvoBTr zGc@~J^Z%x|{|esxJ(^{F-6Z_qL$h>$Dn-4%!Q1}L5_@gWuU`I_(JVl|$pDZ|{I6*C zwdn6g?tg%08UO3MfPao=Un~2gfxn^IS2+4o3JAJS_ZOOF{7tR>ucn*fKj#=OQCW~% z=mRK4l@EmF4f@?8ejwV9B(Jj&5lJB&yk&&g%80-$dJWNpv9)hhOV`y-!t3u1qH)(S zGjCcpJ+rKgjysquJHo5kX(eSMmND|86E#`-hfdVsRVUir`yPd$JK)Ix|2-@Wv=yVx7e$2ed#)=S0XvFA=AB;+Mne377tHz)YsP+}zHRGV(Hm{nxx3%#NhB1x@@S>igkZiz1P*Ao>cMh)l3;kQ0gcupx;ia~k%kwwfh6 zH*;a5Dg{hS2F%7G3I)_lS>LPc!%en&28l-c({CV*)Jq9g1O{DOw*jG;17f>Z%+KNB zNN8)ZAX?bTcjT8~F7@4Eau;m%H0--jf4+7?Kv>eDv-rV}ag(0S_|Bv{m=KGtT9H1g z&xiI-Y`KH}ea2Dmr%uLA)qrNZ*8Oh1BO0AdQQk<;93D8NP{H^!j(8{JcLX;s?>&4( z8(J?iLB%wG2qb<3<3q*JI#FKpT>?Y!orGb_NwT`rz~Gn;op+ZBc(<{wkYQVfRes%T zP1nKUD(h*f^5rhP$!u0^riz1zhy37^TOKOYhBew2gc)l=7e%!&Tjw(|#l*^t3lJgl zrrgrabIfoVea4L9Pxh}XBnISgWKifHQPH6rU`61f<%5)REbobOcQ2R~z9v9ck?A?r zOZl1E4k%GrQtoPNP1R4H_cUPADQWB!x4@~j;5r^GDtb3S?Nm&X%$yI1)1L`qoj?yT zgH*ilSV6210B2smq8Drt&4n~R*a(Mle~%)MIq^~dt&rcuH~7?uI0_P-P6=&X#)xtS zqJC5SuX3n{*v#6^p0@lf{HrMRDq@+^1rP=YRyUl>?9L)mrkeP10q-0$SMfj2Q5ONb zMh5d2uw#oZDb#|pne4y2w56hS)~J!p2-SGtSPN;girkU(WelqPmb(s7a3;b+U%@}X zrZJ|x*7XffVGNM6$C{sYt@>E+28JM;>xpJ*3`fqG!`0mMg1^1a+I+cMDuprj0N)O3 zY0)K^kP%=gZWp1pk@ z1^O9A^X*O5l3Z}1pl6041LF?1eElwRk0QK!X^0(7{wN~t@NQS&cLbEP zQN{U9jHfY^@*SQjqtu^FHxPsg)T!JY+$=NFq}N1EjM%0vm&;glSO)CbIQ-y*&rpwH zN9(^nnBK_r&-t0TMMc;_CBr`=etLhv?wMtqN>nP~{tl0yO>xqK?l^@tVWa9P^)sbk z%R9GemErA^k*yqgW01x3E>*)`3gxhrVMUGm)ey^lq2bRb-he|~tK8gRu;licuBUAq zq9qcxGxtxAg@aD(w(IF0Yi=T^>2gs4C~{= zd_w}x*UvnRjhd`DVpuOd`c1#~onVtW4NTa}_r6_G5gNnFI7yFEtDQ~ZJZyA$ci%8_ z$geZ!rsU`#9ynd@i^N`}CNH8ttVb9YV4Qbu$~GR;(NRfzs^DBm18=x}jN0G{S3`&w z3A;aZVPN=(yMi~JETtuKM#3CWK{YhSQiA3~S|k->!;6k^U~d3{hHF?|MZ8}pB>i@% z>*+hp4aB2@ZKE!D`XS5Kdx)ROAm%7FcXDa9=+G-_~1K{T=pr@kRdo*yF!N*1p7IVy1q|avZzxb4G^#ED3bO71z0bjAjS7Ts$#V#87w1O6Tmii_ZMlUxN)i1#faphg zXInc1b5RRG^19cwbJ8~Y1~vdw!zcUG(8yoJ*G49Q6dum_WPJLTx(0v}d3!4>GXryg zDdN)s?hLTY7o&YKVH*=GJ4+kDaJ~vAezQ@)&0fCqVvr1fszZJ~3;@4Zqr+EYc{x3& zq49f9Uaa`n1pKQd%E0tX!~PFST;rRhMHfD_-*s}w;I8gRh9#kh!d;T!r+gDQ-nML! zP(E1IpI6KhOfjF|QY{<{!0qp!aLV+hudvS}Fy8QnS%+e&`M%xeEo}CLV!=~vKGcnz z013ZOw8rQYp0ZEp(XTUHUc30Y?hE%0=mLx5Cof}iOX)@lvjzF%s&{mTtQ#C`Gv~GL zM3E}Io!_WWlyl#`YB^8$b}Bs0cMHo{GrKwxy$G1${05=SfD!SFcm0n^ct3s~)~*#= z2rhAoQS4=D)t&V>k~h)^r!@qA&Q^i+y*9{+f2TM~%XDX@_cw2a*CBWrD}blvuU**d zEc_ZkzL)2#9X>k)8$PXoj+KakiIK4#K#l`Q2=eRUYllz$;#{B=GSe}##b^3sQePgB zFB3#fPX|!TP%{CfSAgUgFBZVU0(dsROremDxrrHIb^P%-$A5XA1J=mvUrfvZv(3Qt z`V#qbnMmoF0|Y~V^9m5P(=jvAFkOQ&7fuUcO)Ah|-T8f)>E{YwNoBZbtAe znu^L^#D}T^k|YN-D9pSw_%8f!gvPvU*Y+PeV#pBLFc9F7Xw)M05Sp9l2Sg za6haC1^Z4)MV;NF^x*hyDp8|t|DvL}Vor&eRl#c8gaK4+ScM0b@b_d9Bk+NdLh*@nXSZ^GX^iv8Gr=X`Yxn&Nc^9mIk zDU_##U>)}FFEmx(uvy&gz+j3{_m~vs64rTQV=D>{Nal@)k$j+4jqb9{5}FramJ43V z&t^6ZWfveKwc{|4QYF80(95%yE9B5E&XID#97z7i5h}GdC8TUjEjG-2_PiCZQ}FED z^%?EiT%kn75mU5lOD2N(dNIJ6Ym`ZUZzR<@Vp?QNZeGu{;`R_Dl6bkIxraV$u*T0} zT-5cgqSH*iGR7WzJwj-j1JLFVvWnVnH!~f*?^fZXiB5L5ITWmkJKhwvqfxWH8zDqS zRqO)je!Ka_F)ZO$n0i0-vJ^09a%CIcf(8&h0=3Ifra#7@3kLsG5P;ahH+ati=wW56LxJ%27=CrE0(i6pF`&*UHoPf|)M z`@xEKn;zUplTSase;Aus;wq82oUF=f@IY4LS5U<;dWt%e*eYK)G4R?Mek)j;DMdz$ zs4AL(y&=-|K27FFRxoMkXL0wa!7Q&^6fbaTfu0@kQSb4DGYTj@TP3-bu*aE}NX{z; z7>V>7r+u4+Ade$u=_R(+PqA!_PpK1mxLnfrh*E?$Da|evEKA=HVjr|R6;4|whYQu} z^0ybpqTq6vqAB{wEh_?Qda??a4TBB}yNjV5w2ZnQZW{ee=jt%qQCMjhY3QE}*N|Ab z!&Wo>!ZCaH!_o1CPG%q*bZ^9=Jq*>+LwQivL3{A3te?~ax$)ZB={c3XuSX4cul3lH z{1`*7(etWKA$R~uVvf;|?@kszHBAPs&`AZ}7JRaEkob}EeJB2Fl;!m#DTd{BXBm2{ z{)6`^Y8IC_>jL-=D~37f$+=lmyHoRWk=Ake2DwwHb;zlFPf>_6I*-mn{3*i*Syqg7 zS8&|4>K(Dr;F{7>2b-&L824gOr~Teoa4a_y>1?O~-({wsVGk4nw*JFc1cQff*7&}7 zbrl|IIoAQ@W&L^As>Ww)&g6Hu=&5?HX%%s&TzFS>sY{2`qlS95SSv6(aq$O1s~jOI zqimN_J&aNZv#EuSzU6L7!j{tAEO0d>YbHJN2yyN~2-a(&)-DX*q&0PyNIvD=p_aWT zH(N@NYez#;S9JKVc;2abhucoY!^a2ENtbP zVEY0pq$!__GM0la>8H65rc5#5zMM)5;po8-&OXYRm$-kGRSUNCm64xUn^C#>pvSBi zgGHI565$JhF{qaxrlSh2NKa*;z|hj8k{njXfa(l&nTg2RmY?6v53JJzBJ`C$jXs>- zk!e8hrV_~({hG0*FJ2=y+o)G2)>oChz(tAWNN*^5h!IM59*fvtIPRhwnkQF!Hz$uV zm;K~T&yO!$E)*M@aNMJ0sFL)01Nz~q6Hrl;>MoTKQR?x1H|k(W@5yu*^$hpE=#rUr z-{^v2q!}!7jpQ9)=`GIcbBzR4=&8Gym2+U|siO~*Y5&~5wReYhuE$N;7Cs^i7`fDr z26cqt4TG8^#QBg!KjDR-3d_~uM+M<;M6$Prt52e_Xz|915<*tHU^FzLC!1N)ggbJ_;1OsFiiNWmh+ZL@BG)u!p!{n&NEPMOL2rU5>68a za0$A{UcGMWeq{=wqdFevL+SQMJLbbn=LmA@eOFBc+B}6KKMDmBc}O^%l0WWSy9!} zuQzah47oa-is@baAX{tK0F7~9NqX-;6x^H*<6*>^iBq5-n8&dFU7~0^Li5^W!Vw4c z6KoZFlG%8ys@-wn_N>J!X&o?8pmRiU^qeeufJHTZXerjb6rvxW2*SA;ZJ-;idus2# zyXH*rUZ)kC;)K!%to2Vl>sBiCtgwh+2r5zs8#8+7mO!K-pbNHo{^&F9fvjGPnugE% z32q-|bSdOP5ehIuKd1=#r;5vBHXP|cdOk$)m1g2-29oQX3CkANOQel;BGT^c05fE% z52h~r2*LoUuD>(F6OQv7ZXjcD)_FN2b= z{TY+8pGXqFN@X$&`PTtxsg8y?c|K-2&74m(L(EOHUuW&Lg5R+!B>co3#6w%r7VKnB z8`dX>u`lt&;HqXQIoiiY*gVUflKf=M93{e4eIbQzPWfZ_9LLA1z~DQ06bi8HuzcJ5 z)}+uv2v5ecVh~DXCgzTzh4bpnErUhJq`+-3q!}YNvI{(RMMp%!r!MdGYgypFb>7?hf>Di zz$ecg9!gQfel8-<)NE;T7S!99Dn@M{3Qrr7H!%oj*IeNF$DU4u8&Az~HtMy;9rPlEZhxo!JQ&mMETA$NX~X|&9V14t%6cYvIMjy2(eMWt2cqQ2x5*%(c9{KUsMw& zg0?^A0GYpe`f)9`QX{8W8~@2~2%7DT^icI-EEP@Fxmu8rCPMda(eSoU+xorE=yk+U z6u7JFT2(A8Y}!@c1JtHL?eG<8zxY1#BwXT*ecx9uoTI}-drZ~RGj6#}iS+f{j@jWn z!xk<$^u(m32;Rz>fV`9oM_9+6s1jxcTvN5}bB}(XTWjO{o$#@&dZ$1R&tuKO*>{Ll z0h`1uL3pU{S1M`MKvE1%5v=E>TYjREbk`b*;OE-HnoZ>BtfwrWG4bt%8sm;2KClD! zBXS%kQ6&W@Uri{!?|duJ_s?cJQB%k;6|qG$*e_;eLM%(dZu4}F6@nJu&{G10n24MN zPI(u93FZB;t0o|u7;GcDSRjF-5OKwaU1xm#QT_y{K~aj?#dB%c;)$Obk)g2S4i>eM z_s!_3&#-j>J4++Kky_5JC(Rg4Mk;Q8L7{|EhP?4|F*3ERF}QXY$b3uGjq!Q|jmPR{ zFfY$%mHHd8xD;~rKxJiAcVov(xX)W%JE-m&!<-O^({jBs6SMxGYE=iYX>s45?fGLN z3nVp*cd6`nZGJ6*A|<-c2cL0;9`)YaZ z`zO?kXm;xx8c1xzFHS$Oa>!=Ey!tQn))$huD{DR*d@tCBsy5p4Eyas7GHQ7~ZO@k6 z$PP9CfQoXL8S~w63FX=5!K3Vw6ZAqLKU(n1T>Q$}OsMU%$(eR&uIu57b`f;qs;I_5M1j#o3c|~O<*U>$^m^+nu`ahszLhf-#`-P7Y_Otsqt_4-d}hK z0A&DRsST;=r~$;qA8{F87;gI4%U^;|Lu!WC%U{IR%jF9z@XJsCLOT3M!W$5wpMm+6 zqW%vD@m_!Mt9Aci-y~&)m4zfJrA*9q?QH=Bi3Bx(RI>+=Bqlbtb^^vaHZO$7OH2Pj zY5;0~r6^wRETQx3TL4n#FXr%t3!(oHtT)RmiTuB?YbN@3#VL0^rpY@#D6Mh0cz$~7E@Bk&c@`G9-^apO``>9z#ku|*_aq< zm;nTljE>14tRFt0t-nU_m6Zov{uR~XkNNqJq%0sqDFYzt?=O|ae~T3dRPoPP@zO!M4>!~c6AEyt z{4*vLkcaN|?*K}a0Uz+QKY369+lkM>^6Db{!Dnq;>Ttnceu@%bY2xw!s-v0C>>v;?0s3&OT=aho<7HVjZxv&5U(wAYy`sd z@|;Q!5Kcg{G^DtBUbz>Vx6(oNW>kLx$0VQ6lK@D|<7?z`5S za2U?vZP-FVN~YZ6iaLjie|wi@hlJR!0?`p4mOlAxaaf8TdKXL2=xoT~o>uS8j8Yr^ zChh7`zTLbVuIWxArn_FjD!6zW-FALnK2#OtQ$Q_+51gRl3au!n{oUVK$m^iIj2^%N z^w$!9ox*=>NPl^A{&+6F_(uU}3Vxk+Wx!{p2W)hIT>&1qOw7OUcn@6^71S4MsE^t~9!Xh2S!9sCBLn<<2L+GcPQd4FYUN1E1_lS__!bl2xE%gxMDys}{n3}Bd*|ch z@3l&%rX_Rt@or)ChH00ILl+-L>wlUhm$VOz(1D@hv6iv%Lapl`9$jHWPOeUXF4TkK z^6Fq48oEJ(ynBK}4VZ(6cJyKE?0Uw@1A+iWfb)D0e+pk}c?|o_wiXoV2H4em9fW_K zKNHAZQ!~N?;zMUQaE+CbDYoGoBQ#64t`S@U*qeQ%Hp6tIu95!R3rk;V2?=QJ!$(W- zbVvuE*>)hRXLu)Gt~MDXG*UWf0$>>@&kPwzNNtEqhYCkfAak=emR4qHFPU;eUJMYw zx;nNmK(olR&jn}S$na)51t0@3PN1PdJMqNs^wy6J$vHX?+dO`BPRxy;QBOi;xp@wr^&``!3#$R1ndZLfBT@ko<@kx&Fu*T<_o0a zbBIru+&YD)U4EFRCy5UNdkPYNYQs~F)OLUWv_H7s4$3k@mw&yo_KBC2&Qll|(u?DH zP1L>gIg^<7_~6aS+By;fnAK_UJK!6q+go5=eh{Em4@lsrJg5L+2*_JymnIMK!Va?( zU|kQT)3Fv0MUd6go5>lVnw1tqhzI#J;Xw>wU@ttX^EmopXFn%A&fAmqdpwwXxF_0$ zr!@A<+Sm@uoL1^PJPe9Pfeep}$U|vbVBDdf;A21!HAu*i&!bN1NML)9JVQ6(B7`St zM&}-R*+D?uKky-{e4ZM|fpEVjF?2zhWr6Z}9y}XN^SlL(he3z+Yis*3V)1RI6$^-l zo1o=cYHvY+H;k@4X~8G$d9{r$$%AGv3k;NffmhoB$eln32p$%~W#fc<_M4Unm9_8AzUv#Y@?RZ`+fwOSmxgSPwh>?VPk6`$>&75W1%Ix9=gTdgMLh;cq|s&K@x!pvHAc zZOaj$h`;y96XwL3Ik0aUCBY;%3sK3kn4^6F*0>z8-B8|(vPLxgVb=wo8eO3V#I-W1 zFSYOCIv$V?e%Yc$#uaN5cd_A$=GRn#3y+KQH7%`Jm~-sb+TdKx1XJKkM7e1mLt;m} z1xI4?qhf(nBfajx^xn^)z#8yAe?6^JJAY4ggM%UL+Pe9sFjC)Tl++ddHeIQ*g4A>h z1}UVKS!@allKR+|KtF)mTXUzd*$o7T}oj)(MdYhSFyHWK9$p z%sX1jRkzZ+#RSMLvJsIm^0@>5>L(P7N6>&SS{zE1@3hGri*l|0PIf`>m&ASUEP$-J zU3*qY{LN9WX;#KQ;2uxs8V-^lj?t1IcTr~_`VxVEj7uAM`>fsu#fyXbu&mI(TiZyI z{+!9IMR8y)h>N=;x!$I)NEE_24~Kr$txY1;T>1H6N)duGrchL6YTSQ;;Zjp>O0^FS zRNQ(W0aFLLB=rnCk;CSylCh<3r?R2>cDYv$^AZdRtpTZXER%LhxbS>G+jNWq5DsPz z)6QW$c$${>TMk1n9JsaenS_gCc54U^;VM_x>BG~Y%4mxUb!FL7xH%NCA<0E|1mCj( zd0ngkIH~!zSf8!t?y^-~P8VdOb$4*lsn3r^VeZ}5=;Y}thq*grvo){2sT9jWtDzLxrG3GdjMa1)Xw(G>jY;aVvT&za9TRbPGWb*7bX3jBc@mE?UzH&CXdOLlZ8S z8`B#_v^w+^9zOGjhvD5#E^q!Q@g8*Ew|;Crkgb zHt}Od_~R0{&$?>lc;oxM8JB!t55sE<_790BSP{JEP-tQR#u15%!3rau zXdmp2D9-MF7*rNofLjKx(#1mlTl(F3C`C|=#gb~)tEZr$gK#Mpu;%X)V$Ao%DpK=< zOjOBC_78@=#QeLW%yWr#YQ>7}Shjk{Zz{TQq64{PvByIV5Lr7H$YVe3R{=y_(%RvTIvgI&JdBbWvgDUV@97?ksG9EJ}wY);JaIYoeItn(!dg#=V znIAEnGL{*Rxh@;)Y#z7sXEWpYxN&#{+kxR5T26#{Q+vmU_7p@bHopLA7$*sJ5zx%^ zMr<(`szq`fvPvP^szXgIW!(PJ;}K?z%+tGVw-XeT`*Pb>nvo6UplPJA_+^1B<~2dL zX=4&Z)&_YouT#p2>mYkNn1rMKE+dYx$XYLZM;(@lv$BVjiEkbxiDsuI>*PODOe7)G zWL!VG_^kNp2_>+(ezk5?b4>CVNA!JK+en(rcXfl1VxJNo#n9J+M=t-66NxNB@rh;M9f~k@4H%+xhSH@G%+1HNgEbY)pc;c_WFjgd(!V@BB3n1XDD}I9=Vs6p5-^ zWay2Vp4UP5qJ9kLA?r>5GrXNOAdOUyn#SuP(le5w8u#OD(cf<2${XvT*zDl zywOj@7f=XrY71wd1PMgC%b*r~VAuzi!3%#{W?3+0`!%wykEg=Lm;9)Nryq+v!GfX} z!Yi~X()oV$9&+Og&o%Z?m-~k9*hwWbAF28=9!C3v7^YqO<}f9rd1bv|wdKr&@eQbv zUDruOcLchaFqg2fUf=*M9HqxbiyCP~$fU*u9noT3sl)uU%1Vuq{$wbSc^tO~6l%rL zwWf%QL?)7QlTv%ND(U!-6h@`>tJDl*qp3V)i~TG%E%h(5HE8$t-1o5pTDj!5UbPT& z)sZ<3AmIC>7KQ3bu$-U^7bG_%d%%Z#Fve726+hy0oms{yEiTElfoTM8nGH&o({3RB zzO7_<^Bi10n9`Rfk%L74=(jBxEwi5TyaX;`U@DXt9lgJt9?}>p{WPvPbSYmt_70Y9 zmfQQ2)|rmQ1HrS2PiRDbxMp~*+ zVo0u`7Hx|!$6QwQ`-qv6F2eFv{=AugWChCl6HIx9i|ilOQVxwS24NpwGHPf$Qdk`{XcHo^RR zB;jr4Xv1F4cO52f^tyAM&^G2v4?zo|TUqhXMI*cK78kGs`MCFsJDI{e2`oYYgs!&#v4H^%51&PMjn`KxngZHeM(AY^>! z?yZjfVtTDS2dFOfn3uDal->m5rrRy-&q_|$Tu_RdeI=N&{YDM9esNzNY5 zl`PbjYM1~S$p?Z>7$~36lCfd}LS>PMth@X+NCd**VHI$fnxDvYH;S^4>rJ3Pr-=Q~ zbSMP1lU8&wxt_gE%>lt8jrw55hUB$T)ncojF@v2r-v{LVnCGe4q*8W10#m#pXP~QH zK9;0IW?C<_9bTF`d&u!u)OS=k(O22W$S6HMT?M?jKayja;LqQ^uYab$%1*|Y`{7j3 zlWJy`w>8AQVf)3jOoze_dK5mo5ahNWc_mUGR;oX}Xgb?+9r@`{<|~b--A&~Wfa44~ zh1&=_eRtE@1T#>u#dE=Ak=;h<2h~-oVV%9JP3rl$PA)w|u+e2^D7(p<{EAwmst*j< zP&wHrUJ)+W=qkC?u6hRNYuk!uL3K*#Hg|G_qNWQoqB)lg9BZi zWrxnB<6HkwM?YIk{+w)B_bHeDg2*kVmJ+o-*uJ7p(e+!4=cuyNH(Xc4D3g%n(3wQVeBGV|Pe0&pYg9OeH;-JT6+j zypJC5HD`a^Tu@wYq!DmcB?r4f=ZZ1^`bTwinpA9M4vdIty>d=^e9DiN$rB;!VTJ`FM| z?6uqxr7S2v=Chm#L?*MtQk#&B)Icw|fFPlx=ac)Xw?c7B*=q?%>vF2n2LicPF2I`M zi^!xUblQ?ab*B(~nW4h5R$MC_HIPmy*BnHK9q%DAW@h|IDgdQZ&u=be)*l1M#HewI zqKYni5&wVYqbROf(am?J^(N$#V(bd&3tIA?D5P9>}N`M&6Qq*~}U=ViK z8+RX;!Z%gl24dHUCmzG~VXJD-R@}xnYm%#wbw5aDqv-M5{^k#9Q0ptMwAnK9RIgKa zcR!bM1v@PH19Hb0oAoA%>;6p`jchRu`|x8a3RI0$YF7GyJ}j$)S%ehwY1X&MJQ_;^ zRTd}ZFsNc_d!>s+6VUQTuGPp4NHXf7S@?A4H9F7o9Wh+~=t;88Ez7yCD!9^%jE@#d zWi4}IspP+3}4~N{#N0PAp!R7oHacfzUk8 zzHV=A0SCVrV>*W<TcKk2VtY{q(tO2pQAl=rF3i z1r+Ed2B8zf@)eJB&aecF9TYmR;jXcn+k-n8)#j&WDUF|w!_hXncRd*g1##o=*y80? z{ksDF&Qdprm#QN$=b~^B9rV%T?8bGX${Gs6ZX4Ca!8j#cLQ)kE<;V2IOjQrxQC`b` z|6Ji-*qXE@Yoml89BQq~ts5zZYWaZ}-sR8}?DG;1YbC&Nui=(G&A90^ zWlfSzv(AMiSAyibg-I!Rz`09d7-ikzuhdF2>dMa3uB?nzkkb>=P9#?L?`tZ|`*yjW z%{c6$FGZPx{BLor1Vk%6M0WGb)-JpFyLM~{reW&Z-MO6Z)mv_BCQ+bP$L~`joCISv zIeeR-jR!zXT`Fx)Z{b7B$l)UK|jjDg5?4QSk0 zpkrZbwJGwsdIH&Gja@0lq)2-*c&*$l>EjF~S}^#6<&-!|Z0x$^dNW3_Wy3N<%`+L@ zlWC=W{ppq5#zXA$;z4xd1cGl5ynJXwvCig5Z|?DKAW`XVo4rZWNhakody_wU#Uc!v z0%K0wjn$a>eOcC~l-V^RLK2f@F|AtuCck3zfuwy4BW-_W%;L?tr5S0F52mio;-X<* zt?0bD&ZN33WI{h~t^Q$HZY9uMj)ni`7v>{Usbkt7VuSK?+7s~(1>)_Q?VziU?-hlE zuN)gv5|&_KMENBXv7Nn^8Nf)&X6UwjI2w~kHn1wXcOwpK`8&i+=#wVBwYqouyAJw| z2eZ^0CfzFIaWXEH8*$m`r0@h`XFS^58Q4y9Kb7R_yNOGP!TJ}or1iU?!TGT0_y z>-G~>Acuc>;`tzlU zDk#)`>e&b-P}W6I@}Y^Dg+na!{)Ts78OR$Zd}rFQJ67N3eZ`D2P}~9uL=*>A8IJy& z4~Z!wlc{1&c}`lIV)P!^>gx2#8P4UIn?=6Mo<`kuBwQh%IrttY79eUP!D;KR-?!e&2-#I+3?{b%1P# zNN&NQa$=CM2f2&rY~}%El+ZlJaxr7~?I(247~T?Fv4gM)A=}L@w?^;n72rZebk2(a_#YE*NS1ykh_58 zgjco+)i2z+5MBKqHRa!p>+SQJDdRXN(7DCc<L0IQV%F#<#8*0u-k~i+p}nDs$<^MP zcd8}5KWy2&q^xkdNH?FWq+n;=9;PnNxqD*g8872h4hNTQ%%G_Af7;u{>j1`^R@gfA~fGq(cx7GPHt^`4#g#I01Y;eA2SGT z-aRwg)(ur2*d@vgckdznbYOpjtgnv0Z3QFR&%nlOzc)dNF2mATw*7q{m3)& zLxyTy%j@Lvg=u&}Fp?lt43$f_rE7K=G&+j+wr^x?uVi{-{fqQB(>I__gHUu6VwvCS zwX!TBY(RcqW!+EDSD+bLf{2N7N4|t@j%RLqjvny3T-Lbgl$E^;*vVc6qdDpe$E%Ue z<`D{?jtdi67o8x=LiZiZ;oQ|;YKN>hbVXy0c35E$uq`?(*i{p~5L-9jqf>4oBuW#} zhv#Z{JubC@Ghzc7{f4;D!P??5xZRUg{3C2Y;3lV^mA71!_4!c=!p_F%W}(sbI0Y;R zt!2o_+Z!DHJOT>@XXdGeR2dKkCKKsHg%}p^Z)F8k~PhhfHs_i5ip4 z*)@Ghq@>ZS^6!|2aTIH$n!EL(-;Macy+;m;GBiFzAC=h;Zp<|XvYsY4dOIAB#V)1l z%m+$5gvRyI{Ud{mkeFH)XzwX~U#N!$p>TG$N!nJsPK)P_dwyq)Lr!lxd9BKRUiYczr6`sKw-(KQ@bmifDT8n$olGi2+;B|4v)s2@%%HJlq-j&}3Hmnq>xF17SLlIb`xQZJd-a=( zfFj&N^fOqJR`LqSuz0EN3-xc8S1E1SNyci?wJ1D~X89UX^5$WC2*}#y+(+UbW<9=RQEOF1Ry1&qXGDp@?W_6Dhrg+ae2lM)K1+O#4!CGh3Q5`{ilQVo z+9cb-QsuT+Yu6N1GHbS(O0G_^BhzIY&-wa3HMM1OU`F7Qyp~y*=PTb5L{!s<+mlQd zqecC1i%Q!t9^}r=sk6Q(v5IMzB}5UoT}Z6E($N$~gs8e*Le)sD z42^{tKJX~MN$RvOy+%{USN0yRVNH`DDIoK0{NnBYJd=N+c5zy+#OUWc}>|T zx7mfCE-tMxnzud~@0m1JnoaKD&Rmv zV*2(xJ*Ax6E@^=ffNx#rh5x~G8dahne@DvAr zx=SsF)=z+fv=&M*KtKJHvrG1e#O& z@;6X%JV%ht(0FJJ`k5az9x7Pg^w>q^&{A1tN0%rn zE+0{9At(lUQlV7Yp_tC6>P=>bOI?e@M;L!TIw6$?C)T2@32$clEJo~@^{pD7V9qgq zV%OAi>+B=&2qCet${LbWt~t|i9TL#vkwkwtNeN00W!wM9+*w9vnQUpBIB|D(A?_X^ z?(XjH4pHJB;;zKq-IcfzcQ@jmkbEz6H_e%$`<(7GbLRV(RX=J~Rq9>OwRhe7-lH~< zoh}eDb5gs}%CI>un1RK)VjoXBB<@9f}Ap^2gk$mLi^cmD@y27`BL z;P$_I-pA4nf!kGBeorQ2w0LLRdQie?;*<|($1vL?ITh@$wi`-QS@pzk|1 z!jG1kD%MK)9l>^d1(lw{?W(p|^M?e=wSSG~szgC~~l8w7E};aVx2 zPyqXG~Qs=-hQ@pbw}m5ef7dvFLuGE6<4*b+(cLm-0k+^5pyMITUy&;)vACdg5?+oI`&b|EI@jFu0K?;L$)lnV(4r zzKQ_8ss-2?7#V)o5d2n!3Fr{>yWuqJzgp4A_Dx0dqhO#!1z|~ zoGwkH8^*<`^HFg0dU1AnK7_=vi(2$c$^?nSImtj)AZ=XB+9D**P-cj08UBYv@wezR zMpFx6T9=;ju-YcmCz!+R+Oj=~arLIm(TT1*h^@VMGZ`bEV;$Z=|Net zNmD7oVj{BTQ1DQ-GxnVxipO8B*D(yaT9aK~?dCWz99$|cpqPC;u%zUJu0CJqmTqYu zYTA!*=!(A}pIb5G$>_R#H^{|liA@M`QLu%CenB2hx`Em^mcPCtOEFWA^r95v;yy{& zpC*9!adnedZ%0x5UF@t`Y4uoE*-qJ zB)?m$C8Z5a`H{CHg}EQPUW-lwU#?{A_gH%|!!6Pe+tFo`gf0TwFZV6I{S-v0#nbFm zCXO-yy_F=VG1}*RQWY2Pbx3NppQE+Q2hMTc_!3Wa7pUmSOxKo)f*-D?yj))=cL5A_ zSAd~v7NFptz~K@XcM6ko6$yf!SNHwG9csz+npjC9nA~BV7nLlF9{6P+ttR7SBnYQc zj<1!;m=0wU5)Yj*mIIglo^KT}NM3P?iG`JG2?6eoMT2Y*B$Is6K&0`2BR_JTIH0)$ zzIE6p4g>6TDr2T9*G4h!SWZ4wN1V~AJnboDxeG|CooRr8ZC5flX>h^omfZO=Z**%d zs}Di64P(+smwC_Jb%cWULd(F~N=z+k-Kh6pm)x#AmpOD9%Xomb7}im9#km9O(LM6l zU0!36E@v6?mmAIM9@OR^oxNUaZ>{u6^FZgJYwrUgC8fivKpV!T*KEo5oDnLDsNWn} zb5>E-#I8*X@vG)imCtsP9>0xJM3qFDFD!Za3S=Z>a(L*?W&1p_Q+gMA`e_ZX$(zo& z$vhmRB7raGuWeMh)hwY&WG?K;T|go|%xYPr=-~tFbHCW$5)phMz?@D#vq1)Ty`g%U zjer8>tvL&1GFHH*Q1@o|*kZ(_#{ZPq$cCdR>?1~w3HAX8f>jYuQkjC~*7F2DGk@U{ zH|m5e66SX3mdm1GC+ulrJGP@dVU?O7MjbN8}d$}P5yQ{CLYKU;rfCMME z%A%Ci4k_NTZ#9W=Zf9z@#~VMR6Cps0D1Y?O;R_n(!N1*_*+rVmJi>uNn>&ix!Ivq& z@1k-&tKQMdXNcDH89yr8K5H2pko z`sr$wK*N4AI34?IEO(t3ijeuR4x3EO5KulZQVhYeA)_GPXjf^1D>`(bJ&(07t(o;j z*1r|I z*EVKzlRWdl5sxkXIgu*k#UwjFLNF4!Q>XGRGV@f;n~5!%kMa^7;baRRgtHr!{ATlm zCB1H%&b>q(8gKC-k1Ra#Us_1?t{4jXY=IQbrh1m%B{#omUzJ8R0Y8*nWh~?SWHykS zg~ZN{-fL5RBh|)mo{n*<&)>5%I-14Ydo_2LG|bKV>I41e_4jwFHUFelb9|AaS z5D48FcLG{!NAdB+DOSP=rzUyea~#Rzq5K-Wmc6w7VD##BJvhh`mSOKh9t{c`Qr1RrjU1e6^I63K*9pQJ~%dFI}=N&*j_H9yNW&3j7(HZ0Fh zW}~@ccUnzyf4NnA0G8SCko%U_+|N1TG9}mXojD6~d#n~&;?S0xYxQk+GOgR*yzQ<# z$Gh##y=r~LzH7ZfAnJtA{%~;-zOPW&b{7pg-$xV4REW%$MulBo;IQWio*=+Tla-q+Y}2vlJ^IFR-iygi($m@ZAYNN zg`T%r-oeZJ-nc5f6^jJr=p%$OgfxK%0<~3wL8ZuQmNH>V6O982k)J9JY&so@&?RVe zJ1A4=l{!g_@IExoa3Q`y8k2NZd_wewXEx@!fiO;u_kW%Z1K;;tO?C_=mpR)e6u&^B z8N!qCG=ZU>eZNVQ`ym;se$^4aRJNOW^wjOL|IGNy5!$AW8>??7g1>UKA;7uf0Jt

QKw(be}U8n|9kCN96n9|FK{)*bzGvJG^E)c|zG~`5dfvcE_gFP9=*3fZq zGE&2z1702lAkt2Dr;wi84>deM04P zFgVDT@dPi)7~K8pfzCx0)7A<>4`iIj>qyaLtZqp=OJ?Q#M2v5Cde|8S3t_NI(|ipd zG#)pSpf=3ix@)Vt!*MuDEt6j#ZXoe%vd0`XzvZsZz8-aQQciA1MJYA{sBZ)ed z({yiGfS0kYts2uCfX>}ha}_SfJO{COSB*kP>Y1g=8nT>Cn$juThIfH`-U$1xwSS3P zK7|U0ZN}^!j2)s(8=wbTYnic|Q|IyYky(;8LZMp)F`|mg9n?%TlMJewoRr5PgPeR5 zjbW~ZA+%b6PXJ0rPHvJiQN_JT&SSj5ypSDj*#k0P9GpO!V46kGEplTOSa)l+Naa=* zizEHE5(tlY+se5$L87mGl59Eajt!d)?(p=8EeJb1_LH+mSiR<&@YLaR2|_K2K49XpE9er#QSDs0@*80wcZb!U6Q24OV!@ma9+U$9{aS?0tU zXvcPtjgnH8v8wcEtf^AVCRu63Nf8Hz3$53}Y#Ua=7`rc&S2imdyqz&WC84t4!Lgwt z*SvRT(sJ)p9kY)^sZIqT9CLQ{7w=gAgE0Cz97NABXZ;hx_oax<74}* z0*r97(@{bjvwSk7E)nubA!+5lgB$mqUG*UIozlfdD`Hc(`lV_(dgN@7&n}m!$0VDI zZE~>&@rC}vCA=N18+B0RcF70md6Da;%;|T^DmKGA^|1mGVJ2Qx5dU)M|3r_6}k+elS>_R$rCUSiS;tT2G=dWb71+=CdkTaFAgJ7n6v zu8eV{y=*PMElg%x#6c7J2wGR_A5tWwNJX3=(1oBr>dV%>JSXSken8$?fZaPQ#td$( ze{aG*)9I}=($LY4(tO-a;-J|V!KP)OakZD)e_tm3VG$o0SnVwp`(9Aw$PsW78@{V3cqA0BrV!SGGT}kRekGOh$(#Ib#DS zHt$@@J16K*hRVKlx5W|edZO=0)b87JW?f2mNnfkqcCc6mzQqW0KOQCvoMx*hc4qcu zX5G~a&JrTq!R0Lr2M3d8)5VKTtvkMUd8A;3!FM~?U*$M*jNCId{k z7}>rR27X=U-@;_T6n})t-)8<9?EO%-`Dfhpb4+IYul8;Kyddx2Dr98&?zdfs;QX^I z_H{kK?(#2VGQgMQYwhbdI{fjT{9qOK5BKEHG5LSF7b4qV4^aLDw11B-e%SaoQvLn)z{#>mu$Mq5LJmxt>F*4TtXEyM%1(=^$ z(d7`+M0(_8qM{XgMRB@Zma(;9vw)UCHj1?Pkb>KnZ#Y9IE_Y`h``wBp-(%+EaBHMv ze1=w+SesU2VrQzE;b>K;8(E2Y`-F|pmzdP)g(zx@BaR_$GXj4r)|Fp6fL+Dxf&X|m zy>(JYqr^#fy(a&e@g{kd>Y!kDvH&9@2TK03=BbbNm<^Z53XAp*(eTtO7et8eDkE`c zxL#P4WIaa6OHo%tP=z?lfQa~$8k;|Vwm9E_AWnb5CSphHX;%;O`XU+$! zgV$sZA$+Aa%2Wq${07c=@{M6<*fgrm`W1L{+k60GD-+SlND~qRpMwY36REQ~AF;#} zGk@nAXx{uvJCHZA{gI_=!@ZtE1&H~^8DuCLZ_G~1Rs2X-P_MD(njJQ`=D84@3bW2> z+m7T1649fRn0JWpsVFzW)6AD8QY=+`@C`#N2_>^ha@Lx@kQp0v2L-x>GB0gP%=ac% z%+!t_bII46)zI6_MrJKajzzfDuf7|m2~EA`tP;$*UMp#>&TS9cf##uU%t<-9=+QYv z4}2gm7l=a2L9w>&vnp4hp2TfOib51bwM4G}V5##ZI#JMX!qu-?xQ6#~nb^(R$z0rL zyLn7Ydpj`aMI3EemvOD$jOT%)ZHNAr1!MqI0Cj$&fn&r}nI!sYdrJ|YUL%p*Y>7#u z=%Lh1EIUM4y-d@bRZ1?XAbB>dr?7$wHWdVNSUz>dkyZF(a)+trUhZaC+yDzVA2Aw^ ztHHJtedRglOl)F?t5V~F4L;|#@wtR1dE*uM@WWJ$-xYH8WTOB+FN3&xg8Pldi^Fmz zclr99VWSStgBCojGPeAW*NqgK6>+=Eqj62SJw6tm!+z-bXy&)Z)U4qZ<7d@0lMaPS zIn1(L`K;j>K?d{lafa0&BOYCho-wUbcjwwOD=F_TKE7<19_}Xc-)`_=CFnC^`?Rc( z2|(lXF)c&}1#^F~p57q8zAJUB5E1SB+2l1zhGCHVaLv!QVA%?GSjSn07jQ_x3Po|Q$#L51Z1A7jJKZQsQ>;NX zfwgpFcS5kDA6DC?-f6%p@pUK2wy3accbF(D8ELh?1Hm~HpA58!zQ*3gi`ktAYi2d6 zkLcLem)R0-QS%qcDNAmzEwN=)05mG!9MH;ycemyBeAYOaa|aO;-tbg%<(k0iU?&6-i>sR?rLr>Snfconywh`%?AM_O zB^_7fxcFJn%m>Y?1cEy*QFfDX8-}LAy@Og_JitWj6tT}mibat1S)c96Fh_SqkriNc zYxIYBP|;nT6iu3xW5cVwo`{50gDaP-K3(NY2rHM!&-W27PA6LgZ-6p%FsRsRm|z9v z&}@KFFKV#|`&>1)REHOSr%ipwm}a?*=u*;V4J!`;wmw^Yo%y^CW7SNPMlXyI`h9%s zb0Ighpmw54fykAZvi0{k>(`q;?7A-HM!-+tqRsMSP0KsFdnhdmt^N$FY8| z?=^NP^1{LIY$!OKm_j*0TX!o(xe}yaD{+uu%(dzZOcXWn>)(h(f-5Uy6oh&^yY#s= z?K}?V4onF6u8T37Mc?YkLMAIu$Qb;WHod( zbT$E99;?YH5SOA|&eW6VJnlW(k<%2ro@Q?xka%8RfaPUlk>;vDQp0nvRD+D>5#FR( z`llhY!-dAxYN`jHtV;sn+hPIepqtky0tR zCtvjh#%1(S-bFIDp*zET9@ZsBMlC#jk*LmfskJxY&a@3S54c0vjP^ZD|3`Ga~1X%Qnqm!8PGGeV)#L z%i>kmb+n4ZxgtE1>H+a4b^7ud!~n!X2S5J;jroFg4PB$p$cdpKyw1b;o<_AldI_(_ zhG*&sCf4~3erg2G0}uOC(I@kk?Jl-Uaa9T$gG0>GQ_#3xio>nK{%gHHXe17I_SKo? zIvRcG>to86XQc!zH8>dB=rq zr81ymH!vIF9Y#J5Ji7&#;D^T5{k`IO}LiEw1r zA!x6h=Y<~0RuaoWQ+6H$wsb;lQzh+7I3VIrWdqtdHOT76dH z4E+hWho8ZpF296`(l%44DB51c85Jy!_BThmU^Hv#gZCZRhfg@I zS9Sd*CfE(#lQWNBu>mi8Z#JH@P==UbnQ6X+%HoVZ5DVCR-%u1Et4Ndlr(|(+@ZN=+KziM z&fgjs9d;Bh$XSFObfy5%i`Ub?LXNwz1IeLCohvSNQ7M1p(h%kev~Ycn;zQscGEfL_g5oZ9l;B`qQ{iQ6=e z9u&<`Ny6m0Eb0(%Q+3@l(^fEy;3n7^XM_EpP|_!k zT9uc~6IXK|@e_xk!1DrAB4HQ9h3oPAR~678C*SgB_Z`d5p5ax2i-lX-7~Tw3=2u@; zyR5L{Zs3J;JjGjb(wc&o&2ELbHwrpxkvLekK+>)20*}iTwhtC{9)zC9*2cqUM5L`b zm_+3@#@5M7V99~j)5)biUVeE|P^&FW#s9=&eAogKa2I+F)VW_8g+($qk*4`wM2DM_>3V zxd#C=h5ZgfP4>mfg>ec3+k5GB+t#T^puT*@pS8~8`1(%!uM^SVg6^;8e*eV2576?Q zk>giKtRE2L+t5Fu(*Gob1Db;~a{Q_V{hP^Sz!ZPP;NNEcC%gSK1_x+czvA{+i?i=u zet`9U3vGUID*WF}9zTOhcu~J;4+c#5%zN~w5{qxk`PVPpKO#7wS2rWa_Za9qbAB88 z*}?svA~>J{>0b?g{usf(P4>$cfBb#?7JC3h8o!I30sTO~|EB-!zA(}MdXzsRxVrs3 zI|@Lgks{3K6SmcK9;d!gO_=FDw`Apn?@RMCga%P`B<|skhmp`UxasBOot+}S+6-x)guQSGJONcuQ<0Yp2~NDuyj?t6QWM>zF5{28 zB$>y(q87>za9RBAkMdRcynuGchwmnB16-`8Qdl8u*TGFwB4MJ~ZK~KMx)6F5^Y|Ry zAZFTK=J)Pit#^*!;J^K}pwV)W_L8rk)^* zCNP|)P4v;&K3iQi(BUM?MlvAewVa{|h8)cx5~ZzHnNH|Qgkl)=K?m0We!T?|E3f>Uj!yQCe3^^lYD}jD2lL?7ERUA5gb!UHv(C>+ zPB(@mtLUMgcayx@IH|OpofkzS`S27`j5@tXdk*OgX+|j?#h;Iqb8~mO#;RD!T9{I` zsFAjWL%i&T&4-MN+?z<%`*KD+7q8yP?=r_Yt6B|B!o(*O5IbPY9yQ7fNrtr!wqgmM zZySkM>XGKVml~0pms({4^4omuu{I;v6H{PO5F^Et83jY*;db}&hj7#01c65hm8LWE;`OFs=ZRh*j}U;_^K43^xU8-N~L$}|9Na~DwQ4hMStiSr`G{#<1q=44zjF%Etf z1YCrvv}Dp#7d1^=mfKC*6;hdYQEb?w!uOPNjqbfoWR6vmj@nXiz*{kP=oB9>7^G3j zaVYr-;14Tl!qPd+Sdz|=c~hJhEb591?EDH8M;xJvtgtEZU`c%GtQ~M5gn7|X9fmbA zP|<25)9XAi>*_=L^f9v&1#ezoJgvDQZD?Mft*--D0inKGqNcCYd1+YF$?GvZ)FwwW zO}WLf6{<8}Zk8NKSC6k7@&@`OGrqBVpS6AWSg?emI_t>L9Xiy93MRXo{Nq8~G|TIl z`5Ll~J0KxJBuj21c(I$r<0z64^%~r2*>xeBo#M?4j5ukl@P&ZaMp*SOpFXoRl_mCn zJ`Nt&gOByh0yqmHP7U1qRCo2b(7q_Nlum#Y4<_sqg`Xrx=hmj@!Z`us$J#zmrD`iZ zkJ8m7=jh|ui!%kqw>DR9j8`4JnqZ^gfmtGXH3`=dA;$*u&O|#$|Lg#|@UE(d@dRGX zb{1XJO}cyuzp*xrwM@z33U3X2OQBl7aED@Dh*pBKAceHotQKTqrp~D!oJJm&`macpr zwYj~zxEi4*fw$Yy^BEi~jNS0K6WmW9b(s7!Ps66|3=a4ZQzEhg+8e44oM{zP&vEgOuXi z(9baE|0pBPz(BzCwM!hqH^s-_%18sI_yZdLW#)gf+dre5pJt?)0NuX-DwO?O8ENM4 zp3qf4lT!S$o?mzQm!%Yd?r%)=-@}pLaQK&@pPk&Fn1+Yx<_Apijd6a8P0 zBK{C&f0^u;1^y1RztHH{Q9yH>??MQsZ~pMVZsyPa-AoL>9>k9@YYu=}>@Oo_F1%5g z2vJ++ZKQL|bF3e0{j9cA9gKx%fr~0g_2vxbwA&(J45E8oy^ZX3%S*61cQ|$e@0WQ( zd4n3`V*RH&O*%}Fyd_yeLH!y&%Vr>u0Cn^Fh;sh|+hE+{L1vHD z)B`Q)uX+^S7Wvd;PqDBkUSN>EdOOz&N!+8rs%$fb3sLgNAd_hyFt&j))KoE>KY< z*-^*(&*{kZBqu#81rRp=$0yQ-%<*riiW-W+Z6sFBCiTcdic`@83`YH7Fe6+iO4kH1 zb54L)IGfUWLNN{(fP0$_v#+i+iTtHru4y%t0jMeeEQHtzvuAA~ zuCzJBvMf}(5#=X}iO?1g>X~s+#*aE!=9r}17Xz}G zy4XAwW;wSWX4Aim&Hh@S?FAYf==%V<;GTNxxigm!vTJtho}$Ez*nEQ_1DkzJPV+qr3MSZPRm4Qa=P3k#2QdIgL(CKj0!T)uCM#S|{D zb43Le;a4^C^=KN(20o+JNyi((v}vY)37iF|FE-&dSOoHi^>}S0yWJefW!b~XZ++?G zZk{A;ak0dV8kI~P8s|IVD_G(A+|FYNGE7FO8E#Vr-EKnxMNdt{W#12*N%PA=jyblb za{bG)l#gQj3Y~6Yw*H3m^Pd$|ZmCk3&t`1G{f5EZ1XyTuaNDsonMXomSxB>z#err( zbxRmObE9fHrzAD{4|Wz;0%-)HD)F(>J2y~lJ}Xi$+bOHq9gmM_EWEF$h$P_Jf5XQm z^f^N2@d=shszDeyqHQ?b91q{1kM}a@m|QC!e)SD!pdBUs3JlEzSOd|7hAAW%)(58; zo5u&=muSTT$Tpy^BN;O}Pw%nu+cHUMWBpec7AXTB$=w{~Uf5mtXRa8~+s5i)k#s>& zUYrO}%X~5qYEeS1MW5zTA%Ds95R=VMgLY)9_#_-eZ4)%~)~A{;^I1fBPSHI2ySLB4 z;&k0)P*!P=xQy3uQC2lk*jcs|YFxPRg?Hi;#ng2vW>^C|a1?@qbZ#}I=eJ**%OCf> zdt?cBg?nOh#D>_`b7s|*Kn)Y$$`xJrfSPRQEVG;;t#*T}Oz|h& zt2TMA&)1^QU1}<2ROc4nv0Oe6K_nfG(5mRNilGmQuu-=04uppe5D{@gkf0_D8Z^Pp z)5LS=xTp~_yjxhS0z0&Wo;?MIhO{22EpFYHP^z`6=VAvYv84j0%@$|I)1H;!ay9_NcVdAPg0iYtII8Xcc?f9PyBRu<9;BeRf~%-VFZ zQTz0GX!g)Q+c~7{BGs30aw6AUE3&Tg080;!w`Lw6K3;ZRYx02UE-V=Ipqn5-XuGne zlXV>lvEP5&)2rT<LC9Z?3I3REzfJbb7Jp#dZ*cl`l!1+j{rBVe zL1Of07lsMY>-k%h^*h_vsA$SB^rN|5m$66W4n(osO}8S7ydoHRcS@U#ebl9brOs~< zhUs2DQr5=9Xyp6~F7lA3Fq9d~G(7kIVlTU?12>G$MH5<6PHo&}IE6ifipF#ZS4{~d z00%0{OE3LBS~Z3S9PJ)lmn_a4jL&w~uN@j~CAYlZ-%bv;z<=$~cw)t>6OQYUxAY2Q zP76nfI+Nn!kmN#tC2!7*UQ4CakAqnq#-BFCf-_+$N!FO{&d`Or`_9Du4EmJWlDQ15 zq^93MOs1~#G+%aNK)aDVtT&84RD26O^ zMQRhO%^AcmYW--+rDom)@I&h=USIURBV8-v9lTQl`tb^{Nul8)n~dL2ZnMy;-xt)Eb0VL6Ga_UJ%z?2}xf+HB6Yf zY`in%9*L=r)e5(z)3;?WP1peqG6e2$)pLIfvw;fwe53tpF@dmm2Q=^=Qif`du>P>N zuQ5I`WAT|R6z2e+hDvOG_FL7Rn8Hd$LH+g}uvlPUhNu0vIc-YnC+R>qoZ-g->Rxp0 zaSdouBZ|CP*F~s#(3B221BmoxXSo^YxkERfJ*dCXzaa>rVr*11N4}+;vCr8f?xQP< zUB=;tHH5G-;qcxP-~+NknLq8)f@tS(!j0frIMA}dWpx6MBxWUOMw6c5R~f?L?PJ>O zvZqoi1fv$oc5arkf77NCSiPLH$KdtJZdunvatM@(6E$`SYw>dLQs*(3RC9}arwB2X zSdq;^ml)Yq6`OrDDe5uW-tWTlMl9|^R!?n&slargDpVO2JQr4lg#$xnStj-p>0&Y& z(cS6<7{W6;YALJhJj1gI{8OHZ#| zUg_F6KWE!}Aw1vi%lAE-4_~U_so`$rUj3! zyiMmuceOc3=O*0x-edg=6=r5>=h+k0GkkpdOt7n+u3$Q_(i}5WC(YQm-fxEb>}i#k z;D|?`BKvnG_4mC#S{P(BE+}Ek6WF{jP5RRQ8R?|S%t{X7-D6pP1G+9*;jMkxQqeY$ zJwVH}o?&P$$~UUwsXy6_7uoz+2$!zzeJnma!x8jJ!ys`^oFtY?t^P!zWLjwd(Q1}y zQ8E8=;DhF?8vhEh1IwCEatd_Wkf5JRP#h=;d*YE)1KvH(6>}XB_}OSXL&&!dxk8_A zUxNu;$HbG4?qx-8d2hk`3vIsld}s8$jXC%Zc%Lt+l>WGM>yc}3pX2oCq-HNTo!(8n zv9UtGkNP6DJOI)IADOKA^^U0%m?R<%2Nu@uNzgo}!=xV%%eF5PUeVHtFlAnYJQHBaC!We_1) z2`FDF0~{6lsD}L*!UgkSDB}tb-i9V)@KVr2iQ{8A7sUAi<9niG-a&*ye-LWjQXR+Q z;_NUdieoHMY$>^Eu{u*JKIxaAg6EC@Bwl)kTy|>rw)oiqXw~9w-_lEr*;Z@;2LAiq zIS*_2wmw*w)A~a_eO;th(SU?KwCjd}j0ksO8BE2G0dO|7vR7O?>^lXzq<~0bGtObX zLVXH5YK+`P%3f{mo@R@fymU9k{P}Z=SIF)gw};86)&aqVCjs6>q1@?o^jEi&k< zk0a@%kln-&kX!BHU`tk6`2;X4#}&$N=RMf)Rz@t*NLitR(aD3hNHe^H_9R`BtQgMj zLp)~#cGfnO)7|7Ty6DExWz-U`EKa@R=33)G7CH9C2nR?VdflMZOj?<9wpfPO2{X{2 zNfqhIPmuU!8Sh?5SvPwTif7`BsHT7}l2}`Yb6PT_6Pn2qNE)M3ugWQ!K-#JHvyhn1 znTAl zDx-XmYD_M(5k)ydYr}bf=L0rE{qGrLrf<2ke;u0smYm`%(Ei%G50IGy1OPS+^Z;Qi z6CfM-4HACDj{m=k@8|(SQYMC9b)SEe%K=mT5to0P`Db+bZ*e)pUsuxpJ6vAQ{9VI5 z^)mql)3^2f>sRa_1QhfDK@Ahb_rU0P5dEWw=Kq$<|9V99CtUu!F8PNa$TzV5#^oH0 ztly6Vpho)F1Qd+F9_8;`p7WK<8^8L(%|_58zleKSV$0yMxuqf<+De&RhUh`72(HkW zgWFnJAtm>dkE*u8Fv94Iqj4>(nNd5d8)r+>=r5J8&27#R?fOEJ!CGcYIes z4GBHLzJ^I^+g=G!P(+p<2}NT|yI}Auvt{G1k%u~!wKi5gMpUIcU)_x?ew3QA=VYp- zlr-Gtj7QZ}vCn?t{HWzjid(3rnmYM52aA*bvQxo5PcrQNPGJ+%G^@%G<3ipY%D83?(5P2pmhHm~u@i52U7fBpGZua$6deE`Vk% zv&~R&jGbC^s_l){lAID0cn(X+w)mt6eCVa(qVW<6GfefDns7w9nBt^JAH}&>Pn}K` zknY~;oz8B!e7LC;?}~(1CQc8c7;<}+&jF7F1gZkiKZvkq5k)2ByifsN9GoOk4il`yh-c9B@D;PK+KQsg9oJ^af1E9n-MmJO_!pq+p;ktd!QZZQ=s z0o6!3oIL9O350;r{fApb3DdTNo@aR}HOOfXevkFXYgfd|Xz8$%zOntpKl^ zb+_b6N=jrMqVc&Uq=(h=F^>75 zLgE}??xME~&Mln-=0P*1^2OIMSv_ap5TBQyRrtkKZ1q%3A?p%$e{1@@{l>(q4;cKz+ppDLg*AZ#^BV1YYC(p`canJ=4b=it)CvkvY&gsdYP>vgW``7lwCtR|E`*jjAQyE2{O@nl`b@j?v*}=KXcO* z;WXk;Ogl9*E~71Ov{o)_iA7V+c}I9LeG5x*xY%9LkpZHP*=c)30~5rUu{{;r$-cav zqK**-9$j{85eXz5f5U>}ZB5VTeFTb4LBuNCZEBIjJu~ZjBmg2?)Oty8sNo`fD<--C zr*t?llbK{$YO86fcqJ_N(0(DaJg6rz(`dy-f7D%bUo9VQHd6Y^SN&2QfOhn#g7IdV z-gl80T&Zj3VuYLQ2$?D~(=-_1dmn1;19ZD@iV$EB4g^pPu#>rG z16L8s?Hdo8I&4!HYCsAVuACz$)h?%1os11V7UVL~3K@rSbp(ak1>60wvdWSTsnnhB zp(N74#$_Jm>i9FG-{lPFR2y^6@5?BJkESx$H5J z=Y%>}OuLbl)5QH08|0x>2~2iO-t^RXzz;MciN+N#$?9J>upop?! z3$r@to3N=TppFYVMd+r%LG9;0CoX5;y3e!{Q1+#WB;ww@&;m~loOH0;P%>36G6+Hd zdaG#HrxanRdT5O8v<9$adj>bXqoewTm*4Un?{bNMhOfHO5%}04meJLwP8v(<0a;a{+Yc-!+d^?U&(Nc0K(<1u!u8f!tk#_@O2W^);E;4rM zX8AN>rs+O^wq8Ja>w`yC%j}0g_*6rqvqJN(PW3HX(=e+v@zum)UzZC}g;EjP@*)i) zWEX^58osY7N-MZWo%$wZTzRb^=))=5O+=WGKCip&Z6J!m{FXE3{Gffn1x!3A#p(Sy z&#sL!FEs-M3)GB<^rKhmt&xvPlv%&B`6q4S0r+b^e_nk)TWC%lb5rPfpA%)flZRnqX=x?v>|AFS+0LxpzGQZ1a|Bg`O>lA;4-QQ;ZC%gT# z?B>tY;f#M>J^SxqH+|i2*!}J~g4a(XyKn3HyVwn|kziu{9{2ptnLoVZ{KG@SpJVrL zSLdff4S+=Ye=eE*ZL(iB{UbF07A*i&$KMP8m>7Qt>_091WBTvVT%z(-sDT0yYSe!% z{EO4ByjI6Em~T!(b81?0$|DYKrxLOwP4<1fYhm=}_9C(1Or)=+KCm5a*}=IrJ|K4} zQ&qI3A!Hh(2|`DvPxPfd*U?kMb3w>KBq0PE)?K`ty?f4Akw`X0I%|T@$$YVt&|BMQ z97zLwB`xg)r-#kGux2twl!nbxH!LWzL=Z$N>fP&|+^dV^PoGfJsNd&Y4iD2?!YZnd z4hPO>QB-I4WxjQ~a2=Q}%b=$cN^Dq|sK_^pUc)OAHQhuBlEWJx>ey>kQFftEg6<@z zk@}E{Otpf(bG2tTMOVR1k-$I7v$xis7g7!5FCB?MD({J+uaIhCIRr!~&EYDYN8TOeG!sFrlh}dq z5!R0k<76z@t@@rLC5+g~o)#Cfuz5kZE4ZX!F)8tiL@3tdccH)4!sr#hDFrL5euoT% zu0iCxqtS_)Tn<%sn%G;+tJfoP2ud5ehu5wP0{DE^Jwq2x2nsp|+k|aTj1+ncv;$Eu zrANo+EnbU$j09WkhpO^C4dc^8@^PPY{EG*esopCb)oRi*tC%!aFlhdqP+SAmhEcsY zywOo-SK$=?~nG(xMNngeG+Arbko-KaBz{zKC+y-)+ zbeg->WBaw|o34v$5ON79tyKYup%|PXI1Q*&%7cy`Ll(n3@8M%LVBR_+q|pJuQHYdgt!~T3=_#c%-t# zh=UQ9oQ!>)w#pQ~UY2}fFci~$Y?kV7Y|nFB`u~`F%dkANWNkY@a1HJf+}(q_yE_DT zcL?qpT!TBoJ-E9C2rj`jxbr>DOlQ)2(mm5N`|bVx;ehp{j(gqD0?xYDRaJ>4PSqiE zlpJvlA!xk=f!On0RwZE@T|>*@SlE(Q1m;RuY^9U|@gcQ#MZzubmeWv}7m7u#;k9{- z%w1pJaaCV~uMN$Gm*`Flj#zEi#z2B3g0bC_T-;>1mo82s*gRPDFSouL#dX#DCyG&FnlzwOEaDBkD8xP^1K4%G$UlV$f+UDEqFwAWogD?T_`CbR15$K_s--z(Inh0_)cYuOo&V%V!3!(`-EFR2tTW8Z8UrzcQwE5!Ap$n6Y2q}Q0kb%BcN9$ z8YmB*D4nyX{IYYg-vVcT1hw#cc^!lW|nf`83Q%r*}OesGeFdzjB;3rvbnF0U;p zKze_4Y$y~{7=18liTBcjcK#b*dmK8s1hu}uNeKPxhN?7n5mfoS&YblA7tqNXn$?BWq5P<`|Q7S1= z%5R5m_Xx~%2*;%guutBDu@|A!9Hf5*mmW707^}sQWud`*pY$2D$WZv-s%o#N+5R?IXr_G>tW#&g6i^ zIg@UjaH9t<>Ym9n8oF7V->Y_qU~g zfswzG&tCz}fOfu&jNi?*{}D6;Li`>yKS%x-yZ!Uf3}|KYXW{R^f@Ve)n}Kbpm;5h$ z4t~~egYo%!{!?fMG}vZj{GPRZhvw&{pS`(1hUWjr9uh!nl|TFDe_yM94)(V#{(#+# zKV%DkbTk;5{+^Nks7eDmYp|jH3fQrmvtC0ao^`2#Ot2Q!Uyeo@J5fJ$iwaf^OzO0$ zfW;!FO zd)9b{ab(LThLpMffpeEwZ?#YCpu7R>bJptP@Oh8e;tN}}dxkyPb&(OhGh%6vgkH!I z{E+0#9zD!Jx)kSIB-)b`Omry@tx)r>iJh5<*u$M!D}ru){EuEJ;w^p(Z{9)6VQH_6 zTznY832#IJ)x?aM?KK0jPsZ|ev^@xfO2%_7)etL-s`9J}7dhG7a(W-VTLY*lz;I5lMVs5>secAezN?kq^t zDuc?!jp4ck3jx!e=-l_UWcQ_W_l|ZcZ>IgiLESC+;|Jz5n^AF_)V&EqlTVf!ijh@{9olz)8$*z z;XZk-!(Q{)(X*^KsO$C=$S z#1TQV%8vS{bTX?bbc-)bfUClf=3<&(=XkxUj5FL{fp_7450!|#NT!L{K(U%L37CpvN@8ht8>>xd?vM*{ zP~^6f%%hb)AOZH4jlO5XD|Wcq_OwKkeM3=? zh!>*HAUpBTufORgIOlFP8gdEnuS$^Vfs+Fko`IlhsFhZj4@NM=pjZ-q<%6sUw1wv{5*z6y?^IR^AH(!0SO<5_}( z-cM+?+7PBjtSqd1${`q7%7S{J>&o&fv3)#?7kCm(gwLjmA#|gX1s@L#r(#ndGuf)q zmx}7i1@9}S^X=M{1IO`|K~YJfD1Vz_udtJI9NZK^y2K-N^d{tr&KKQOFL*u(s(Sey z8tB%kCj8n|&)xR@G8wdEDM5(A2w0p7lmzAt-{cO@d~6iY+q;Q!Emc+iyZR;;Q3r3I z^-y*q(f4d#5#<&B*QU{v%36@>_7L~wvdYm=eodusdk%51Qx{RrQYiMnREqyz;D($Mzi4#(n8>)$Oixi(imTR^L z;3lp|n_=^#M1AR{T@I&^sIUX6$~^bNR6>18<%UvBbQY5$Wn5M>ih60N40*?%h!1xt9b@+m>6jvYxsNE zanYKnZeViG53sm_V-Fg-YrgF6amL|9Mt(g#D&5*gUL+w`h26-ekrOKpO~qZD&bF!C z%?}x2)ikhrw9nUx9MXs}9WJspZ$N~$EJk}T`88$^+V|;59w_1jGMjD`AL1K{dTKOU zER4k6OfYC(lD~FW-#h(&DfPqXY=y^mfA#AtE`ixHN-s4N%B5gqwAa>`ukN5XuES%7 zM8D#UUtaUCpiN?6As(yXwXSO7`L`&o^JaY6ns+Sq91z4682r+djq@2C-Qle1D;QmK zf?JcB3{esrV^m948SLxz<`z+dK+M<=y-!4gL`N(78f(p*CvxK2D2g#tUcHxCCLS4u zusiuNI5Q?t61x=+6jckeO=xIXT{{6p#dUnwLr0R%$Ef4-N0Uium9{NQN#E14M3yO$Br2JP>9_kYC1 zfDpgS#6KvAzmY?K*zF&I&(DQqrf2W$U)2ACj){)>d&S@QPyHLu&-0%$F*88PXZq*d z>3QjA=j8vCiJ6|Wr$6=Pf1in;gZ&c|Kcnt1tMqL2|L|`x{gO5O{dN4&fnfw3#^LkK*+L9rJb?ck4Xvqzh>;9=R!^;xPRU4qtoS%^*^V_9`Rd)&{P0!Hf3 z;@%0_-(Od1yp`_UV7bJ9#UGbfgINXK#2S7}kDMG>{(+8K9huDE6hf@MT|B9VyOz6y z8aMKN65X4f*7Nmu@5v$;!zst9!H&kHGXW)6yi5Me6+L8-`+5%qp&^-q^b!P-0-c_i zg5)r0Wk=l`xT8L6xu0^YL9%LI7v+%sM)N0pzms7vAbjt_^_`94ng{+pj>y|JOI0=9FrK~8Yx6ivnuy5UlhoCIVQSuRa{|5{zFP*FI|-W^ZNjGF>E ztZ%>Np5vXS85@BOK`p8pi$edF_}(`tH;Po+@OT`y3Cv~DEY6cxtlC?>>Eo<~lm*!! z)=C)y$Az0WmN!!Bxy_$;;Rd>8%ws*}lxJczDO^=v_{QfAh86^Kq~S_<`I+>GM7gIC z0qy`Gvje59z{DSNOAe>uJHH~t$;Bb?NCESh>LxQ)zVQ(+35hkZsVoyu{;ctqM0Py^ zHCc%jj;yLj>!Y0bvA|j5pD$M!n{RF6dLRjLH#qkW;laOGS2I2IST@ z8QLi=>V!OIjm*|bRnze1FVvM$yMV49N9S%vkBx6Z9okmb%Uq)(0K08-^dSF-@Iw zg`r8IUxgrz%IUyCM&XNwjQQpFRQNLo^l|95PzLO_!NM)-Df&vHEN{a5v`3X&b?F3v za2C=DBo?<5*fPpcxs_CTs0NY|m5jdBL z>Kj4YY?Ybf0po|YuY4yx7gK3NbK?EL#L4%K7rmj+yMUrw6=jC5Vsu>+l?MqU%Fzdm zN`HcdMCZj~6uBt_Y;Y>;npgCbhTI8FsT#Uuo#SAL;|s|X{OOJ?26WMLOPZ@a4q1@n z9Kq=ss1jbJ(?^NV-gHQyG;iTRG^b&->iv!$9X|xo6ANd*gMRPHDpY{_s>Tg)U$8xK zv){fQ>`+hiMv5EKwH*$BOreyQfTI_TH1T>w+KO;zSu}Z3Qso_ccs!~P?f!gJVOirr zW_#}BKAFuRhc?QU0h5v%4%7Y{@0qsD(1043QBWUT;m?aozRvntp%^d~)q|31*&-s) z>(iScKJ~rfLK$s!A|i#;XJ1NO93r(6_i>%C?$zG6CfLJ_3FM|V=MTZ(_b$R=HkVt1 zK80+eug&wvmtb!7ezR1CDnos}qOr1IjB?8h$yM6E@_CUe(=6Eg)hm0~D{)I?drQUG zlec{sJS!p-Y+U|uUECDBB}*p)z!IHYgOsH2r6!jYC<4+45u@LiY9t1z;$x6HL*L(m zE}gEJS@i%*=XWlLU?WMaBQ@zryO|WH+nF5~14gMh=pNS7rnMxT?s_adHW7?fQt$|0 zeek=*!3(HM7HkNEOWV&{&)31g?C8-$cc>Id6IJA^nFR0XIwCKob_Z!Ggv>Da3}ntW zC`~o^$b`$1d{qwq*#0R|S{+Dh7ruC30r$4J*u^52Zj4o^q`V}Np@a%^sj9qU$K!4+ z<%CYkVDa$wez~|4DwM#Uh8WS=j}!aCe?J>+GQt*26Jo!Yt;PsUC;ZEb66Bp4b>?oK z$5=sI#wV(9vGtrOsrFj={qns5bc9Qt+R_}}@KBntLK&06hPJN@3*KfA;3NbhKbiZ% z2$;X|{h!`niLQ(QnU3k7@3xc5h6>A#l**8H?t1=Ft=|Nmaq>2Iya090cHq&a{80KUt)&r3ghbAODk zzaE{RdLWqpJo5iOcs>XF+XMa}#y*qdFROs&$KMA^Gcy0+L;9nSn-MT+@qd8l^VAMN z=e1u3N^|Ow$Po?AdgkOwXprR*EhPka%32VkcFR*NWbJ-jAq3|q7zGm#!q6Y8%H3&B zlew9>0rOLkQ<^jrl|UcHxBw zd^kxuO_z2BYm;5sdHF@QJN_m%capf&DsSA)97O5^q<2XbD&vK~E;&SchjF@?hO*19 zI$R!0fVdu6+h_TLD*rL!x?{&joKJm*ml%*meWovSN~=>=1!t6pPjZ^*FPmDldfjYC zIQe1)r|pk59ah55F(U({?XxKJ+->qHn$GHmzJHk*a5wDY&!&JF zV7nt!CR&02njRhzSD>41P3&M}laP$2$yI4&hzA@*udM`>-iZTY!>*kl6KPOjMdy0| zCi8_GI=H_(eOHgAFK)fhv8oP14 zv;!SCZMORSSn)1Z=5SSlHXjqwqT?V+!4;@#!NOI%-ILpQrvsa?*a$A~m+&_SajT#r zm6ilBX%xXkUc!BTX*^cPzZXOq_(@*a^gu0GJP&K}7IL9t%oj~Jm5AsvhYo@OCc3;* zgnc2>(n!&ds&&Yu6MO@_1UtM0ixwgSw7SnWeOSw?Gp(k>g9NQm{fmqgedYlF9@JLh z2mWkFz7T!8;9}qoP5jAZq&-Qf5-oc@26p?CW5?STbG@cXkH&hW-G_iFDj55c@_qg} z+_XVMLndr>a>pWuFMgg{TNf~qX6Yf6jS@AH71c0-XGR^?0XT%@QJ{}#@s&dn#7lg* z7v%fy6CVYRga|@Zn(?|JFD;fHSh_I`uLruKIaM=)Z4lP$rITd3r2Glz15b88F>Jo| zO`-0PbE|>#0~QIw)0beU`vH! z^N(>{r}@|;zgTWt@{|FwO?kqn+&EAd%4{Qia6`)rSB7jXq~hNmCb|xu<|>0s83m5X z9E*f|JzMd4TwIl~{c#o`2r!*#of~|SJbR$SJ z?E+HIsgXML^u2l%7Fu>adP$hcb^JiJT8iFi-Bw0_VZ183t_7+t4P9cjxp-`Cjz)| zIjOCIyA}k>hG?b9f_uLVa1W1IsELL-fgdlI$&sg8>}P~X?rmD+l_7S?ROhgp2{m`0 z;m-kDG>jard@(Xl`cyv@c0`vrLy_o;+?!@ky7{qoh*v#)s<}5-aG!cDWLe$6fwznr zW(`77nqj%qJ?RFn(|&*xK@}83?|WCnS0E^EA!{5)T5TgMpCb4$7v(o(g{XbhGyOTl zScr@^|D{ftdh*wLk-}OloWyAM*-n~o4RI`sF&DR&8H%->FnqW82F;TT(m9UnZs=VL zW1>7$dy>dJ(U~IbK4T_(MiX5I8lgTAk0f+VjMb`xJ)>viyWjc43i_g>(rh+1rtdIw{+Pv+A!M ztt&3dnj2rK%z14s4|x@0TAHmmmJ61bD{XM@sC@PLTsqYg(trvoUytF8r!o%XzB}<^ z6f*xJkIeo8gmf8&LNkrIC{9gb(n?}4q+w~`0a>a^wdqo?Sb4%&;-rtFuY&r~aZ=xg zR1Vp81zZ!?Hve<%zIr;vItaM(O4yR*o3*IoJI%0I7Z_s*VBuZHJ^T#bilaL4;fjcX zkz=MRZ@;g;(+^IuuQL%sIYVn>FXSF5^;)`9#ihShBZg%L13{C^Vf^U}}W+ecX({PEh|rY6XN!BETu6HiV=PzlFo1YT+kr><6XsLEv4s*w3-282Pd}aE+-_$V z=T1bc?Tld<0?kPZ=*T9`!GS`_Ve%m|&>p_OQTpk@WTNqY6j@C)xzKer>q+&yU}%bg|ob&hA!`kyO}w z9t-bo>_}aiuzaMHE0XWhGX227zc5^QV(& z09kk=E`2m}wvYBQZ!8|sT zEwah(8>4ZI=tbEk?LBIV{M(xr0o#}Y31D)5Ya5eJTAf%Grr71O3)Ej_G(|Dd5gMq< zrVUICnQ=>(KHOWIH zcJ?akhfi11vaNzUIL-*Gztv#zOa$)BYlY=ETWdg_#|jaj;bNl{ynud;^5`1lnrnl7 zolt}*ty}0to`bvGb(_ywD#e~iM$%WpZJ&$t+E}jToM#`{4ybRA)PW#EY;ZlU0HKH5 z6uI*W9WVRc<>sgJ3WyYjR@hhBbVSs^(}rnvm8+)?v#F_7u@IX%@5Tnjm(+OaZ8A}d zQ+!QqGHLpRH)(4p=P3luaBy{+Rv zbtWD(48h9ae03cvv;0EEQM@7s3}6jMff05e1&pYQ13H7!UntFWfMqTy4T(qeKDP&&&D_shg~!JSKUC^(M}j0;LY>W`@mi$jx8~2Iqb1nz$ynI~12f*c=2U^Rk(jLSFw9@-%An_bmF|OXs*uptxtC>gdk0CRfO-jN z2jkA^owHP(e;^e=p6i&;kftGGyQug&72c*}4=urls#e*V?5&RNdpn;d70W(!Uf6!j^vu#PbZ>+@r+KoO$z9CUiP3G(ZpE-={GRx1jF zVfYP87ZhFm%^hm7$xh0)2$e|oC_8&8Gm9!R^T*1K8n!-HuJ4~>*8)!UAq>PfM=mrJb-(%kcgyNn2nEu~UyUe7d; z7L=ifHz}TQf^tXaa0zK@36|ZgHFU^5b=AHz2(in zs3zKpZxedd_sYx?-J{_F!Jb-jrO)^XRc}Ez0ygYpKwKLnLa^gp;01yZT9l+dI*Amk%c0VkPOPMdYC&Eq*c+24$kZu%qTtl{k%0 z(KU@Iz+;g<@>FQN$5NWNcIZ+B5#ONi4L^GBH&AdJ-lsn*cz{7h9AurgaA z-*=Oaz0EuKpsueuL4V>t061g6|13&9Qv9x_+1Do|k?GD*s17`%Al9 zfTi}o0a`$a-vip`$p2!ue?&Mx2U?crJmIec+M4e`JN#4k!SnO{eV6~n5&JFB(gQSf zmhTzGKhDVyHxIw@ZTuHN%kt+Ff!_z(=U{)^^bbJ$Or*c8(zE_h#>Vmw`2FKDwqIwR z{|0DP*4fZow^dpaUEuulSC%*hRc2$!(EV? zfXFl>>PoDwu2udR;2fC7=mEVzU94h5u@&ahnTDxTA!rSRz`dOjy((SBp7%_FXlL~Tq4J{ zg!OP<>pW|s=JS?S`;M-Ixq6x9e!-=VnjCemb}_%Ss262zutdCZoJIIlx19&; z(?ju|wl*?of~A|MSthv{VT|LB;Feq9t0=qp5sq`|kv{nXJ1I|&L`o$GlW3Ku;c+`j zH1!_Dbiw3Yc_XIW4!)B88l;ez)3` zr2P3z+4gb~oDM{Nji`|M+#|+Dwjgr4%jO&tNm4U@>*d!u&8LkUw;e{ld>PU^ZdtaJ z4irqnUHwMG472-(vxnfaJN6584J(^325&rg9TPX?hV<*Uh36c9suDE>?D@VU<3HoSf^(_O%zZDImdw7@@Kl3Mh6@OT5M;SoAG<=thX-Q81rF7wR2X;Lq451jrK0Nct&Zj`_^)UlNbC+vFJC3L5 z@ko22-_-k3ENE2l9`tq)e;vlYQ=(gya_5EuuLNe)mauApB1hsP&2E9;TX86}9eecx zEa}s`fJihxI&IyqhB6ZsPIx-#2Ptav4?PV#wkuFch5~sKkC+x*vxa(aZ&VlI>M}n2 zAL6Nz8=p|r#yJp)^F2uB?(?JMi^7!DC@zI5#s%8g2p6hYH+{xLO~vI^&Shpy8S1)8 zpn2%w-+nEfQHI>Lr$Nk6sF`$o;+)+|%HC?N$io0)>9GofCveMntHG_>&+WRQ92FJi z{?HK(fc>6)3OHb!NC zGD34}t{hhI3Zv;uXE_obXO-4yX49rYbCjFm8(c=JNy9pjuF9O*=)nNW)p}Fr#f|1% zGo4SzQ<^AC`aUo8R^>6IqWNT1v0|QVHwAq7p|t6nBG0N`t_sK{zvP`YQ3qmc=(Cr8 zY#pU{AWjr7hhm7|F-|z@7;C=NFsn3X1b0UX{Aw440ppC)&^MW0k+_=-)Mqeiu4^03 z>e&a^IQ1fIDc8|&2!sUmYuGdcnwwP7VC%SQ0wq#~QR-nf@rOboISD;z-F>yGR=soZ zl3KTZI*5wIcf3m}m?M&(g?IF~PPGy~%BxL}xM=lbT6;Lf7Ua7ad;lnwka3Zt!$cj3 ztshCR!sY8>B}^PKPJ3N4PddIsQ4B8|(}Cm#@1U53fQ;PLmMc_iuKT=+<+g z+O5<&b>a4jJAHN`GN=`On=UgX$W4boO6q=b{lZAQUx;6HBAafLRffUuGjA?`pKB2J z3)Q#0uK1booG4go<|WX?o^wmM7GTaw>+5VW+>Y>Z zrKEzx=MiGA(HN}O_%YN|g}2983WGfQIIEpV99*E+vw{v*;*~fLvB9IZb*Gl^i z!wS8YpKUze)b<_>mss(EZaywBBVdT(-P#^*ccaNNI0v3Id_|h9h*V_$9L=znC_+?rzNyM~?@Lu(={4*1ibIYH>)ma_1Yz!eulrlDop?0_45B3{sLw~FtR>h8ULoK@$=HZ z0K#9dq5q14v|oDavVQk|{zv9pK#1SN-{;8xVz+-De_8*$1@ymiFwlK3NUQj%gW>sk z{=UmUjlY1#*NlLUAOBO&eO~(6oBLz@{eN7L#`@_@K*Y)B-m*;U>GkNSOkF_pw?lLDf&sXcg0t|gl5mqdH zCEf`b5un*GspL3-1=mIkGrND3!hUP+b+a@D$u=DU0qF~|b)(DXxHcHDqN>H<3d34T zt=9`W+k7>3e4OKj&~w?^yq0HH`8WPT1kKIDkV9lQ~% zbDKb_tD9hAn{cOGW*cPPO)nY|zCu$K0zPcJR}RtD)|$_g$3`qob`kB)g;r|)#Fg(j zP-tp;Ysn;za*g5>_F7qn%zDf=F;hrvpK}`+>pB6w?|1=M1F>hrh^rE1#OMBiwP70* zrn}B+Vbc!Sx8hrxm>Jf3rAkW)S{-TNG4@}l5W|XCqpJhgwwQS^*&bz~8xN#E`MuBg zn(M%5#8u3uqNX+TfJ=<{f@2y5Oaxaa8~F`lzCC393lz)LY&X z?+So!n@y=UXfT9+1LnsVxSf_j%$oU-WHIIpgE#HF50PYh{MR>$2t47l)@F_P-^p&=v6I{oKnw4<+>I-DAnS=-%WR?&h0<6~Q1(+U&H!p$ZtUfGXLa8&-N-!C~ zf)T>@9gd4XI%`oZj^EVnEOxQX&A4Qwew%(9ASDz~F{2fV7U4oNK2uS)%Id@AJTeoT z1L3W^j?N7guol^oDQ&=TST--8>^E{HaEZo_Oo{dcbtjN0Jnuk% z7;z~AD7ve+=@ePr7#HrumA?K21S>ScmcNw@`Fo>#F-jrJzJjiID-OHiip= z6cDsr(IN>aWh=Bcloxy~E4=u}VFOflm@B0_&}4e@QYmR!=`iK?x6n>?^9MBddel{h zHlJIAk3wYmUt=9=M&tN1cRxD6IYq)u&9_4AmO}^f-9Qw#T+#_L--|+BNw=#pnc;?A z@vTTiV1*q=%E9&bORpI(TBQb-Slv7IJQUdD%hV0( zb;-+mj*@-5S()*|!<_0Wp>hYghn|<(YhN8^W?;AW?DL`S_mQeEISgDoW`G6i-X?8B z`&4&@3G#Z&8N$`qda`oM(9!{G(%%o*?9H7=>EkexIqt}OC83R6d<%{!ZSeR=H8TVy zz5}s@R6WM3;06mlUgC);IG{}@fa?D=01O5y-GI}=GG}H*{!Y3kz?ya3%0}7zpp--^ zUua2n>|N*@SFAyunH1}eL)80EIq2qCJ#yVPb-sfnm1t@wLl?R%KC`C4TQq`?q;`qh zLNsEWQXdE$4OWUD_L+c4PhzsP5%Z-+~`WFLJ81HnxHVfiJUjqbQH_c-H&c-k; zowXQynEk~|Pap$L1&L-u-xYU|dYUxT(M&KOzmQK&OK!{dryhOj=Y52RzDP5dJXZR6 z)URPKecrTV(_$!b3iK{Yzd^1LA$8@s2CveVV)zL-v;Ln{t+BHKLC(>`Yw%?Tj_(Xr8eGfa8DrhMJ9uk%n38mzKw-Rt}1;Hir20-}@i` z`s0tI$^gm3kE$j6FD-8Y{kQ&e)%cePze~+Oq%6O;*FU1>pN?#1WP46V{vy%1qUQUI zI@>b@<)0j{=g0ZapqU*ox0sRbd(!e=xKiJ!e%G3&`)aiJ3Sp_p=AYU#&dvVexh8XfoOLo>*rfM$fv}~$uCgius zQtp{Ynt3v8Hvq`mi_7`+Z_9%@^8Si~!`cF4;k){@Ih$ggWuRw@@& zMtW1YZI*4XrVp(w_8yJ%3D3KQm7Y|lBI%4S-1JRHc(W3*M8ujyw}}jSHSRcg_2lR9 zN6oF-L61!S#Z3N}dTAE8FWHRmQ9u#kH{m@XJgWq~ z=tGluD8!YgWxlUzosmx2Oob((pfI+J5Y!@DmFYQ6#qLoH+;B~~Id zXa&xM$F9+nY0p?IgK_$HzX-xG#T^5RwdgFPsWeb7Xg zfQ$by22*i)u4RqGI%f}Yx9|HjX>ho#YC{_EQF)p)F7yup_dY&pg@~h-t~DA`?-Zd4 z?T@h+r?=Dv*70Ig4>I5%`)i@KwBgf)TKeAZ&%BbJhi(WdFON%J689Kh+?MQ^$jIw7 zL9$!;(B~eq{Un?gB1ng3gpNi68RiSI(+1gInrm%sF`r!=;(8)AO_|9P zZ$&ufQX%3-CJ0nyg*U%JQrprAvYP6IoYG31neP&=N$M=+G)JllSNmu^Dp_mLnDu8e zGQ#(CTWX?N!K#x=Ig5OI(xcFe-R-olE*4iW;B&I@a=l(=TRa+n7NF z^BrzicSH$IYTet)!P4or%#KFyuj_b&1sj_lywPuqHq*FO$Pho-M}c@_uqeZkgMxV!Jk#}2 zVmfkU{)`qumUr&mP>XOlJ&=RN(%I z@Fs=}dnsO>l&9U#3Tz#i-XO(fdZyzfaq!a!HZnrlOe^QbJ>uG5vVyhrj)PvSekwlr zCa@!Za-;~;T;WQ|Jh*7L6;;jB_lD>HBOrA=9CpXG$XJJ|x{ z!UFCYW$INwQX7FHaSl6#W|%npG6=ft(5mvjfW+eb4+?qWLe%*PAMRntjo31hVRrl_MV#^{f-hh4E}&;olIs+Lj1Kt;a+RYjj%>}!vkqTzB?f#P*F6qUZg0)J&E;WNJB z(}gGFUD@RG1<0%5!#M^9&%q3J!nStW%WuTd&5L}r$T!O+6j~hy34)QZIGVwn^44wR zo85(VjLr?li>*H4ILwRGiriyDtL_0{FAc*j+#`(nt*6W^*xfaQG7#Fl?rppI;~(74M8`H&Z_VhzYaYL9gz9d2IMJrc>6O9`Jt-&-_Y*bnxB8o8|beW*u6Y5 z)6UH#zi`C2=cFDBd2`{e-utqm#b0t&2=5`!9U$#$dgFTJ4+NNCnlf)PODU^i%cYyA zKnRDqwFk?^3B?7bmFC{8cCCVV_HGbKy@?Du+GKPdudLDB(QM(9uuhCv zR)@oy7LHy^x8O%0Wym3dYoi94qED7$N^Ti=hto5q!Uv|2;^GWH+R1#m+}*UPi*rz= z30E4al4t+KTt*|96J9oWh-A09d@WHe^cXV?!izdi{`5c1>9lUFG?_CoRX*JNY;f!g1=rA$?JO;pk>W=~2HQ;5OGa8lIrM88KRb4+@Y8TD$qK3{eQrI1{YU<% zHIo$|n>698?JuBghalIQ%CnO*61CaROr#ZLSluoJaphPXXgBhvVr^o2wUf{s ztN~Q}iP5dtUU530Qi)r=#jlhr8?#J78hq9u(06-g^SbR6G$VFaieGKw1gf}r8q6$g zUTt>C&OH@gJdhr~z=~qWIy>*n+LT+*p7)0rD49*1Lt}=OK zsl=Xfjg>-DmWL#k+aH14D-*y$5ISc%eJ7SWj_Ehm5FGiQv{)kRFU}gQD~-h(Iv#Ty zQ0hM9rvDEGy$BMdoHTr#=1 zxYNc)eGe-IsxW{|<;541GNMm0P6UTx6*vP`xw~3R{br<}a8*YTgRqw$U)YeFk!%-M z>LH&z1al?mTd>~@4P`lHEG!#j^|_p2e(gkl2jX~ih2M7<&6!|#A!TkY=cc1#f=33? zV_EJ?HtFC(4K;oui{`BRmNVO#_fW+klwIG-O2lE$C#Epp#Ump#Ebj&BO1!i0C`vbo z{(&l6b~S4Iq+}faRYCcW?`E6A0gO^@bSxRb}?t0_lVPtyDBzEmh_rD!Ou$wkVc4&ErCI#6L& zW){K^RDc|ZP<*J~2Fo|~7vi0cQ@?J&G5F{XC|8G#%|BBQw?upOwKB%|YN-C7+6GmZ9WTWM<^picSYT zsb&1@Es9T>ZNO|+#U^#at>BfK!B5(5*OETj818 z+ZQ67z!_3yd#vC=w4A<%)HJ;tA0VZE@4`>F)f3rIaFcutYP(v%`TDWlq_IkzH3PBn z(B-f$L$}lE__%!~N&ueI&=xPfi~GKyN`9kBi`~SW?o2XuK^_%4L|SME6KdJ8L&5VH zeTi6jDnemNA==Gvbco`qc~FYntO;RGH8tOUUdHi>0^;6|hALgv7;-Q2^O!`t@8Q7hbA!G)=RI}l;4;|LMNBZL%I@MVF#8Ap*U>b`6z`M3PiYO zC&DMFd#T0qJDcmDlBMz3Uv*!L^p7^f??QTL!N9#S@$MtiDEMD#t5?IyW=&3Vup zpb>bInFweWSHLaRG5nZ@Y)lUAu7-I9blaz%{i+O`e?)$gG;ack;KWNxq^xX{)Y`wt)jQErkt`BL0NH>*1Fh61c?K&7CYy?P}t`cTF1{fdD^BFA3-k#V;@{mx^30W z3KoyGa$vOb8OI<6-{<&ITsG~vhaY=a7v%~OOseM8M$XV7VmTk*a!nU5Z9QVBNUjK+^eh#kyUGvXk z*ne(ps9<9K)z)j~CBA5r2QZzX0IxKV~%p=7IlP_{;P>n*BHUTbpDKD0D*%Y}-~j7by%y zxv?5!6@YnZmr4XC$3~@wOx8vK4Y2H%1Vx#_z}Nf%|;EWPfgCrUL0}Kd5um zoqX56cA(9Z*IcjyH#!Z1Y*9JPhP+M+PGDUfI^>VmVAi3Qk^7;TK@M6Q`_b9yQC}V^ zQ0$F3PM_ZwY-3K98=N2{r>--Wb0Lx>f1n`=0%$h5s#VAZDgD;8?i!r270JoXs<#;7 zwhzFgO7Rjc3iWh@Ly9BuR6?h2lq+znx9!>a2IH%@2zyuGN`*o;IS6WBY`mRs4Ui*j zMRQ45K^LRU?Pjr!v)&*bX?ND?p{E`){8ltIk;O#yUbRllwoV6_Ud30Jii53WH?YPa zxz*S|(j1YI2(Cj0eHsd~O)iSlQ^4(0J8_)>>%m~Io`Msh%xlx@W z{ua8SZ8PdATtX?b3;0l+gi?7THe!5n3)KgTbN7sIxx?|u19}q2PN`0eHChjr@baTo z3_W>>@nd0!G}?Upp2kS28*w4A1I&J=6v`H(3mmg8lAw}vT7$}xbp)eAzWvA;V-a3T z$oxpmTdMXpaNpL2vFl*&KEMj}p!o`<xAS#7^QtV`v&iJuCfJ}KNwIzq4fIn z?O=P()bmdKf|uA*2g1<{?n&~jlg5l%%&8rU)Vg{IHj*|=4XaxwB`6D`xz=I$r5wvK z&R(}ioI3~4wB#v^g~-r1I4?eol8w{w9`!1+zmg_=_Lxr-7Jb6%GaffU#*3EO0M1I` zVwItHUv7X?9p5BLy$lLY_SlWeZ(ZBd1B(@V)&+0}-r6x?_m3lI>tSZciJz$HAgzxG zThASN?m5!uJfoKr`E7Vvhge^3dSXTQB|VwU;2^L~eWRUx+OnEYfWh3k6P=Q9IiW_oc!ew~8=PCl$?*-L$RPxGfr5e7OZylRl!A}Bc4#29+=6jHM zXg|?8#-KEBG}sc1)d1 z2Kn?-?b~{^q3avp0_ylmSz31&s!9Y^W5A5loY+-T#AfEORpyTBT(Nc2y6sL0GYR8qD@s2g{wQ|o3ZMM-U2^9*;6O&}tS{CsUgh?yG zrXdMdo`c>eIthe_GQDXKBdkM5A~W+(Cp+vV2)o<(Cr;K*#4BS{*$`>Purl-5v3cvK*+rx$NVK@MWXj}Dlcj`wDLPP`%m)=bgRnXZDK25M5vt+XN6IZTKjXk- zdO=#tZ=OD$3QlGca+l_kq06kn+p5@=eVY7CeO#yc1foOzc>T>fTeCZtfOC?U`12*+ z`dl4O+r8_mvruWpVrwwPhF?x)(cSlRQp5{-_tBJd>Vij?LTd{a%M0v z41DS1)wC&2#df!}k$1Lajk67f=$&|kY`&3Kd(dgA=6hF1W(0-js))LXw9DnDw2D02 z`7*7h)0}%U!NkWw0g^_1Tu6Z6KywCZD%}_Y+c;nLaTnZ-8Y9yE!TYV-@U(co-IM(U zjd8g7VxHO*xyu8T3a5%eJlvreJ4a%nm_iU9IFddZompUV9<(F~ngyOFojfc4{9Kx1 z;y2NfYl0BDEz{dCaVAZ?Fo*ZxClkx#Cf4FrkFu3TlLJ;^QG4U><)|GFr*qt&m+s+M zfsWSy3}pa@U-icS8&LM&Dp&(;F#teG^FI`Q$x{?}rSTD3)m z4R&PDFRGXN1+T+4n_cIzb%Tex;NuL;&ET$qTISakd4%(N=QoRX?{vvVLAk1h1t`Nx zC2#iZWiH0teb!!*lERgwuJGH)VnnGjiYnRZhaNE}e_~i0zP6+7dp09sviUF4a0_Buu?woNEo&p_9WX<#Q2uDGJbu$D^xB!Hf33J z$-Rwz>5}KO&1+n_6C;Qr^jrIJS+- z(@DZx#4&j`GatOa2{tr|Q z=&se@Qo1(cm-JrnsueTO6FR}A&=Nj^^p0F3cOAC-cgcdG2HSRVhX{&v_2l5~?Sn#G z?C z{#yAOSmF1Y`pEr@}g;6 z2#o2G)re(u@(oUs)DtMWzNJTUAAFe!6MVKja0|oj!Wf$EBZ42G7qAtcL!0Q*)xEJn zEI+S=svIvg$dR(Qg^iBiMJ4W$_=p@X%tI^8dkKn>ul)70I&xBDUrPp`(5x4I6)yv*^V zdtt&lq)cOBqTLL1k0{Za$+3v&#oH0{w(Uq&2hbDwq`VSb6Y?Nx4FEfT7km6$UII6A z>|I$g5!BcLeEMW#5?~0+OxHk{36|nRtaYv~G%Z~p6 zK7T0;7el0e3mmS^P8iIo8w9H<8-uF59~ZBwP zTI6ALK`snKAGsZhc()O{XqHu!5!-=i^S}ebBqYzji?qDLXk0Oa2trr>JiEgW9f|@I zXJ_uT)UmJJb4O}x9Vh7cd>_7hdmvYXJSuM_Ld_;&kVWDdS&WT?j>uZo>Pxf&`kncL z?J}Q31N)ZXG8}t39%9s;!nO-d91&U1FbDJ|c5eDOY`^{{lP&e|LCv{&4cR^ETmPBZ zN7N6dBcel`mCXz}Cu6VZ5kuRLB2Z-YJ{`EkD3B8mOxXpEcSkD@vcdayfk9-N89+qZ zDZyx1URL;i4xYL5gxr+i*q`ZlkjCV z8NG0fi#rMEj=OTvla*KdY^m6U2HLs$2|#wu_*Gnjm|xY$ZW4Y)ipY5r(RNKnKn~Sr zn`hHETd0cxX7UI>KI_||*iGiOj{Deg)cOmu0)|&Q&B&ivp+i zb`w-x@wI^G!m2E`g^NVM9Pg3~K!~#zE9<1rO7aH!B#6h>aAi`y6|IVIUYzeuA_R(G zd9k!Bnt?4WPR=a>*EqU2EqZ<)?=I6d|QPjtph@dxO%~6nB z5+4@|Jc_QK%;BfMg*aW}^M)&o%6S6p6d?3T$E=px7xp4BA*o>kYdPY~c#f1KrJEhI zAH6A}FkjbC=6{X*7`eN6F0!GAO%6t|xa@!o?h@TO$@ix1@)^WImm=;*wENex+5fbA zj{`XR0Pu77-LFN+KZkxqGJg*30?ls$jK8!P{wK5x%TMh(9DW z_`lFD<6k@XY$tyt&j5V?oWCsckJ2vCof*LR%U$Ca+Wj8-(Z2nuvuKpBdWMgIg+jjiQ#^d*n3*h^-^}nldm#S*nFS4V0%$6sMHcEhloT5D# z^*7q%n$Lzh9vH+GqYDYVeitjpuMky#-M``!dl)HPT5A89E+&8wbT=a-@>*_uAmy$LtIZDi9#+!2;6j_gDM&6E5g@KHe8vB00vwK9XIg-@$??6J{Bb~ z;M@$l>$V;QIkq(eCf#_UWqoHfauTtP02r~@kW-(aN*{5K(;fU98$_vXf7xz((`opx zpSQm#s1V;qPunC6X(kf0vn5#Asdt82DDn7h2zn+>OR2jkCrhZp8HzG=9*lRu1tEOx zXiS`b2Trhy7Gb(gs1|r;_;{%u-F$`Sm*ibGwru+ zH7i$PSk0$fHP`A=`htbfb5spF-#C1r8YC0R#y0|k3JEJXs9OoRmCazv+v%nAibRO* zzQyLz)f5rrR0Ty%Fht-vWCopVsVeWB;>#yRm7r+uniake6&LLZw4iN6)83lEz`o;2 zpyu&s^o&<@pC;YV(?s-^X5@oAd-qv1>L;f#+>Q5+pid??6YV2?dp07*cKO0UFBHmB zFYLBLP5BChFZPpfomuR$iq$b&lVu`kZ3XLa>K3sixIaICdjYE^k$xeJ9QK}ph<%X~ z^0?dB$N`i4Rx$|VkjsS!9~mVO_aKVE0PrYrIg9#5ekAn5To6LQ+J?{7C^;~T5CrLg zmBg=&!CjFKPWc_OW5^6ctH0!Cr^DH0QnR=F1lR_KF!}~ue2Cu)XruEAlAz}HzR`kS zq4&}y5oZsnO*BYZ6P(-})YyRxzYLe0)oTk{bTum*yUMRK$JK)`2dIW3dxtD&m!Mkl z5i*9e32gD$<9bowcz~-$zAc1(8nq%(pf$2|+K7;`R@_ezK*x|ZRHBh42v~P=)uakH zF1%L)>vy!Qss%e#M#hNE>W`M|?1Z}&9f+e4e--@V;5x*q{)<8^^o#^7mfkHj5k56c zG4Ed5VIq^0j~D1HBRIEVr{Z&jC!-T{A8#q(U&v%(l^EhQn z{*K?8RgBqsmnZW^-}s+uvnCuMYHz8`Ms`+I+Z3%zybI8E(X6M#pelv+pfQ`ZurPLd zH6!i)jU!}kX#e9}=m$~0UMH~BSOp`C51AN>MO~-)q z;-4J50yPA93u{MhND(-bz*8EkLHGy$v*tr2C7}}$zeu? zc{a&;gKnjE>shH~e1)gjxJrxlL2Mb;_+r9aVKjWbU@mVSQXwXa1Xx zZr-_}S!f4bx29`k86=L&2s-Y)mx#higLtuDv9V>}qYq(A`A1JooP(G|kASDi7_#rY z<|aXBcrE@uw-rB~mwqsct@f-ppEXD$^5tP|ESt@AwtJbw@{TrF&_UIqr=M`UV+JDm z?YT7Slg743$pdYDm%f&B17iwUA;PZ8%V9^i$`B{_I``1+^DkV*W?P&rRUK&cirJAm zs%lyVUC#Sg)+qk;S74BQ&{xZ=-hyM-G(W&k#$UCa{~LVp-zv5;0Sjr2-`}MF4nMz# zeuNf(4t@fiXaS7B=u`g_`~+tBP5k^7`2W{>-~d3cu|JAI{}=pZ{p-{=+xi~~EPno+ zzx<&8#_{^QVk>Z*FM#P6`2U5szK4EvMt&bZe@`6@VEW_G{@d{QJ=xD6_$zq)PJcf| z8Ch8v{T6R|w_j&R_5!LSCNT=bA`S1->-nqZ)8k+c<VbV%-J20vYZk!$J+iTkdcEYAvW>I0mKB&tmFa!o5!s!~k1p!Ci zVKhJi8PS`&pSTBIv0)!k$XxySG%lGlCOgqHdW&Wd?(;O6QyMmO#%L2A=1UdUls?(-jeW;25Z{8yG!tLX;=lnA-5~I zoRWqjldQj4@G@5LEO#*l>munh{A`GTeF&n5csjpC0YN^cR{woIPuW4r-PB&Z#H~a$rwc8S z{1QxDfPvoe;t6*ozV$lZltk4Wl0V?1LF8NYikP?3%5dfH;L6rgzW~^eug#F68y>9*?b&s$HJ}n3>;`-h z_RU>;nFN1do61nQG?sU@4&&bq)ATyyn|x)gc>NYj>|%xMmCacoR<(HEs@h7Uq>dfC zSkBhy>^`E5C%ZWnx(`3=QTK{O4v2T#WCNjgV?xEN5j^kvK>n*RA|vRjstJlo%i%M> z;hasdMr6U!K*=AKgpGBusnUdKG>(WDNOo0>4;MkZ~ zr)4m(*nT?rOOhy`^#lKleTY!ZCS0I|s6TT~D2|gMo8&adMR9aq>P2eos1VC;Y672P z?1R9262R+b8=~yBeVa%)2Q_cW;0SdTHE~I4*HcG;cd{4|hzLD4S2s{L+ft z^@rI9a-wM;c{e>E)$TOtAHku{^V+;1P)5V!+XtPE3TX5<#= z-^S6Ly)^807Ho}2JZE`y&>=9yVG~OCu}Qt@P6?>3HB2-LGefItL^`OkJ|X}Ndsh;l zWlRaL#qQg4L9pMN6mAX3xDrbFd3@qV-tPHA8F_Ja4I^!89;uO2SXNOt!1ogQR2{nw z^iFdyrXX#?NEfiqMME&+{&`taBM)o%=|;_wNj21~Rx6keTj`z%F>y>)ta)6+(Kv0d zz2|Xy7|(-$h(zWIJ|}n-uP;TmY8I7klCNh;pzl~Jn551OK9=IciPCxCRAMCMAgjT# z0bXUE-va`bQxqXCAu1ekUC|zjpE>IqWN$DJ1FQR$3jv}WJ@%qfIeRKB17 ziZ32CwQ7e0HQh6k8n4`YKL5Z?m?m178JK5I28uol{9{mfaVDCzI0SSP!F+EPDc(=$#)ALo!TVIg% zHpb00p3d+vFe`K8ZNmYac&+uKt5LNR+;mYT z(aRg?XMTOgW9|950M6bV)U73*`EJR%{_PZd=VWgb%dF=ntrv>Cfqz1L7|LI3?pBB?#X`KO&w#hl+uS31Ip~`1+qv zF)+h#QStZ0|6;AbUuFA#s{mSJ(Erl^1^nFa5dXg_%7UM3oi~E|BEPUrLze#0>H6#Y zI2+sFK=4<$fj>u4NnZrzJul$8}_Z1)z z9D@ZSed?GnTfPZ{RLX%TPGlwpT&p|K0-=q&(0D;<`Mtj{CVZC0&|6KoGS#}61fJWI z%PGr&V^)$?c|Hk687!%b&eLcj;`k@q7f_S3{u_Hfw=}Sdk0InAL5yASGyBDdT~1z6 z5_#cPzL4$9=`p61+;FfLtK%^k2ZXC`*lORuF7~k8-jHI^TJ>Tr_ENZOS4+(YKzrUK z3AcnV242J-LSg}=T`pZ#h{(O1={&^iIv+pP9SNr9RWSuCmhMB2Dj&rL-Fee)fqEro zn?}|U?&Hc(F(%oNSs9puphZO+(eb2(D&WIwrfbRQ3r-T+aXhiNpf*nAWt6kVk!Q!) zTTaa$4=fkWrj(Fs4oZD7pC+0MSCWEWTXUPfkuWJ{9Cpsmh+c*FHxoS-F{h^v@Nbc0 zX~V#BMQATD1K(*vRwg|c+=LoHf)IVDEtVr<^KRKy0tD0Ts|45mqClNS29-Q!sSlE1 z^y+6u#$bDbvc;mzt@C9X1 zoKG|}#&&Mrz*q3dE{4s_4`f}3Yw0s>DA%HIMFQa~NGVuT;3eER34GfKa(oAJ$J%2L zWw+@rk=w~SX>@me-kuBc30 zlNuc%nDN?QHJT_%Qoe58FxHOB3WdEf9`Xz|U!DSY`(R&Ic~Jv=UvO8VQ*T}=+pBJl zo^vF?ejc5(zPTcdr;8`igHI)b#7Z?$x7}zMsiOF`1y|+g+Wq%kjq%`v4P^|x=`Ted zbYJEJpM4LhA3h=#hsUwWeqQ`+ME+V2t2}OZJ7F%8Y4$Sj;)T_jXa~ZHN#;A&_I79e z6{*SOT%rKRC@HJV49r`|6PEZR}C`$15b45xzzkmn6gCuZDg2kGH?Ni1I!%-`aaFid*CS9_ zN_w}Ks0+J$TjSiMHv8q$Yg4*-@6F}SPc(JON3V=N$4S8Gw^4E}r@Gu*h)Q&($ z_9e;(;k=$tO_wu_sHiAz{nMCoB+%?;jCgE6&_N|Z6R95H`{5Zx+aWAXdfU=Jw!^W6Blm?^paJmNU2dieq z%x~Y+`(s^_&>xX!#zPJ}IWZJ6FFvX!uk+5}6>4TpSy-7F&raG~+g;g8{69KGfTui?&rN(?jd?=fPE$8c!n!LAt)!MyE{3&fGtkXi=qKsFKHb+FrbBwYD#B zdq0x7@X8u$O>fbJ239GK)~PIjF0v>3X2x>%l*n8k!K-!Rw6|cPm)p^o*Cz9Yp ziP4tRU!BfAoJ5Bk9tJIH_O&t5S`M7~PBuc*c`{AgnxAG+q2xO;h z@!*?Zt^^Iv@7`s;9;8Vx)-dR_18;}kO7#UBAK>^L%xFSW0WmN?EOr@p4iWX3MvC8X zz!vE}e3kZqaw54k3X7z9eTLr(f9&MVbqg2W#{mBWT4DZGiR#}VlmAwQ7^q$ZF#ogE z@_XoC@Zmp}`u;zZk@2l*ZF{o8fnS2bqfr0su8 zc+2$jq~~7?Z%2|Afi7vNK_^$}o~Ejy=pY$4qcK`_>)m8APE7MI=CDLWsH`v)FA5{J z_v!`I>nz5hB5&ReDJP1Zy}0~#wX)0y#iw@}%7QS9$^@1p9DA?J1Vg+npa+TX^4&)m zSX1V`1u`(-nFG*AxHIlrlpL`Dr#Gf`oq7NS5ioa-+m3h7Pbd*DA=5xCkW?l@Akf)xfcq7Ten??!=O+IN*zWIcxYYC{v zmnG{`k~pvG)J%G-s^W?%IT!|u5>OzgaLtq=aF9{kO9M*~9#N%XuGU@{qAN^MxSBT{ zXFHPPi9c1u$%tLPC*3y=)TTESHKIqHanrIbn@7M+Dy4B@37DJ#SQP3;P;Utq6X6bO zUg(?-1|p?yz0_$B1gUDRAT7vZfZp_CcAmnjNXvDTSQg4l0G0h{`yNkwKyNLq>3}2DwZ`dGlXf`blEw|k0F3hkjEEYpDb8v9@rWj3B&}k<#q1J_V@3yWoYu_Ty zOiV@k(H*^!7zgniBe&L%blEE?4BUb@H+?@A@Alw>k?VUf>2vLiap}8FP;TDS+h^C* zHKr12W0t@d@R`Gs>p8zvJB&0VGc>iE+!AbIEhH$wmHt?wPjSG=GV)7ro$$w%!`xd$ zjeb~FJJMTZbN`<8iCReVyiWg|bXCwoQ`N($Dt%4uP74BgyStLTCI7QI$Cc_^APeH)0gIWTHUPJ{~#1|UtvFCWg(dNOeC07U419nShCoSS#g zU2WZGXl$pTOY3p`yyuMc_<6SLw^6bRP}0fwGUreTH*0Xk^Z8!Kr!TM196JFb^ST{( zaSKKQL&xg~CH~ayZ1rU=mX72v*;s>hjzrrCh9zwES2ZaEHqpaX{NDrEG*h#-s)I=1 z*;D0i)@x={D)SwZexmgf+c{rA3dP{S>1ueF>d4gH6S!APfSi)>NMA!fj`h`8Is_x` zI%VNlsOwe#(dt>k0D3xea`gR$1d&ll`V9cW|Yz~2L^Z5;&{-&E6POA%j1_d1( zEb+qRl87fa9^Y|vMTtO;()2iqK|TjFH0lsjiO+nIq*)2a>o+T%Z+eP)T2$a>;)@gl zBkkJt)-qrV-GCD2d$a*gn+zuV-t(8$iG|zg^eE+$ZB1KZA!U)m2HBN)&F_S}!=pI* zijH$>-s9F9q`cyNeVlP^8|s|x$;fU_?T?HJoiQnpVIP2~a}~M>{@|1NH4TKIQN<(; zZ|`8yf_&6U=jdxbYkp+MTo-1WoZIY7eR1KNCufGRUWRi)F13mGP2$#)z&f%Q6H_X^ ziBOyg0TU&{(q@VTHlO5n4io2K^NZ5MGg9m1$~VeEuL*MzoJN`4PyH~>h&1NSZflEu zMH@}t=UynJgsIarkv@QYnlVsy@tB#hn3fQN`m&Rjfosf3BVjz3d~xWi#kSpjdGhYg zFP`0>)LTaa^()&7NiXk0V%>)&@At8Le7#j@aZe1ysIDeVdQ!2Z^P-Y|(sesmD)Q-L z+K#V=r7w{+QW77%P=}~++i9LG6f2%NGJGqdmZgmqOu0Pg2tV# zakyqkL=#*lB=p5RVoZX%_2Jn*2T$9?(m2w^46gCUexq!gTD9Fq>&9l`T+#yYPMNiT zELivMYQ*qP@G!NqrASpu-kPjqtjC3^`Vf)&g+-?`6wa8vfx0pDa;MW9L=EysUcmss zpVE}Z4s@IV0HwPBNI2g^KO&qzr6SA@v{3*6ei4fP2Ne8MhTo##?}`7#T7Qp%0l;a@ z!2SD&iZD&!p5FMdJMh z4+Ie{%*IHp&d6BrAYLq>^p(cv0anF%op#zSFfsy$wL=U{b|;$pCW7`SCocyt{%>g3 z*m~$v=DS}e;{|zudI#|HrZ+zP@xt$e&?9{MM|VbzZ+$<-0nR8hYA{<#47}2Nu5onu z9VzDc;s`W*G6=8X{KbP&`RoXm1=4Q?u*m`bB%vO$My6O@x%-y{Jo{_Ty61C^2TX$c z&=dJSig|CfyX5BaPDRWDGe>LO&_xOgUpzre6Y09x+JazGB}qUNDwagMWqPPx={TWYgFIj8WHPBcn{0Il(ktmW=vay?(;mN(15bG#DNh{#OrckU>m8OoL1$chh#|7E@$VuukPbzLgQuMa?a_C# zQ54f`()2UnzJ$cGy@Pz2Sx3sofYvD2)sWp0x9K7{oQ1wbQQ{MRw56X2LmohqC4H-P z6NltK6R&WFpZ;L8)=eN%V>aWDLr~_RHGuDNwH8iCU|za3kMi}MkXBMx$?T{0GFgAX zK;i*c2rd?YKh{gwq@`fHDbW%Fi`!0KU53WwhZ1|fLP#fJual5|4NsGzZl zin4B7_5i)@exlGbcUz1HfetZ#=c4NTPL%<;cYGubP7NQ1EmaJj2vti0)H?+2;Hhne zo){#qh0!1$)LrFWZz&m`1SD6?@N|g?$&YQg<~k zeUHek(-gOLR6l&VO02V*AW0lhw4>Q7J!}F8I;?XEGM%4V zKM>^{CO>)@XBW6T#TBqp3V3}krh9ja<;sT(l1Qv!CM<=<*O?8ucVJU775xgX>QyF8zWK)NTEu39Z3>ku?D-x#CQR`7-ThoN?@%^1H&3n}y1X4=c6Dy4%)P ziBw)@`GGZ`G&~;M=`YhNd`O0*GcJN=vwf@maCm#K43@<#Tvcw#gNm^=AT`^IWaJY` z)d=gCL`!)zVJWdNQKxru=dbvZW%+uY$dJ9G>w|$n%FpBNcm2uql z`p4&L8lj7V4}{GHCgTrThAGN503iip)H^+kv!|oy871n$cPNsk!%@D>Ql|{|vR{3f z7oc`TQ!u~gGBl0A^&pK(XwTw6RPmf7e#sq0$i!rM%&Bv;LEg>NEe#9R@-n5IiMQ4W z#be~6&CKoE@q8$wE;i98g^iNPx6{5F109LDg-PO-y3L;It74;NS74&hXz|K}D=Bv> zSi=i;q>ETCo@kvYt*qR@|6)j&hN)&Dd@lt-il65V27ieq5Nkvd^uCvZcpRT)Q5G`4 zP)D&VcB8IqMODEwD(no8?p9T64Y2l?N67CF ziFNDqkwk0^bRD>aG8MU$Ew$$N!;&u*?-=w#fj;SjJBZ1vyn*cqJ;&Ixfnmq=DA9;yx&BKC(cDD*Lu`ZmV6R* zE%JIpV`U?;y|UV3qD=Q0U$f4!I}?zuM|lB@OYe61{Owd} zsEZC22WQPW>WNQv9qGw4DVbCV0n{$j zD>;v8Rq$3`w@>VD!418tLB%>TU8A2t>7T;!uihZwm)>tWu`_Z|yw0h!$R&S?@#UGS zE~R{lg13)VX~X9oIckOJmxKhA{y9wCjzBczBpcBY?R+oG(BcyB<3~&`YpwGi(aT@o zv;U_pZ0x{c2mrwMqT1hSh`)z^1U`Qb8v{LV0f1i~=l>Hn24?t8HvUz?@9&=Yf3?=% zW1t^bh<~mQ{;$}$;+Nrbwx>TdZ20~;{~;W+1MBlF|9pS?9{SN4`BUMT<@*)t5AFMJ zSBSqS`zJX5&b&WGfwSlSrb1@IN!%6E2LsLf+ zhDhg9mq|k3tGPx$6t{{R9whFND(H##{wNi@TGw=ITnNAc8|(;81i|9F)$F0mcx)ft zaF=wdJ{% zPSmoVw_yYyF>Od3N$og3bymKWAPtByb3M&GP`t_gc1z-M@PIEHq|JSa6eN)8&E<1l zcjov--wWL+1E~fXpM;k-oYWTkL3~F6jL*d^!oj@AkOh0#KFRQWEN4YJLr7lbWwHw@sd-9N*|7qS>U9Pj#W(&W3 z7s|M`InnfswmWTbV^m%kj$Ci3*5%{di{&%Ih`1xob}DC=F^E7xha#K}-q1mPty1j! z5#eYr50t);D$Y#;(3nZE_=||%O9H%AHUY7SNSYllT=+%d+@9HO9+O@aWBJ4$`Iav# zbGhE^A>Ab5$B8nMqrph4i+ym;i)7$ zF(u@9WqYJ>Xmq<)2TX>I1;=?5)ls@IBTktc9SRQTPB~#PJG7E$x6@i zR-TXNm>ceOQrNSjol%RWJ;7NB9e%^hm) zWG-EhUdnubxY;#jLnlnSS<29ycN~)hIp4Zo5?Hzw>(qpCx)|nAUE)xfgmb&ts?`2@ z{k>va7gC6fSh&PO7wR1^Txhi?o=et!@1U3hsjk&ab&U4#UIsXIWQT+WFD&Ou7qS_k zkB9{ff^g3ig>oxMp_p|MEe%`_^#C57S+V$jA=%ah?7L{Ig2g2(MlY>|C^%wD>7JgH z_srt2FJ2BNZ;GKOILoAy0o0{QExpKL(h(BcK{U~;;)@fC#r@zV%F9b8sVg3Lr-r9B;RDZ z(|KEy5}m3Gbz$W`kdSDRq$9)RX)7y%zEGCgqu5*eIt0Os2`@3BfTNB%UB}=!I6H%P zhoqb8cB*(OYHV+Q}oA)_&VVYsaQJg znmSWIBbmT{-D&Mfcp1ZdN5#?rYB!Ok&d5n>H;~c#Z5YXtrlScCY?33v2Qcc1T2p4~ zF$@LqTPmgM+}3!jXL^$~iLS`u1O~8%{5Q-3#@Y5ICDk^g4ESuxIF|J?`|3XUBqu%U zpQ$0)-ieeQu(Tdz;>#dGn!{r|0ZZ&pH`H*(4WNY6R7d!_pK-g3LA^Ai zlyS@L6&)eJ!VmG`0k4kx$moVLP2W&hxYm|h4ozS$i|JuF6AdLmAn!=CoOP_p?P1kO z2cy(pn&nh&C))w?u4?>Z4MqL(!m!#ljD;3Dp~(kM$+^IcV{Up#C$-5g1ou^_#G|9h@y^Xz!iyz3|PHCw`eb84C94+U-RMS zeaD86>oO_rd6O8{H+dY70!hUk;?V%h?&;K2`_dpJn)fSTJ5P}g&d&=dEj@Q(cQbgU z9DJ1nIXsLutuQw|pcpdbgGo)9YK++H;Bw7`yLZIeW>{9LvaJ=j-72Wh$h(x%oCKHW zvQYy~4Fkjaw8kls6cy%c9kz_C$;L*{617A3{RP(yr_P>r0{q^ZV32u<;yhN#xugu) zqn6#dpNhS~;N-q8rFUG%bKov`^Sr8Ugj2<r!XmCWJTdK4O~!H|j(aaYaz)JX2t z7{pD{mHZR|O8SjAeCH4b6>Zx<~sxC;6cA{JJZzH7|eUiOwXc^&2RTU0}x# zfF5WO_RBl+zfq3-FG)VwS%H;ImS2R$KYiCvr1Epx?32mgnrKnAv7ngXYO=t;ryH}L}fzj3_&uG$TpatHvr zu>IqM$oJ5X_U-q9{vVIcFOqTKYXN}uzo}IIp6urj{8g>`J5T=THpx{mTN`)<|u?1#m%Eq9fSrH1tzM4XuN_gng zfpKkziv?l82BoU+RP5Wd=NWDH`gZ3wMeBi@?V{|G{IA?F`bE`YZo^sX%eDhq3<#7k znhNRpACNvEUE+1yqAt=wjtEmDa&v_d+xOEd<0TDerXHpDz_V`$P0IH>9#-&JB4y8A z$5A*bh7qsG{N$mAo$%r7)H%)4n;g@ffg1}>d{1Wo(T&xloVB~vDsMsiUh>zI*{LYy zb}ELN)0g#Q9E_?oO1sZ`F}^P0GhA35B<9p)jbm*H{9 zAPWIfR~@a{zuHiy(zWWNLTPah*`CIzYTdtKNZD%1R-Dh%(Dp=2w>sZ;VJUp@P?$x^ zcH!7dp&@f%RwcW4mWs^FmJE{3&uilIdNr9ueDAi&VP(LP%=GLgbws<96Q%F2_ns49 zPg;xbx}wVomox^`TFa3>>d>I}kT=6Ij(th(O7pNNL&)9T9^@vI~Q z(#5(C(Sgf*8RYw8GIej9dEd_KAm(L$9=4X%4(ehw>@D0O!ID%tZ(e$#zF~G#aZEIA zQ^wX$7q#HST5m`t3-4_E&!d%ED>I@aKCr8;-;(4@eugX?D3wKMLs(sOM-*6Ia2VQ8 zEsj;@nkVyWNHc`C-uRE8)d185*v@@g`22pGEva8=8=~pnUwCS1P#I zx!rubg<-hiZl3MCyTp_HM4VnkhwG%B65>0EDf!j8wxouy`RR|k)cY3~YcNcXebi-Y z0_x2ow5sW)7jgLm5tb=!aeJKBP}%5jVbmwBfqqpSXeC8L%Xvh_!-Uie)!$xP_;J1b zu;B2yE!0Bi9AAq~y;JY?la!R%Ek8o$;Q6%dk}kFyosk>DJYfZJg!Ec;4BEWjI!78z zt-rvp(SzN0Z2=$;xIy#vxHQMhk8e?gOG@{hJO_B9Acha|| zc2FOM$u4;YO^YES1GN|+RGOY4?C~okPW^ZRC8>^sxs_^$WtH14D7V}0p&780R6=U6 z8%myG#QV=%zGdd7 z=PP&>3=g+)?wRM>_!tkDj&X`5-i{d_9os=1jK3_ItX1{CYB8zFbvY9{)46Wvb1jEI(NL`A+u7}Q>#UMq0uRIL(+ED zrYz5LnFyZifkxT4rpV-znv6reZ*a0m*^x2_JB<&PZLswtivv6q);s`PTU5|*ke&|b z8oN47ksmOwaA<|qSzn5V8tJz!tX+~zXC0t9z$nVYfVG-~SLg=LpE!h;Fl(b<<`-6N z^(G1HErUg%*Ljm_*Vdo5*N!Wl!jch5Cf7<^=gqKm1>9t&X+P7=TAcgTr;UC^2h(Ig zVwv8wA~I{WCU?HPh0_VDLia^W9sO!9T(N*gu=~Q+s@PRBBtUw9LLVm-yfoKy5xkJ9}WzDHin_*431FS957 znnHt;qBrF1=^MEAOEd-eWdRM_s~In3=HUuuG@KXOCTqJB`OK%P3Zoi5)-Q11=^`Oj zNrI(8R-sz73NP+G@@*XD!|byJEqx1cZjLd3=pws0e-Hf#bN(C@{>cs(I4b2YF!rCIFfhY!q44*_|6;9w9EDl` zr|M@0W|hAPKWrO*=#TOJbN;f(KZ?S@qxt`cdkd(#vSnQu2_7W4ySuvtPY4>^-Q696 z1$TFMcXxMpcXzk&lXRc%+;h|Cy*}r^?+(V;ENX3P&pB67)c4JrRgA9d!}{vrC+ zy8Y`=_<#7=AM@XbAb-tm|4{bZ6n_q;|FIcD$I8U~2YO`rDRJvxnG55OZt(ww+s;-n zmfjdZejX{E!zmwN+|RmXw>^{*^b3&3=9k<}I-in3P}+M3pA*SCdw*?j|6YjbtrI2H zaOOw*8ztxBX%}-B-@E;m#H)=u z%o{AD#dv6IxXcX#13>2%C-z^~M2F*6mK z)S-66?j(ze$en!4?l8N~6N;YbxHssSH>_gLCBf>tW{7VUMrJd~M~Q4jJOQ-P+XOd= zd1g5_5>r@_r+gDjyu?fu*-$Gixg6939jO2VB??uRGj&5*M0>8XTHUlzuo%M+6!?$= zoD!T`9HJ)gs)ktj6e{;gex(f6l4=lUX_hHbFG9y?repW>pUXru#0_xvV|zPs zih7;A&7UN)=(u!_$L0q!r1QauU!n14|9Wu4G;B#A24o-oXf(GI*Ibq7;iXUQAJBXeOAhUA`9o75%K2hae6k_}2JYAQI z;zryjse{T9sm`)QB!-eo!+6ZjIGWgsf>D^v5?#+r&JBhgJ~Bl*5vWhjB|2&b(H_!j z1#kOr6s>dy-i~zBJ#oc`5=pE#aeh@Ew9#NK)r*FzpBUsBpUBMdWpfCRX0=Kmoa;e} zg20BIAXL_=Y<^+@?yQ0~a{J;=O+aj2UBj9pNdBN`zKJ8g`Vgb@foqZk(cN^^1@;=V z=YvLrTyPMp?a*N$4jf7V3XtX~#VIO9y%KmPjb2k2>;}<`8P(vipFjzmM6etkiiMgc z!I%<*j5S_h?FVR&Fx*9KCMYTCku#T!JNj7}R=hF1zS{eUTn8tq;Me8HQ`HS@&U>^g zbGt)65E~5=2$72$x+RBU5FJb#0fV!r)Z*}nG57t}A*rO~@3%XF|p&6#_8Y5({d?xsUmm4RxYIW~2FHVOW*tbQT zB^9CMO7M(PVeoSY#1iwA5YXmbXbIp&#n^9@2e3}zmXpTr+C}Yl?!Wxh-S94`le<8c6+2bJds2>a2V`Jjuw+Z`V}x`d&*4L*oWE2PV5lcPrZR{ zxkI{dy$0nGVlSvYtAcoD8WO%^*kWa-eM6d_+WeIZc}`bk zIUo8C%tF~ThC_uyY}%=V1jA}?wc6#90XRFR+M{GXlR8~vt###=RQVwWx>E)2YJg); zha55Ew`+n$)vlgNZSpSDO4s=_y2v?7OL?dx^1dy;4(t|>H9iyzL@DaRP2h#(Cno~Y zZjd2yNM(HFa@Q&*c8roc52wj+(l$jn;Fk$5adcTq=Dx`u#{()W_1215u^V+C#~YGfx#XO1}j^k(m$zOc=eGg?5Phg2>$ zU32R?S6!PX*Y%Uff1tt(F>M~q&pSlu)wUJ)5zSi$Z+&k%$#@G_A_?R`?~4!2&HFe3 zU*W@p{4qwn-f3#TRy5$VzrMhHYvh7eQ*R*9M)Zk-Xc>wDKM?-X@ld5;I^NDcU!~~e z#u>996d#4jEv@q<h5B)b(yHU_hy zkZT$KrO7*CtqyPO*Mlbb7bFQI{Z|YaGJvJlT8wMG0+3j%_4Fs%OPZiv>k5{#Gmr+l z_-ja608DJ|WdgddSQd|u2f5`z#2`=uaywHm;j4F{wa%iXHU$o3EPXM)g)8V5%+clR zucnTLZp$338M6*`KW7E5IWXNc?;av7ns18~<>5R%3~WhnzG#UBF8l%=yfV4}<3u?w zjUE*}6&)`9ANi`UwJjtyZESR{%y3^bZY)?|Q+*j({z$+6GxGU~M*OSfkbfpX|L2ZM zvVN~7%Fm5}kD@a9Z^9F%#X z_v#>^;_cUan|lNi{ix@a>WnOnR%(ln7%`A`d9kyG@_``m zoS;Rd9Tb+|w4Z3pufDG?AaCUM<}s0`Cc>nUHgE+X5_a`fL5_w4L%dZwp${#<9aJ68sUV_6>5`UW1;X7w}J|*gA&&f|S1m`2{ z>yshuk0hGTkwxk^;ZkLtW*pfd*{}FqI#s1e8Xa8_Kj2A>C@(xvvy!`nFW_y6Q>2Z2 z72TM5#?@hAidB%UAXia+0t2aQQe3H5wt!*wqA6yX39a(M-g(cXGTLlSU#DSZY1pkP z5hUG@1J^D57@o5&5H~ z4&?R$5u|86f857R^18dAk&~bzq70{;qYk-xd7VYf4qwRU?snZm$RnyuJjqC0*Ea{- zs*?o`=UL7ou^YM>uO$_G%gseK;o(j`4W@z_brEnEC$gRI;0koz0MZBHyLhf~nUQ^c zaJ_!-@0Z97u21zy+M3<{Nq2ZbiS?F`qeVF(qf7L)8YG^|hDRmQS0H0Mk&`(?+u%3o z4ku6tpZgAW>Za5vd-?&J{OE*<95s=`iS~)2auX2j4yD!WBW;dJN9Rk3W*8yi$NXIm zJ)Q}S7iRDWEOFXp@R+Ypbdfl17`QEWELmi*5#m1H<+vg@0_e_~Lrz65LlL`dM)^9^ zG4VDBNP4L1m;7acKDNgbxQGS&D`<#Qa22gpebD6X)8u6ShSlo>h|lK`&b*--QljuYvczqcCs0pw zw*nZkCxvRZRS}nZ15bG$GFgM7*0jYf+4nGE9j_Lspd52 z!pJpWmno6mPmjbhd?$18@57Gli$3x+kZl*^-O&#%ZqFm-aP4lDuW zmZRhPl7|j@Mmz`(t6gq+6q@M{a1REU^s5;4AGFy2+wAPua$Ah7KX%anOLq1T(XRrI zp8~c2he7FUIT6O!Jo5j5!G6{7S3&6y#s7!7{;MeC*Fh=k-_H{MPlD1K$$v1|y>Op@ zVO~Fm^ZO+KE`!a$`nvjmK!QJj%Ma17*6m*hrT=U;|0S7&_3wxBe;rc(Q1;sZf8w8i z#Dw3abj&RDe@^EA_Bi4{`py3wQZCvph`_t)D+gN8g1C0LiF>&qj+ zLPvr2#k|l?#Se{?Vl+}=D%s~D?;X|irD^irx~@q zOC=@rq9WJ(t;l-il*vIxz4KEuDY&C(`luq@BQ?rYA|rcd&14VO6b4hmtWj3b0BC{3 z?BmQ@4fD0t;|?zXHbbbhco`(F8^~JF(SR?ndE%p!sBcBosbMLpSjsj6R~Xs$*O2Lp z&eoOZTKfmmD1Otdeby8dNpn9^^)1|RJ#pWlhU^4latfuX=EU5IF@2$3k?Oq<==TUO zp$M7EHp!#=-s2ugqL1Wa$~IpFH*u6&jH{OQm~&|SyG&8oWkHfF*nOsJ#P_Ibo@x1d z{D|}$KIqu_E_07Wmk)-8itGd2uc_W6%=ZoSx$&tc6IJM@ijRYR<$PeLxack^1ltb9 ziE0eTSj0+<-F3GyRNpn^PN6X-o2kw!6eRGU`pW{}r1Xcojeq zx#duxlrL2KK{FLPg|XRm4U3#1JX;oCHR@Pq7+2m)2H6LjYLNflPg3u#Cd4!g?3Y2r zD$4Fpj;V=_8-h<}iO%zI(=U=yMxf`5ECrPNMG5H9N@OP#(mPeXd(zy4=g)?+QFFI% zP3MatV)G}#zQ^pmE5An(n@Z6^gx+G6uM=lFW z$!7P9Fxv0rGCq3Q}mA)U=e%yJxL@^_^S7_nFE*nT^mA5r`sgm%r zH&2ZOxgfrT+>QZS58Jr+P>%7Am|>H*PPd>$kz^Q_tI(0;QP3Amd??A!c}2A8s|__~ zF@o>n*AHdEHqJUkeNNm@Fa?N6J|_wy`{@yJ`0r_@6%BhKK>I0sp{*nj;;DCOFgh^| z*A(*y_p~e|vkkdFya$N)WV6X1tZ*P0t`J$|D(mG_45qTWsY7i24%g(DyRoNBbjatk%!MA?HjFZESc1E$U8NOUAJBPj7$S$7}w9jk23 zXpe1w+f^ui`>X?W9@6a74a>)Me$p2~~_K2uwd9mEIehrw2f+x(T5s&7Cmm-0vP7^d?P43r5lAq8C7p$otyu6elPvij7@ zvXZh&0V!@2kGhpVn{4z@X(T^cx?G+=o#C7shK{7`9TQ|C9*gA?3*Wv=I?eW;6_G7) zSF&gzdIb_C!0gOcdB= z9gSbm5ZOSQGRY~Zb6#>A%a@TG1n!xaQ@}~CbJ8SAEJ4~^Y(T2&o641-zknJ0F@5}1 zSo6Pgrt!781taT!KhyX_^s9*IeG%dHop9?+^6vM$NA!vKiU_z$||Rnm;=kslZ%qBdxzTtUTTNGA}^b^=u{Z+N42Z6tH5zqqEj0Q z2{N&5l=|3pyw-baIpklN*WW+D@qH`l`$p5NwfyMc{hA5$$2QYzHvEr`CfZlL#`fP2 zX#5cU%JTkqywhu6Pe!&s@}mBojiy%(f90KiDE>dp_22MLe?L3wKk-hCQvdKyp>U$V zGN1q8fW~W+OvcxeME|_e^h5Nkb^G_d6BF&<_vwG>oqkjH+W`OcPQSUA?^1fkKlUYyB1}|%Ch$NSNqi0rZ>S@?r>rRg`PxA94p@CqX$_Uy&YP!*F##! zmv{l3iLw#JdN}$x^DKUyq+8714;xZ@Y@Fc0uDvL7Lnb*d1&6)S?*-w`Ng>mF*QE>? zShiZW%Lx`Y+)DKiW0IhO0>MItE2K0oNi7uNF2U^{!1xju&HD77*)?mn+Yiq}-omFg zA`sLdS;X3)MsFsVjC}Y=0@*wgEVW_95+t0R?p4i|a6I#Icflk~$LlmXEwpHol`^=0 zYuB;{E@oT!-QqJHH^iog@%u6?z6O-HA8vsXoz;f;0vqKxIcWtK>}tYmIKnuweNS-5 zaY0ge@-Sx+vN*=y?5k~D(Ao)SD_~@EO_=TW`Bg2lO+2nZJoK>#cUA~L;6dq+=B&fK zzUHn|L+QVK`N~0=^vnn?-|C?ObNJGpH`z_8m*8JMb5fsPvuoX3-Gr+b=wjEzei1?m zXJ4LxwUr@s!lv?Me`Cb^RLzbR1$GtTHYW$(%DGx?AI;WWQ3__F=#Q^BI-dbwTcD?p zLK&S{)D%l*JQG2s9L+?DBB)Ffy})SVI2u7eDm$6Ckw7JA)tL}s#w9}te7n+)@l1Gg&ujUO+I1uqV_ZO2?;gH=qT%XNrp4yvJfRUUHLuv=?Tg)S$XJFMe zPngQUuxN^f5I~?ZKID8#cp_u78l4g6#qc1)%E)k_CB*1!4&y zpBF_5te4fiL@*SK2gvmYi$u*Ls%hI$ilo^5@GgRPmgO(jW~Hue>2202Ve~N9Mig+u zOl7!co#3Tcd8xa6cb`8@J~<}h04(%5p!IOVfG=Q?9Frsmg&k0BZe}`t+c^qV(tZ$4 z0RkJnDQdg-2zK}42y}ZnAX&Bxi%uT9UfQ!vR)_Udwr!lM2cJ?3bfPi_J%Jk}h7dqB z?A@*rA>@bI`y7{EeryFcQ)Izm8dgAdD8uU{E*lwj$3Jd&a7|)`kFf`WEA#&S ziCOg}qu#p-Q_OdIp+*k{_RfHDTd1fLhWuo;iUf&}ZPpZlc>6IFz!8~&c_77${Y{FS zloD+%>|+*%IeK+z)=$%Tk*l_RV)Ftopa9HoNS|<1;}i-+nQqy^@(MDfY2;14INe8? zWgRJQNpAg3wR1mYl~=uEC*SL}t5qC``5<_oQ-m%hnKOP)!ysE4?ur-@rj+j;uk^WJ zAV)|#{G^rw>dedR%S_T!RG!X9#cX!RfEvYE*=9Km?)4tId9dpSDjS?6CmsRwQi(6K z$7xI!>)uxfHsoCAjgPy*!nTS*$L6z3fr;3Jbj_` z+89%&B8&21zLnH!%byx8-%M34$2pgKRUqS4`0@|laIs+!;R%=X#WIQDp9o^l7_c8( zIP_GGWj>ypz!47BN*#1anSy?N|7=WiZYruXl4qk5IgduRFKD;hPg*YIqt>PUc`p8F zroXFHJSw=vK7x+X_+l}xbLwL8#nxoJM|IUZ2ra7Ygot2fldCU){mo#tuQ0JBMmwSr z!8qwXXH^GpMz~cc?`~ZZ?sA||ma~8Mr=h&bK#pXt1I%h4BLG5VSn+?J) zfgzRd9lfgTBPpXlUI)LwxBJZtC^sr=}kgF8x`Q;N*euJzb1-(NVwIVeJy}hN-49#kb zR$e8lbV1e0qJOd$dwz_nyOr;qy$!6aL@2!80proW_E29=#}`WUAZr@X+r;W6mK-i{ z6kXX{v^J_~*X)}HDSUpZF3QL?-gI+Uh(6&-&?IQu4-SR13-zQH%P;=o55{`uXe~=Pu42ii>;OvUSaxe&(nyUiC;* zQI=tyT}KLNCUi^3rnQSZv9<%D(y{S_LvUzjV=8w7Hy~soH3^;SisHT zgY-(trSvl#j!0Dsbf|Q`5_GOa6(Y@rY-XBO?LkMteFi{;+^UhlFI84eAshTQk9@>) zSmPd?Q2=v@uIZ_pqxlxT$mAI15oFmpfE}w}!*0sEk5}dVGTmzA~jZJWgct zDtREMUOzT_B){}ibFR!$7gn{}ZfoRdu+bOb=qr}bI%{(J58Gh7PZM@3G#8qx`iX^|Y>ztgBq%jvc3yB0 zrw45^{3=VZ%qgs`RT->lq@sd{jNzPDs(rT06B9mAg${Ec#5|_0ux@Z#gzc(@iStt1OQm+>?bZ~lVGQ%fF;ynw7=Ua>)5wR2*k!o*&8G=wAzpY&dawobbJ;VCG zKa2bCn1YG+|F?rwKVj7Gxsi;okd=w{_p?&}3|U!T3(GRSwvzZKIP%;3{{crdt#!Yb zFnG=L6A|GRkfRVa)Y7%mv@tX{6Ein6$E6W8(>B*JG}FhWkvBBsHM2JSQy^q$Wo^T6 zplS6AkzaSbX};?b(6!dKGPJNUx58z5MUt-{jY}h{`P0W!E(Ma2B**IG0zCu$g zTV0vouFPf43|~J+_f-+oYhQyOzxCU#-^M0ysAFSb{d!mVyUJgGG57m{PyQ9k{AHE^({CKkPo^ehZe{v9 z@*l?bdqsuU1-G`b($zGDbV#Dug8WYy1=`R5U=*z3enF@IHk{ul`FHuwSGdje$~^tI zKjEK9ivMCG{|Te;gShxxtM)G#g+GbxpG^CQNBDguc&)W{U;Pg2chAAEY4JhVP~YJ9 zAHO@1@As$}UjKS%giJK`t^aT^-)H;X5m3Fhsr+%D`E_|-m8kyae}ptm4NV+zN&j?X zxZnL4+3$A^O>|!ulV`$4cXM$y{WgA_CDgG%;^O0+%jpitq1k8 zXt%T?r)d)>UXj~3-&rXPifA+qX00_-kWuBo;Y~Dti*b~4WYrD^0x+*!hDyxRadNoi zKc(9I(vGJ3rZ@qV5=Q1E`?mBsC@g!+)!5KIU@>Ll8)j;vFzEz)Y|N0B<#luWSmQg{ zVq4&AT1Lp^=h|5L2rIN9dmI^6r3xc44_1^B|H9rbL_gRLup7*7Yn^7eW{qa}W-TF2 z9p;X&icdt3fp~Nv6WMP4*A!GtOwL)gp3a3Luo0kJhhfl3jbL+pp z&lC-f)9af|#9UdOr%9=txP`i>j=}rIKi(fJ`rY6dX#b?Jemwr)t?*wzkKglVe)G0; zxU5X{j6Xk}X+>ulY59iL&`9q$5lQdifdEG!e7%Lg0K9XIBnZ=no5I!b3U~t@5K0bd zH^r^9pv2S?H&dRRZ<=g}T~mF`yGT(JQhi*cX!l?>d|cFcg4N86zj!27KMqEuIDQ?vVCt!`+tfZs8nAElyW zFF4O?%M+3JXWg|!dFbL6*Sk}Yv|=2BtjQv^R@3OB7c)Eq){wCKxwZH1Z(TJh`1(x;$Mx8aHo=Hu6^MO$usa)t? z>X>2W%V&2W3ai#`UCTzw666kvopoViFs5a7?!}HKUcy9D-=zH+p%X=X1hOyb*#dVH7Z3UcDic6`~QfdMs z@1UY^7o#pc?Y0jq3~J2qGk(F*3l>6VfZKD7m7{(pq>@taz_T7pVdp(eSI44$>D!f!7mjRf0ptpN^(8yAxc?&rME~>*omeO)pRqQg!cdt5$N( zM{m%}#*ubn8YUT~QQ_#_KW#4Ed2ON@J%axW{HKxa z&8Md}6`jtS1en}Ob(0$l?2SvzD?-kxO-r9ZMR!&t+dUhyd^^OmM_2jzLXSc>^;6P6sFfF5NI~817}}Ua4f&@Ha#0me+ZHEovm&qe9ga;nRov9s;+5T$ z+4>7R1n1}8Gmwp+F{dC_?d)cT4fjkF*P_2|f!oMff@2w=OB&4~gV?m9PzKo|5_?Afme%kgyP)&gZVPsB)eX4Sm{LPutQmeU(*GRbO~zed>8_ zy>m1_6+p3AYah#wAzy1PYi$)tMedY?00K2W5)LJ2Z;N9F(;cc4@qrWlsp)jL3cvqO7VNV-V>Zj%} z2W%Qku-7eP(p7D(WxWg>wWVJFh@6A?Iy6 z>kI7Z@=(AIky=ukldj7NtJ%5U3a6UqsO-0|u+t4M&RHg|SVm=uK%NlP^4=NUuG^Zq zeH#l60W1hQ1IHxkmEPrrD_1h-2kpa!9ZZqbZ`5(5+sOos#-!opAuG1DSlHDk>L5xK zs|+241iTAp>$^Vy^T{6`HNdcvP7bJ3R>U$g*LAX~V8s!MEVOcCK^WBczK@kVrETsf+;t7anRR%jSvrP=G&XB2kD*QO=C-q)r=jd;rIowK z3gu5Jc51`5sk|-KfW(*)0;h7vW>lK=#qr+#WqNQS*7Mg99=dFuLY*+3yujLJlY)wr z#xwjAZ{N>#x;@$7V@eYzF2GQCtM^CjPbx-~-^;Psag`N4bt5v%J{FCb(8GzzY2Wr` z99#=tS$wt{j3EgB`1PsDN#`2hH?zE+sF9A?@hy9|9T*9LC`^|&8)OE7*5*xx1DYaI zTYJ<=ln3I1Tw|4;Dpj$KrdSP?G97{Thf&eh6l@87m(PM`bKbq8!eU)I14)I^j5H+N zf(c!0m~Z zJPwjatP*jDmvsj6al74^+#y&n5MUtl3ZKFzDdkKTh8*cl9V(pqU(LG*VlYRS>C-Ma zVR4ByzrpkzeQE7!ICP?$9AESoM>#g(75iu^%sdk{yq&E#4~gXRK7%9#DfJWw3=8a5 zi#614&1ubY22N}8jVI2|FHKC(Xu+}|GN;An!!Lt0FK5E9>0`a}$gxE(CX+D)gO|uk;ORrq{pt8ImbhMB(r9InRr??x=OqcQ&R9^$MF^_Bg;X@YmJ)H8S?b1amI)p2>qQh!#)EFDC7JH zfD*oNJyBeK}M`9iAV4YppM|UZ{P9_l-H4qZa}4VCrzzqpN~}4 z9u-xeWVs*TI{oTa)zGAKJ!$W{d*1RiAu9n`lc;&hRb~(A7KYSY2U^;`NZajABv2W- zF>iy^47>ymd%vb#cwNUPGeHMunN`d%6_vsTy@cs1u6xAI~DfwfT@>)LM*y#iXt|WHY*7TE2qBI z$SD}k+w!iOHIv8o?iU)DJg(-#-zwz_90|_ z+qtxRyVHjK8Qk}TZd|0c>egkRy1pzj-N-Dt#)z{{9|6}uSo6su-+}j{vtX*pomFI( zv{sv0xIy5KPc&pudu^>au3OYrD-__W7@FSbgnmUHDx{c@zAbtfQas+?U-w;U=CIRI zt!{j|D_QS-2sC2m%oF4 zz4mEsd`rQk*d6d--A%x2nQX`T$_jtx#T2m$2LlIusM5g(24i#N;SutFqrQxO#XIp` zH4@7UfGmF&CBFC^eCc9;;>0p3sa~>!pgGebMl-q-MtTKVFUxl&-HYWBN+o{$v_qh| zQbZXtfn^XhTE`}d^Eyn4VnR)%gDKM`3`({_wE8t+%O#z1SR*qY*-$36V`OS+?XtsE5y zsYHVkeofUKaK#;#Xmxqih>%S!@-3aV*ejMMaWb^B`b=-ppjx_IC^kug5%Kx+-j|?P z__(!fW{=Ghn@9=B^AMiN(LmXlC23H-p&ePPxC`l%&GzX5SwD=MQCBk{s+BSe_bcTW zcCd`E_y`Lg@O22co5gq7Pq_C)^?lBF^er(wCj(-HQL9gTq-x%oV{W%L&gfOMO9n4H zV4avK=dkDRo>*N%#NFK>QZMhGYC{-#KYp9Rc2D%h1^x`@;qA3%1>r{u+r{R&2;7Ck zr_|=>u@FT00(N-ses8ey)x@@z7#xkTpvBw~H&bEU9!|ui$im%KV&$e*Rd7P)8vlXK z2^33k*JFWgxe8<1%gu9SdE#02ir5MHm|uS^)8M5xj|Oe+P%;Iz{Wh&Ya+AO>^Mn#|BMh%3;UOpfwc%&$zC-#~p}Gqeq`nk3pE zr3Bw|^TS36Re3ylgN|d*X+~9f-9g7W13u8H1;k`ZRe{QRd_iv#jP)JiZyZa0Z(3h3 zXiamTxz&}b(e1o8#r8InB|Y0<+~yL3Z@oFysw{DL{8NF+>5#(lq>Wz^-p5(@FD+oH z_hdD8%-iv$+fVz0C{*5==%~phr zJmFr!o0>Igv?l&l%2F~cqv^qA3z=#cQV*;6r>;9mU! z1y^@|%0_TX-1ecA>qj~qM2GuiWRKLgPwe>bPtD&sIoubBjLIAKWYP<1s8FM}gJF45 zH3en5NXDF?27zT#rI~Zik5MwhqKX8J*Wsb{>7;D|X&PXnFlt5Pg0wGKdPW2dRMFdl zJ!)Ik5x|>RHmGnBbr)BBG>}S^tStCE!#v0?OVlc)$i6Tbijo?5ny4`zkjFwq-1QJE zsIvg8y~&6gdKy8cRxfD~4pr;E_9Yp%ubc0aeYKSJp(4I#U_)Lm7DM%ZeWMt>*xVS+b?@*b-9UM!gV{Wed zOl;t0#m(iErigQGDTDjEsLs8faXrl6TgY;##-zk^ubUeQ0laAgjB8LC=(!d5GXSBz zFc8)W*d2%5g&=%r;-dE_r`pw1GFcWG+!GxPUY7zanx8@Q9}H?_B7Ot7+5CL52YHrT z9YFR*j@fnH2I5=>N`3QC>hYqiU}g?oZN>or8P)a@;0#gsLy|1oB#QL3EQ#n^dDaCJ zP@dG)8y^ju2j|{2V8k<`ZmT{H-G>rC${GN?8;M0yR<#|kch6F^;|a*qE)^1}5ygCy z(9Yz8kR|&;dHgxP{WE%Kg7V^_5w7jXx_aEGp*5mZyyHYR0^$Vj8v}QgD~=Petuf9@ zQ6o36r2{2svHgo9>hl;t!)pRUSye0MfI&~*;RAC8WCHT}$ZN32vS*4WG~$~k^C|a6 ze7L^i%Rxu94dS>%V&dy!%LY{u7q=ogf47mvH5Lv6NCSw+INhl{Qy-aD22e*Xt5Thx z5~@Lxun#>L26eyoe$a-;r&Tat$TD|lEQqx;o1k3b+hWq6ih7(#!I^@aj*H|_yr2j zF>z1?iDJN_%Yvbb%j?F_^5vcRLF*UmRpTj;s@H2uaw4=o=!tCttt`_~Z!9N&!p~iY z={Y?KNWzU{6i#&NxlE(7N)Y}alPwTV&gZj^Bs5+H9!FT+dRgT+mbWmF2NbQ&<`GW* zDIr|5%F0u;e+~@|5RHKdK{o#4OEjj}7rL!eTpZ~mYxz+P&V|f7)(T+!vu=81bU>_h zOH2gQ024USSJ0gPF&qrhQA4Cec8mS&tw}FY_u{6BG#PFS!W5HVq}229Uk`T8>qFC~@+ zZh6Dg%toPxJY&r~&9O2`fqWJO<82~}Tl}BkGKd(l%q9hb_Q|ZV+_9$g`Qk8IxIf|g zK6aiGQQzMxJT>d=y4eeEsMTN0M6ar)Zy>#x><4VEjc8C9b>-(_!!OB#2x;z8ho|wO z308}sF}x3456nc_4h*`-Z_&84cHL!3S+z{LX)Zqx#xft1SMM$L;aVC~97?6)a6X~L zV&`y*i5s=h_d69`8|q=8Nvd}dvVLHL?UjfU^CtO9EJVUMu!>zlJON2`gSaE%BN3bu zAwic((@*p6vcIk0m}LJAbHIu|-%=hYF@hpDiF#Q6sZSf!)}Hl+_12O#N(w>CrpM-% zzy5i>J~b!*PFy*ftOmjXXkeViC*m%CG|vuP20cLx7#x3WG)s zqRlN$=a{Ub+`Y1XeAm_ZwIN?M)~f?Na_j->*e-rMlBUl?5%+kj{2sn`oHoaJ!S}jJ zp3EDF;Uuspw-&31q872e7W1e(Jucchn{y#_o-h?)_g*sNgE%%iAD=ubfOXctwLD|o zPFA?iHH9wjE(&+Mnk00!1Vn0<_vgKZ1OX;W{ji%o35oYrv9Rh^4YpQajjPsK9=P`K zo70JhBgBb+a$Iw2YwMAlk{!;9B@OcFbnb^vjjzFL${1-nTEg~USI3JlC(Wo(86izw zCB;cfG?pPo2=cxAvUNWW#(LN3erBshJRp+HmCdpLy@)53@1 zu0`{9x^^B69bih*;K=26zsKF9rRS^;(-dB{^8SVbNM7)ZGq+5yIkvx*9b)>aZ1`{4 zrr$YzdKR{ynEwCE^QTO&JB~lL*#6uTOjb~}UaUiMoh@z4TrNj-`+z@P%ec?H@{xI@ zhhsme85R!WW8OzT!1rVarKj~MSvrGO6T>=~gNTkiqinaH>cZ;Xg&6zI6~&~<4%PjE zqIDHWr8<;yW{80bqPx}~Mo0`92lZ`wQN}~yd0&nW zbO~tb1~ZHLx(SnoE8n*2TfX^Z86^S6U@7CbzH4COV5TGxSp*jxm*XhI+3*>s^s2_# za-#_OuvIU2Z8h}hEy~E8;&!LM_d}I@{-`ZeDE_A3bT+p)>Ipc@BO&Of3 zF>|$-2+U$L8=JJ=%KF6b_K@Mj-^7+;&jO1s{s>epnh<7n`-fm#;HD1?2GRdxypBskcg{GfZwx1pH*_03TquAIP#Ro5N{oX;`3#W@z2Xu(H;`9LHz0#j7OZ)hd8wAQ#l~r4{t7!EPHgx!7MB>jL11ytR@|5(IC8(5uPF<{e52`*a z^S+m=K7gczKIu`4ZrwYLykE2+p*#_A%R?CwtKzGVhn@5r$NuE`rrsZ4lYDBj7`8|dKXi!Zgetg_j%RlsWVY`jx(s4j`|!{3W9FG!yb=R zw9$h35N`V!n9U5bOj>ad87D5PPy(?Yx24>>$J)R1V=C-x|9Y_Z%zpP;@RDa3C&wRk zuA<|voXoa);($HV$KvJeQ2hNV_4iZ=#k@qP)cU%Ge6tFs`G}_~Osd|4RMe`!87a7! zD6?XGZOuM-Z-!UML4b+Wj9qiz6D_5S5rJha#`zZRxcXKFvtR1nL&W?$j6LL23cm_? z!p;;N1wy7YKHuC3Juo8TVX|BNg5=Jg=!^I=0c@eiKs4r@CTCk7Yn8_ArVA#Z&*Fnu z%MJL7I;V)0$uu&(A2FoAEFp4=htO?=evp_X9?H4rl7idf<~e?g2eKHTPmoPR{B%Vn z$Wj8c5Zg+Xiaue7xtDZAMtb&s6$7RCMCEA$21aAaR;o@T3PBFgcm;aW)||)nEjeE< z8vWAg#!cd0;0x_YNUyPdSy+@fQ^Rxw(;?lVm+CtMDp>sUl6(vzI3&bi$1e%WGKOPv zaZ&O1+8dQXkljElRpTn=6fi_ilZxxW6id>JChAz19WlXVtgT$K_JkKr)ZrxTrMFBi zW4v1_GDoP%*9Z|$;Mk&3{+u&1I`vj!<9ar7l2k`^472B?r@UhE-q zI=M`KU;~B6qr~)}hQU~XCl*E^$GlG`+M_xlx(INm(?r+6ACm?wk}Vc_h`~N%N_3mh zyyQMNTUeuZ-!y#SgzU_o$jE-P3N z$C%F=geiA`-d-h_YYNeNw|Fho8}E}`<`H_`Fr|U)NLXA*ZDj|UiT1X}-LOEi9vR%x z2%Y#%ptN3q51BOQqe0MK^_uQ-3&cl`MtR<2yO^zgn60Rb#L}FFTXajF_V#v^c8`}k zED;#LFS7KxmOB=p#UTbmYsP^tT_xGHu&iJXIGFkPl6>^V|nq#kW) z_rA`692`(`73}Yk6?nDpJY_bS1er6KL9#R@}7G~hfR4WI32dg<8p-3T&A=*0Y2e8YG^yw!ZSrz_OLU+DCG$++gfu} zMCTQhT9E`VSRxkOe;s^nSbEs%FlvE(;y3LzdRq7c?cDYn}SuzETcX z@J4*wQM)>YXQ$%s6&Y!-1*Q~9j2p28p-XDdhymi{7Zt=1J16w37#t7p%Un!&3_+u}axXVJ^Q#izVrtH~SHT`~)YztEA^Rw= zv?r40KEnw5rAyr0p}K;n_c?!HV6Hs!@6thK38kvTD~U-PH}&hV7{&mbXPB;IYa6%j z!qK!k*A<$lOA4yp5QW%9Ad!5WOz`u!3bo~F=hAfxL}`{nmH#}a{;0&#v`wuAlLRy! z>AI$k1N|XM1mMGp7tudSN5vEV0+F<|netl-j{h|;@=wyqpX_}<%zdn{Racnl|EJ9A z$IvfIk^dsI`bnPlAC%&~7Fl6_l~Dd72K!f}c&`zDPm1>=@IRRAAE{-2TZ;GU>HeYS z{8jb&e<{VItFQRAisUVV@^74`ACL3PB>$uokM-5`&CKvWtjB*H`mJ^QH>7y44)H&` zYQHbV`w{G~5BQTP?FaGvWt4%5;qSooDysXnrt2SVHZ$Wd3-Nbg$^{JAEQunuUQ!)S zBFevctNST}n_F$n7+c8?9uqHwFPa*fZ=%3Q@8M`qAFV;`3rbh;X$1*V!Qgdk;sIVY z-sV($uC2+wuz63uxt!3~+15K^!OFU7SwF&J2D!Xwj5w7Mo-v+Kyt@YSQ2eifS3B9F zTllm#E}O;shS8tD2VBnx?)ygJ-VJq8L-~ez?%;hR6dh~{qLf1cAdqRNC~sz^x~tM` zf@L7W7cXq3pVs+0E2ST%qc)*0@iR9$JOF}AF_jGV^*$9h8pSP!C=F!IJ*%31L2b3x z$;WO396T-%Hoo>GoIUOK++?fFb60}(5cgjqFFtMOeuD0OpN$XSRUz6V$k^O9Go)nC z(%^X~{VC>14xo3nv$b-j)Q|gpXtu=BRxWk*Pz3h8IiR*;x&%Uh!Nl#h1e_|J#Yug> zRQ@^uSFdY0MUv4Szs-n^Xp^p_zi1;{G13w1?OCGub@sb2_Dwvn{%oQ1p$=GC0i7Mx z2XQ@E2F)|i&xFYH9DYJ1plUaGuDcX&Nbc?19-NGIv=|tQVpecS~K=Fq_n+ zUsdok<>b5jEo8U#anaEkaY@9Yh!tOxROL2Y0DZA5jvYti*BVIes+@9Y;QDC#&QGIa zk&bIo=^`aeIuSKPG$2Gru&9umSst@cwL&utgf@Hh#}qF*q5B~?dyw&9^%M->o7VtI zU~+NY`d4jh2Z0`j^x|~yi|IVizTBH1u(RlCD)7hKX0_8ZJdXD>z*-p#0NNbj={P*uIlM(;;V8h_!kFa; zgR?~C4s~uWvD=Ck6Y(%aCNz`PwnX5Tn>P}lFSBS46`;@u^TvSq^aSHhC2F)eA zMh;F5(OqsYMk7GQ7usm9*jXWYDsr}!Flao$!6EAsEsox-xS1Z1k+hXvIYdWw~RT}aM8Jzk$(1#bkf=oCD++j?_OR(yl&Jp z1O;c%un}2eL{X=Z;9T*9lmE79mR#nP3Z?oXI5xffi$u_i6YKtUZ?r8Z+w_FoT{nzD z4ajKDkVzhxqqo{baaCTERZ$j(u@3iiCE2dXlB%WJN={T~#l5p4E|)Tm3uSyp7*G%X!J%~Cjr{Zm>_%hi<>*F1H?hv z0ySGPHvMq)XIVoCN-bVtJfEkQgG7c`czPQ_|Zs(y3Vk# zPP(6@XNkb%m62HWH@fohS*>qcw8trGgooGAyvbhMx7o2-hvZZP! z8}XzMxgv`|X=2s0*ZqY4(W`ddChHz1y-n%(-$nPA?cPvRds41L?WXDoV4#$oHSm(L z;RA!7fGB(h^;(%SzaAZepX3Fa37#JBElL=NaP7ES+!OWg5Ux4-{7HN0?XC%<*62O6 zb?l9J@Lr$>{>vzchQkut7fDbQKS(0qQi#Z=5ZZt>4CCJGinrXVW`^}(P;v|PPI1pHKFoiCF7lJj^wfPF&tOfQoKW&P~ zX&{OwzEO~v5d$83D4yFRP2MNbdp3oj+?Eq_r`UIaP}nM#{JS^f53c^MPLG@h57ZVs*-@ zbMS=JyQijN!k8fZ;LCHa1@Zi;NDk@#2m9{>$EA@8-*S9HV?PKWrlAf0GA+MZwB&_uL+twz{ZJ+j2CDdRim13L#mVs!$0a>;S~64EPumRq{v zOSg&|C0I0uoV8v!t&-TpJ3gOVYp3%7lB(&0!gP4wWSY*eU@TQYk9KS?RO~TX?o}Dv zI2215=~%i7daKayang}EC2lvGnFHc1(%O5+BX%CM3x_|4Nto0O3Jj*J+axzP-4~<8 zJ;~17xu4NQH@{~4-l*-{7olcp$ut;~J@sBd9oa|jGk!E0(YTm}jM5i9Zj8`CVO|G> z_UXn%Mi+)RQUa6+-=^@1POsN_0bKA@jJ;#+`=L!ppcu6ORR++_3MC5J4eu7ihuUXrtqZ8ISiT5|tFEu;%JA zElv9@Eee;pTTIH-#0+DL#-Bx3tQp&1ZwD?jUUKipLJv>c-xE0Q^id}Y^Y0rzzdogk zmKeA(GUMBJ!|lFFQi^PRsc}i=$c^wvyk4_>eN_1q&Hc_5+;;yu^jp{U|366gTDzZ_;oqs(!ulHF z_mJ*K;D0dJKf2}r4$}SLP1y!xyDNu>K_ zA$~$Sm8FQKenj^<<*O}8{0U;=iMAudo#DM{b*~gdPOyjy-4`_(VSQl&0SyjW&LaKDeXs{ z{k0+UH7;rps4$+LPj@P5Z<0l0!qLl~$w#{dtDS1q{HeHza2-PXI9)4zO}0PV0_Gs1 z>!rwC6(NhI0L3%-C+GxH1SyrcHILQC(28o6rmhzcD}lH0*+)x{JtppRg&yOB-!7XS zy1xsO1U|~ZfB$N(?2Z1|mMOHE<+Xo7_J|1=$Vxd{=>v0KP$BNNFHVU?b;5qwntZS%Sd1w8rsUW7E zhK+b=o>lXB44OWLqAVX~b1T)iR9HJzXTj}6kB7=)bAXWV1oQF*{Gx7`{UU3xm&Xm# z!6$@@)i+1PZ&+H0rMKt^=GE8rar!j($zC`RtkJdx%zQtM0okI1VTofTX97DJQ)sw~ zVf3cnVI|<6@yK!1FXB%M6pjklEoM`K1y-m&Y9f$saBxA4(T^DRb0vgSB;S_S7H69i zPv&C*D7h>afmJNS$vl`sj(jD;>^tGtDf=map;*ci1Au7sAnXQPI6%EvRv`JrsWKG9 z1zPg%Qqb&_2Q+uJc+4jZvsdJpPqy3mgO|P{HPwaeU%0eOwq|*59mT>f$CYN!#bNhm zU|J%$E*v@3WKTM^j8Pq$ZtKh7gbLc2F3{llXXQHe5}`Unuu4$&!b;-Ki^~8gJ5!n0 zx|Eu>TzRSVBii%9x{F82!ZdssjKHe9rDb!ZbNe(OY&GCX zkE3xM#Et`mN&_aWY}@4=I*8Jg2TjVN@F=;xwyT6*%JXaGxBQ?W^i^Wm5sbkds6z!} z!KN!Xyr7lw4Woc$b&}PhOW-uIV}PKS3f0fF!Z8nB8epI*5Qq%kW_*`gcpH;w4c8st z6-|}+lr<9$9(=gLw>Pf!-o`xY)U#&HFJ|Z3XiQpKX`$fgZ8Bhtn|hy zhKAd=KWG_*`{0KEDuEyy-$|%m*Jm=Gn*J7?UY|W>IegYx`@#b21MMQBCO+o$UFfT2 zRTJ8yi}_eKYwmBswckx;71zO|rvr28in=GbH+nV&!jS2)o8E@vQ@KrT_)&6QSl-`5 z$85WO2UJB8%G#kBM7_aUyVW`U%2>4KmU)mf(i?hrE=85;R|*7`59UW4TWo)L=KS<68AGqnwDN;qwMINx=p5}dFS?QalV`&J-aOB2$KGht`ybd2b{=jUO-vDR_8U(4n|ECptyTv$9pddH zC`oQ1EIttQ#mvAOMn34SnJ56i^{K*4I$iyqGM58pvOurp;*y$mB18kfo_|C?Ctbu$ zDJHoqWece@h*fOivB6;zn<1mpwwdvOyj`CTGEL$cF3{(dTk=>38f6%LKYq zHCN}giz)lbEhoEwM(Q$SWm4)p-8+aDx&W1# zSvhcl>uP%!;COokkIYEmc)FlT@`r?6nGD?x*ll}kU?sz$G2WG?*SCou(sJ5%UY)8L zA*VyFfpi;1fF@K(1@Y<7pKY3vRZ3_o9UTXU^3q)M;QNbW0MPpYaEBBj9h-Vu z;#M5vX*%8ov_XVgo3+wi)lU$cxi-Yw!=`w|+&%gzZ!J1}7!;z)EI70Ad?*yQaIbni zI_DAKpECvyxj5&x66_ZU85ISih<&&${l%)l_sf#BbROOf9jPP`#{t0E^ucm-j)#g8A3WSLkh?ir zEy&r@L*i-5yXT3)HE_vLUt*S2mrP>L$*NYRLTvmTSbS47Wl_FlV}-?Gg+ycY<5F=a z%39cS56o@N+PA%IGoOQh8H*z~^oAqEvGPqM)y}Xvl8tKe%Og88_o2Rck`n4Zi#fzQ z`c}ux<cT!KT?$y3(gVT{D)gP2zKRGJolp#^T2f5iAPc z13lNm)$f5n*fTyJfup}RgyrX1v=aG6TzEoBY;iq^a~vjOCs8LbL>!H@cK(ovS+e<9 zg-=44t|okRCBjD68R8TCatT>Dw05a%pIB>JXu^4*CNY`o zZWS71Ih7FcMavR(+G_(8{wQhP2sF4}d1ir}(TVS{(^C=pJ5?ssg~Z%6-0|@!-4n5qRGVA`(Dy`kw%bJ4{`X~<=1R^5@VuTP<5Y)A5{E%cSu5j4plS-dVt zE&((#ka`Spwf2hR4mVOwAo}7=mh#+i$J^EP2>lxu zSH}N~q9Fexv-R3!nEAC$!+*!1A4C5D4gQ0fEtXeUWBj#J$-k1>dX4aV81y6XKbY&E z#~{WZZ22Fz70DN1M(3+Gy{`Kg-`Ss1&edvn2A`ib8ofM}efH!Pn`^ zerQ1m)lwmT(dLToO~AHL!#D;X)8#e$3aD4y8*xqjA#37Tty{a>gQ@($HMF(v)KB_z zg>`6+h?WkF(REOr1O0PUTtigv~qZ=RKH}CcS=sOE4$TWT?QV;mFn~8@r1sC8(LVfF?-8 z^SgH8t;fn{7*P1OSSH(Qz2siT4Px5ft-vfS5o08>Jy`eF22_KkV^nVy2Xgs*7KNd& z0=^G5vLq(c3?V_WxC*G5MHhtvYo=vp6GJgW`^4`VlJzHa5yC}r9;*u-V#0=Er2$lm zZ0D7UG^<9JG(p5?4@${zD`A_9dpyh5I7Hl$m&3wV7V@G&t5Vx?b`yoO0>K%`(be0& z-(QyN?TatZNO>qTJwAj0ui`s=aEI{5V3g|?$0S8UfQ6&FH#+5#z#pV2^}7}*Ob0?O zk~dQ*9dCJ{ohR(oD-se=!X^;5^Uo<&G=0uSv(d_nMH+UULB+h*RD?niSJD@o>dxpm z7Eh05wwJM_0yWtUaOim~Jq|53ga3d)tGN)D5eVADM1jO>HN`gEvi&w9r^6MoFS_0C zx#Jcn*N}Qex<4vS=TzQ~Z`+?AliOQxjk#!4&flQ`Oa^OQpOODiT*2ibK1#OG!j+u; z<8T%Xdj^}r)~!O=4UFxJ#EevK_flDR+%YER2met{Z5No&t^(Tai#1a1jD9_(ivwi? z-a)uK;L7#n%C804$j>uGRvNr(*x!{JMzw9y6uuGWS$H;?;KQF?wnSPq!PaWJC8 zKEESf63$<@w}Z=xkeN^tf{?1uEr~2;iWC}D{fKeV*HboQU*;H;kkVE-;N86kCz8th zp1lXWf|_1nUs4@iFxjEUkV#g8uV*Qc3hQ#i&hml?vys99VF}4hyRia>~&cf-&Wkha1T~v1s(e5x{9tyOw z+ij;WNNhkDEJ`?0tWquGbw%5o=8#exnZ>s~-6H$?LU-_nh;g4%^0ZO3`+LjqsvzAM z!d_b(YuET$4g0{@S*;>T5hYL!M84JTCf$DKbzPhaGeIVhq_>Z zUJK{0o5QyxSB~wf`Amp3p*{n_@jc$;I`lMy!iS)3cgQkw#vd7SR4FpWmFSh7kl_~w z(bu_D+l!AszLN$5v8_ECj*nl-hw>tM(u{_4f3x0V9%kJu2-x#=J24}W{^$|`V*?3p zPC3d1;{^`m`cA@8vQMq%2<-_5q^qS0*9-1ydo#OY@5?}MH#gp{V6Y|DR)eS2ij9H! zF0a>DD@(Yhz&u7y=0;yKL}f8V$Wk(P`FR9ZiZqd0K157jC(G5WnS7Cz!(F(T zZWTPkfd0VwJz^_KxJoXE7A_8(=Yv;n?Sv^ug`%XJW^(@+ap=$i0WGn576R9x%SzN=5lkUtW?VJ|!UKrat z*{Wrq{g!vP*-pkgaN4U7+4mzA`e{KTqGc(wh9X@S*xoAju(QO?$|JtoF`P=@Vr+&! z)WniA$Y2fgg-r4U1+L%3%tdLl61Xi|R|KFX2RJlL&CV{eK!Af-pq6lDD zD38;lJ%CE^&Mm{OkYJ}S3FvHA$ng`TrU^%lT~QZGGBwR4B*^0?sn%D?qri$Y^IDpz z23APl0$kg*jVkP`lx}o-uX4yU_3$)SlW3S3&n5-F4d=orRxj=JKtSfM@wB-=Yk)Ql zt}@>Zb&n*-u{kb8wQ(un@2fu^Tn31)rhhZBFG*}YN|g_fL$%uj+8xUB6ASVEqPNP- z6{W@)^8Pg8qK)8!fM%;t_??)m*%gsgU*Tu(J`#4s^6JU)~N_V zTtz8JpT^E!G9itrt{bjyq)&m2RD&7qSRj)f?a9@Tg#=Aw2!*!mmNss7W2roSWemS7i8?iKelPO7BNrac?lnriiuZ!Zz1^#mY7kl5HJ%eYHms! z5I~BdGKPffn^mjotB9_dKO93G`U0g!3;eJ~(AW(Q> zGKT79iHR(AhU$avaU%fjZivK@fCjMfO)O$6@gp{AXWV9W0C^7?Ypw4ro|Yl|!YB;c z=b?lS=#1qhDbr}hVJ1^t(f#!9j8817pyA_rCj}T-J?94gxff8;v>HzKUCb2+%1P?; z8q)c)^fF~kq|$v;tC1NJ8kzmL6xWvtNduBYFl{=k0@ z%32+L)Prz0aUa*Txh8qCi>E0p+}~Z@16u#RLzxSI1+v7I=X9yFKe1;GNip8%HyD8F zr$S@@1OxoDnY7m%F*DOIrOAG}Xa5-b2UztVL`bGr&}I6?#`6~nRxo~bCL8`J%|?$; ztMXd8%?h8M`L&r0E#Sw`3=B*xue1532p~SK6YFb)-$TeBf&al=|A;DnJCpX>yZjY3 z{$mKqtn@d8%t4U;t>g2P`0t-MKK~_?_S&DFndw)6|2uX4l-vE^LP(}R58l6zkUxU` zbz^@%+J7To23GoCyBRVw{T(-H|7f$BnSWV`za1{8v|50a?UE>5>yh%+L|GMvhbQ)2 z?C_~sDPw9}@lDkBIH8(+q*$E*%ZZc6GuLM^@j7D8qta3ZBsEfRi*?{L;Fq?$_p8rp zs>Mx8ajj1EL)k)C`c$i8pUfWoY2Bhxg8R4l?8HmZGtn};9o^D)lF`=%A2t|51!+g* zT{d~MHQxBV*risBxY43aed2O(Z-ykKMZCiHj5g4W%Lm8?QH!-MM^15}*N<$jxE#=4 zgf(c;wy7_})Xsz#>ba9(KDZd`dpeTnsl8nqsQXkbJuU7~YQdVgX7Ro8&T7c;f-L`K zd)o7m391Dj>JlG2J-{`;UQL}ZL216UMkiM)hf-ruRRL)={=^@k=C749E4~6nVX|UW zOgRBJ0tLZfnzgh#Sf}80V^bOS-b%bt-kw%G7=34)il>w|!Ta@{D0cp4ozGc1efoT% zg?UrJDw$fmeDi&bz%163akIIL%=(vBq#$@eLeQhML*kH;wYEL`bX754PZ?t=(k!Gc zeDUHUh3tp2V(-jqwOr21nN%92#-nwh`kKWW$c|C2f>Gq}9i9~q_Guv6sGsVW>mD9I zyXY02hBugwfNoCnS|1x;?9Ve{_llpUT85;sChMCegxED;t;K=Rh2W=ndiQLGvPCIHRYwIFW+F~_tE0px} zy^t|Z*_$gyxV3vv=wo`f&TpI2R}I@g?cVJAi>cpDtH$CF3S6<$N?bAxOW(Axg|)Dq z`x2vyDLMIPZiX9jpAo`s4lVI&T4Y!sy9y`iIiLGvS>i4LM1m9b#1PPR#A*PG z)Iyn%#Yr9ph@ehFY;4L$#z7sc7G{bvU01DL`QvBd%yR{^wE7j`&W3O`RMQ80g(Z?f z=F@DYZxauCziV*$emQ@_$43&glH$FfAp8Ivt{H9^b+0Mz&2G6n)WiDqS^OzU)ioh@5s&gznRQ?bx^3%Vl=s^*$5KXTZys1VpR2UL$ z$WmuC&<`t+Or6g#9&CsBPc#cDQ@a%`-HCeJf4NaHwCkmeQsga3$SpFJSIA}|v4(TZ zti--A)T{wdTRaH2z>%6MX-lF9EADPIn($av7qUil>E{9+f^WfKJ)c76cgM*N{^XP# zx(%Js8vu8W=O>)jI|hgVu}C8M@GTQyoo($V#AP5l$sGdmM*jFMK%h)h_M;;VuLDMP z6?LR~GsmXwtP(_#r3O5ku#6a<{xGVVjn5TdsE_uorJOjsoNl?Rtq9?HHq|u-Xm8Nn zR@gRgXGJ4AdRjC(wfhnka$jvBI+`&87?~Yuh_`|iv>e1qWs!Y*E}tdprha3-CHUKf zS}0(ACYd;TIoe{J2qfY$P0QFQ??nfwhnoFEh#=#r3FKY-6t_(b+++MrX{7KrJ*k6W z)1J%8uyVl5mVpc~!!q^gN#K5Fkq(ccuAtm|peBS% z#5RlvV2-B?7$$wiPFwi);t!&?c+!SYGskux66An)X4j1%stNIQwyk*TkyF;c1@{k? zZZG5uuG@=EaL}+T_geH&xN}OCS$whD@-~pdoQI5=P+~aQ z`Y~5?t0adir2K$#Im)kD)#gT|n|d$MdAP)TCuO&65;l@EsJa||Q1CRrdcq!PY^sbC zKyQcp!Tz-+Cn4^xMyyA*de%!4+FK|p@pB6x9^Sn2Z8A}oFQ-tP8_)qAnxAO7FOE9> zt#8<6A$sU736`TI^`{%ZQ2S!&aKdaICqHPU+UzG!BrQ3)<>)?=ZL`xEVtO5i^u?O` z#gNJaR^^P=u!j&%hNWPl;ZcKn6L8?+5eUm6kO>()dJqP7k!cr@JFj0v?txpHP40XnA*XN zQ~F&nM7T&#XkjHFeIcNe_RPbGUhpxA%%ixA8$M(FFr!w4clxfPM2eTDrk~k0uXKQa z9ucW~cleG6P!En{C*fMYMt%~|mSHeMWeqAyz*Kd!s);QevFDWiMG3J7kb(_eiGS-U zK$f3@XTXG!RxdOoR|wACf#gI5e0L_c^|>!rMA{ZF`~03~2jWY*g%Ro-0ku`%ZmPPg zqcxt@7mwJeZ>C1kAW$Dq5*%<>hM`SAGPVnR4I_4*zZgY$2#johdU@%Ju5>n3IJsyg zOZ2sHBnexZSdfdWf~jY_nk-3dIG1<0foW#rvc4&{06JDEXZ2!>Wqi2Ui^SWbccUa` ze3?eNGEuBL^G>BZK(P7Iha#OFnH{Egf{z}%s~`zNW#_DqIVE}X^@6Gg>tz`{|PS~ z|LepPBi{6PpIl=mpC@=9wt)`}QcwN*7u7|Llm!snYkJ}Bqgr5Y6{|k|Wq^fRd^I?pIF-61yzmw!Lco_*Q&b{Ky4Z?r z(ctb5P5OgXjP7SoxIwNmPb;XP*S@f#3CuSNjA5%|bF8hYv+0?pAD4Ks7V5IIdRHT+ zmJXoX+h6)=$7sSWB#C9Ps;zH@1rP`b3jHoG9Fij$*OH3@APcSA%Qvcd1p=gzjU68b zZ?yoI*ORc-o){xb-;L$1sr%fl=?>jmqd?yje@^F>4oqxsD7y`SmB1XqN?0DMUcd6$ zu~$Uh>1@$rH-+@wb4~dYAukGqrQxRf1ct{%?Dbn$@xSFT{_Vc?f8=_5O^h)!|InrV z+@}7oL;nDF{)1rmnwer|{?&K&uYlcagx>?ZAA$eDT>t2<|Jz{4{OA2!|BghG{#Vyq zAcDkioTeX-^UEavEZDuiQ(%6DyZ@d){xS4h>-KL*B$@v_4*x#b{RsBg2mA@_evsK; zMj4pd|DGFU{wa6*N1M&e^2kXpy zfTI!o>EDjOAs4=XU|+#6osW)FS1!VqAZ?yl$(PCOecsk=xx;T&PDk`drdZv&~3PFFsJ=FL*v@8tG$kp?#pPMr;?iy$E-Xs6NG1uR5t|Rs>jB12MPxCw+Sd< zJ1P@^u6>!CWgGRf!5yt#f*N=$uW378wOpxe+)@&!f+ve~=dh@NJ&^qQaYO_i)7yI1 zZpy(SUwneRy*3y69*G1=ehF>iT`9u*%^e-wl9sw$q&Qw1QYnh_62#FP4Sw&>@7-Xg z#o0dy3s?_mXkRoRC_3Xn=@-Mn#5!Z-+qSWxhQ76%`y+ZP?@~4(O^Bl1b6^s^qbNoL@GVd@*h1VWzfZ4L2#qYIcXu%hH%eMF;&5u?T20{1n}=LYdzS$GtKlDrY#Cwn;HX z6ezm3M=Yj}k98vvmecmo$HbVK+qp!g*g*^nV%y=ZNEI~}fy@yXByP-#ds2I9eN({dWpUlQ_)y6A5+4Y+@4;G| zn(!jYx#_BSvx;%iy`B!!Z*Q~}C?NGNcoCHwoZp#apV?ppH#x>pfh*(9XW5LEH-L3t zBRz39Q!I@)b87P%R4AGb-L+sMFr$TcZvq~wiO13i9TTR^l{ZVcF2fc-uXEM656uK) z^QJZ!TGtF3`xzj7sMcqY*utA&;9+=w=KPQJNG(Zh$*?oaWvnYGhULO zoi3)1`vAB-bT#UlRZQ-?^wo+eW1H-|=U`hH%e=e>86TX}NTc5F(a?*&VE9eQs-|VH4@~`X z*HK`T$*6jtFh-Lo`FSP-W53-U!d895T>O5oHIlGRD`FsQxS- zOTzB)zN2jbOu~r4OBHzfjU~t|omrA=U+wl|GNTC=&?}SM4v0YkFZ*fH=zDqGkTe&G3aOBITWTgNa7s;HVcv(p53NX0Ar{Nw0Ag}F&RM3 zsH~5Y=@_IXcFN>HELSlF*MOV*Tpb-7D@Hmuj~P5p5X;0HL*0&JTP=oPOWmwSMJIpj zFs!lYj>o6bsJzqJ92PMG8{eDDk#i`KXk9>s_kEh+b0iG8^OW^V*kmBmWJ#X0b^c(i zb-zJm&KS9L5fF4{O zMR)WZT_}vr+5<8+#^^QdH(EQWhmh@JUfc*ik;+x6e9=0BPH28+rbpd1(tfF{wTvvMoZ5TYPr7c#W)hqQMM&%=c1GL-hY z(@=DOMuKDu@}#`XPzrR^9&$#n!w%F~j@cRsh_zyTnK<5l~<#jE? z6J0IxRkVv#G#cuQ$16UD{nowrZ?EzHJVnv}=mq{urorHMT7&dSu+z^4_%w|niX_nJ|B#bg_83tKB~8+{91=hxqU^gjFX!N2VOU-A7l1pSY4 zimy3Hd3{?;dmCMS+t&yE^Yr@r9QGsF|KPA6AoYX8n3$M;%@(r!lz{xBV z&05u^m?ajZSF75(Tp4C3apzHv?Sfeyi<-4_-8Z@$P$2;%9ZkKcX`%avG;ng`5@_Zu zs(A7Ek}OQvSCytM<`cu?*VPxwX6fXV9U9kWDYwj8=aE(M$kL_cJ}xwUL~l^2@QEjKZdaIjmXc<7#=ey7{>OU7>(wQp1=yeo$< z5FG(Lq0Xu=P4d%*UXcI!(N$)U3j!2wUlw}g2X3je=68?FGTFp9p$}6}Q=n=$ zI5UJ+Z`p{4>ezjJ%YZwJP9QrLO9yMIa6f8Zz5_ozRLSW2f zb@effF~xq+$9$*+)7aal9LUzk^UK@!0#oDEp;6&H+Cnf|wno;K*T+1&;5;c01@E?* z48JRy4=ozZEO5Dj!qE-hfrTo*q@t=h2vs1v3KX5PZd{I6c0GV3y_B( z&GkS@*N4iT8}G=^^FN30M+u}yuY{a5jz0=4SZ0zqXt7YO+PmIZ1RxTrfkdffDU(OF z5f3i`KkSd24cevVDVA7R%}W&ywn{_^e=UT`k%-xjLN-mqYXDu26_Xg)2FNN&t=9Vu zmfDyhgx)B0ovc#^B{ubr?JgNCB~Ds#&_9KfSKv8+ZM^oi;LNGyL1it#l0+KL;hIk& zgmbZ13EIaz$qooTl?x?s`l=GCNyie(*nO=ah3eqTs#LYg=v1c^iYECn{Y0hIiWnzp z3*tD;{*eX={Y!bnxl_NgjkQTLr0_84c{;fPB2>a8IQA)0#8a~|gikMWw0zm?PR`Sf zYdV`PEfyqpy3@uYK!QXhgaZWH=~%inTtz!=+-NVxNIrx{SiPAn*JWeI^#CtJy3<6f zf=pi?LV_CEkv{9q)F(l7G2dkUT_oyAW8#Dn(pw9hdSgK<9;b2PaFl( zj5eHMdon4t91_d8YW_0%=(}I4wbvfCFsA(_syO9!gCW>p5W%+69MjJBI)umY$H@7W zhUST9osKboeK_5OHvT*QxA$ z#GNh)A0jL^+tvj@F0Q04Bo*P_83iDh=SZ&_#%JMO;D^ynMtnI(HG>F`E~EDf2M1sT z?uBpY7b{&$r{2p7J_b#VWY(z>HvsaBNFCUTl9wnUn(4lyBc^DlF%yXvRO=;v+fgP-w)X z=UvDRK8l*Lk>D7T#>yz`vHrlyNe5d;mJZ@r~U$H3Xa$cZ+pLd59 zQ5eB6;~p(Z#L+I(N7(2f3jq5pqSpQ z?PG%bCzfYu`-_hMN8DRRg_&$=+c?49oj`DRcPCho;O_43?ykYz-912XcXxMp$Vbxs z_31NjchBiFYv%l8?H_wRRTQkcYTtX`=`1cuphJ`M8%`3GM;Q8@dVJRWegL2{_~BTv zdna~)Qnc8OA8>r0o%r~dnmEIrR=BU-Pj8uY-f@vxK2~c{Bp)FTH+~As-zpS>N?^o9 zk+qTyrOG9l!6bqm`Ktbc7kI|nHYiq^mp}3B;qr0P?H$KEnG_pxm=Zq7`+S(*xvxTHQZ(-<|?=_24Y_UXnPBK$9(NKhyPx9}q%XQ&! zqHAg-4P%JRCg=+xIDyTWsZ_Hn4v6)8SE)>ysy~e@uGioKVJTa2Yqm8WvW=2+kD~M) zIv6~59&4R?pC%e{igf!0#DzcOoI7D^Tiuq|C0SM^srBkySu|=*p#)l~zlmHbFN56j z4&KSj!XROph%&iAG>a~6FY26MxD*`@NjbV_9pzg_uedhESl0ltW82v;4KA6sq-FHl zeq&7KZ(S0-7o;hK_<-+A4jQG`=X`pTcJ{` zymF9Maw9<6urFs}h9I;;P+V+62fg;&KW;|$+^u|B zYj%Ty8~i;Ij%S`EByii!l3CJLuc`H|{-M^KIy`u&t_yycaW83NEd0mvz5-&PAPByY z`)p~>ku|z0N>^+;7_{K7Nh?z`^;@$NzXZeA?&k5wg|Sto^?E+@@(dIJfQ zydScjqk@^QU_(2RSwD%qmo_$=^${}PP7o&8E?thP@mF`0F}a9wq|qVmlw_g{hdy(Y%>cj#ny6j#efmtLiH)tP6>N%r*s5TgNXqq(zu*o&8M7&InLlB{0Om2{XOa ztxjMBS;VYQO#;}!g>37Q779&x9~MZ!jXDHgE##Zp7F6p;YMKcd%}X$BoWdcU z{OQLgna#S|?ZX|MKTnKKMhn=TOpNd7xa$r%Pn|AFDcFSwS6~TH9m>GfbInvQH@LXu zgNpU%^oxtTh~V2LdY~omc5F6NKsA1>)f%4c>?fe}!#YWqwOI#t;cZ zv_>YiJ!JqF?du(b-_D_Y8$q5){2tH*KrXAOinCRZW6oES&swr)l}h$JitXJesoKp( z%4qdQXV>P`_u0&|@y})qS1xYS^(T=4MS;vuI7uNXt6TdhJ{w%%I%q|W#SxjPnir7X zp8H4*R9rATPbODw9+X_`1T(@PZYMUJ;*ggTu;+L@3>X=(2FeT_ta#w5er3 z!5#$pabvARTR-Lj!W~N7@L-h&(>w@bZ{pxqQ?});_vqy~s2kMER2<)&07fzLP=v8&1ntljiFLjAkAWkJ`XpR=w+|1x=I!`kOx>`?rp@-@3W} zg^Jb>rRy)eWI#pxlk9&g^nN&~|5bhT&5E3f`R5zJKau^<&>vCTe}?Se>};8se{EX( z_sIS&!=I7;&&2=DTK@{w{)Fs*K9~HT311n18zGbNhr(Cp-#yj*3#aLygs*gO)oJG6 z-c|mAvVZm>^4Imz|7Egg`TH*_e@^y4ll?zr{}auBMBf~jS%2FOmVb2!4a=`v`Hu=~ znaYw47RxVWUyQiXq>1a;wqRdxQsW?0#QE6+z>pWQq-h*j=B@qGwV5v0=pvuCJWe+_ z)tIqaI)fYL{%Az$;xxxb3NEd%)Tv>3P;R~$=}aDf_x1z4x9E^OvKT2Cl>tkJ#hVku zJu8h3#-_ldO$&2eJ3pe7tbaFB>1A^#(gG;gqSj2#31mwv{P| z2t}*qIBQwX3}TIZ412ndcdt&y(kd6(rbfy>(=eJf`=dTC$-(8%q%-sbz}&(Omm*?vYRa`pS;zWSEz#@Ty=N8XIW15qlfarm+lb= zxBq^%(e@d(Z7LtB*cYYEE`K(g&i2vXq#2VNg6T{0Xk+;Ouz`n5BZOYu2Yq;mW4f@b zcN_`Mih8mhjF7Sz{q&MPl!(qgtb6*3$Q9jISbG~Q-QVY*2b_BO2x5aYdCU*4P7+-w z;IMn7#%Av55XJ_Z4cAD+4Rz2wAGz2m$OM9H?PLa>njwm4?$r*-q4r-&m>5os&-5;BDf= zaA;73jk1|fd7I%f--Dj4)^$fqrO_=?w(BRAY1NESeEa#UCWsu!Ch;c72mO=cJQDaZ znD>n>z#4tmq8FWwEQ<5(Gmb1zPAhr*Jiuv{@300pJe`{5(rBr8p%Z1PYdnmMM~<>m z468|F8DWJK4VKxySpAdN4GYV@Vt1*z!ju|bL3W&T60#Uj1O@TnIlB0o z{6tfj&hEF^0C-}u*QZi&r&jR!I%_EHk1Zz1-cd3JvVO7Zd}w0G-nJT9`#Ug@Fnm(p zNGb6t%m7+t@c+qYV0x1IXB^lXlf3aa)5YW_gR_h zEMj)Fygd#*%>9y@;$FpceGmXrMHyIa8VouB$MIcsn|vTEr@LCiH6y_}4mxn= zXcR$^w4gg{vm4dL2qMyTD3da6%pP4R6{gWn2>+h?>nji}&*nWn!iluBtPhxFV1xHu z91Yde_cF9~4ug#EWTz??3S41hOw4HD3&@y7q3|NXN$igctkOP%gYB)KQlTM;R8dV2 z(URcXEtWlPK&;6_;b+@$BX@gf-Um)dJ#XT1PB@SmE_0h4p3c_nRCKsqJGC|wcFP-R z<7VUK8jNYi)o0_Btu zu3=nh=ph(F65*93r;2ZFDtFJK(?TJ+c=#U6=_1oD!Bqo;2EoNz;F$bKP}F`W(_Q=M zeip}GDp0cibMlmM@BB=wT#bVJfw&YyJX5SiEJ8&)^kV zmwXkuJ`IbWE7UG`SGL6y(9#T14ci%_IGssF=5c)5U~k`IL)Lcm>{;f6n z@iv<1>qqNhgdzRV$_>7o4wOeeS3_6&;SN&R?0eXFM8kMQ$Cr&w!Z--yZl@lHN{VQ> zq^?;%|Cjcz5RmJ@Om)i4HigOST$N{IVR85eV!%DB9MglHav?dL5>UCk7B)x8{BMb? z((ZjE{y_#iTtZJpE=To@{sEwGCI-bT5 z*Rv+r@?5SBam5MZXha{MC9Pxo00(-+8&4WrMeu6zqXd)$WCMnz!2ueUSEj2jvg#HI>j_8InTOshuDw#5_o5pqlWh zcJ83Z)lI-D_B zNCh}ffcM4m1cQz2#oDMt!{GRXvS@tKgq?x1AE||z1iwWDf*4YA=}69}JjSM@HsvDF zh>y8G23XUH5{bW-64=Sm@NoLx<_`5e3TtMr;ad9BY?wnll38lT)!`Qo!p$R{t48@- zM`IW2ez-e>L^B8A^84V0mnoA2jR#_lfh}8TeQt<+h#_$|x5F61fRW1wD`Yo1!V;AI zE}K*zKclKgOrPvTmE6fGVq4jy&?WcDgz_*XADMhwmUE9@gn@i$?ptHuN&7qjMk3TF zWx1jF`h?7?LukE;t&DEGAcc(i_TCcI+YYquR8xvu^~+qSIarzVH$Mo~e;$r`U*Df~B-K=Sr4calHE)?7*`a$X%`#9gSqWWyxfbw`q&^UH! zzrC=Iydpt;h=HL7&;qJ(LyG_~O{kDt-#h*X2_(!VN|?QwQ+&|X6Lmy4$KGrD$B!S^ zu5En>yjZ{F695HoZW(biYcpEwYSoT6@$E?$aE4=s-JBiR32nO z6w)?Pw#3jU=nn`ih;eAv`^O6r3m45H(630s`Ia}JRRJ4+=xOF_V!{S2G+`gE*3djz z@8I!&?A{>p{Q9ZlK4@8KJ9x=EF;~U&W57b`sp=Xf4hvV~b|Mqb2;h;%=Z|UAvJyL~ zVTnjgRui12o}1N1d7zA*zQCS6S$3|}==k1hz{I73)M&eW7_ z7SXCo5FNzkMNadeQJlUdt+^@n$D!m&3rUxzl2!+4nmraRhxS%38(4S^3uqz!hO*l5 zScfDqk33ngfz>rfN|%xrbHFJBlSfBDRk~a8TY;O7@Ap`OJl3q;{=v&fh1T@|I4IgF zxBx%KI1)TP`8AHB*s3ZRQIlnIQ?{YHjoBrT!-0FX=fd35r(Z2m> zIr;CuPy7WZ|KoMxcS_`+p!;Lt^sLPP7}m=AuTCjq{dFt6Rk1Gx?CHUwbJ0YzAj*CP0ouGpv6 zGKH^PC@39z4B8x)Zgq!XO#SwiDH>(_TJnug9o4Z`gAL7b&)adA0_>mp#)Cl`5T7dQ z_?BC7iTv@y^nD!`zRnhFZ;H@_9s4cG!*B-(m(O**V@l=T=j;D$Di5QGQ;DvVt2U6+ z7~LXV%WM_;a^iphM)$>2+FxpypM(d6ou&l8B!=+IA@-8Y}Oj^`S_~*rxhRqQ39fKA(Y&H~V|C!z(7hWLMcEmEHc`X2R zc4Gl;^Co><4)w^^t*4ABf*H1FD**-LC4uYgQXmO#z{JVfoH~$dRs8DJO|dNILZHnw zeyb%5c*5xQK4j-Hi1(c<#vo*Ln5*=}U~oBg6F91&hRoIj^spv}GqKZKf=RM7UC{Fg z0ALD1`2GG%*zpb*X5OzakOgpRw%-(e-0}FRwxKgs>ax3wj=WyV=Y>gE#6-tZ4H8!( z&i0A5q_{U2AZJgY7|*_zf~5wk4mlP`AM?IuUf0-QqsM=DhQCoHIB}YTwh3RHuqJK* z`tGPl*IOl4K@97UpJpK$=1-6BrqffyJNxEd3m5NhQ>7D54`)ycev7pRL1B$NhS4|E zJo`RLOptrHWo?GHep=4Qx2+TJ{uN#G=@W;!0AvN}LU?6eG=-j3@!AO*_KdAlMHa;i z5s(86054VpJknkGQlilXI18L)rh|FsDeQb&o`UH0Xm)PU*RPrVL!8u7m-JLq_>ei2 zk;-5?Qarc_SA+whju`3wtQJ2!O=iXKYDekAeL!S|C#;?QaERc8YUinB)*6q?b1iOq z0lNobws8Qk)BR%(m3%oQF?Ialwiew2E& zx?Hd!nF1JuDHcnK9-yPBCPNEMkyL8|x%IpWe-?>aiiJZxJx4UuNe<|1Qp4$E)he{i zWOUIsgOz#Mm{(>7j5bZz5H;%9Z$i?@S?sc@VYQ=sz{zatkU8K!eJH98)n^Cbl!5E< z_LbJj#9Lm3{n~Qh;I634s{stDbCUAJTzaL{w>Mg9CAv8Khlolh8&I3!KHGG^KFpn+ zA|Jr0PK~X#BKq3+Fa~p2^jgcv6JTCjYh%heKn=UMT$gDb;)Q2A1UgLJcQqaY(Or!A zMZ=Mh4Q%2UXEP2fW1{e3Dmg2s!xvn>Q0^$m_z5KmT9(;Kr{QSj``k>C{!?AOX|%l*CSQPpCppfzQo|mj z!Bs*Omx6oiRCodwN^A{;BfpfLxr|ys&VO8Zclv}rVpFbAN(T8-0}MX2+t^6MmHQIh zkkc3Z0sEsNJB-s>Xn)j*Ofz8vd6?$<&NF!a;NXUSPmKlomGFdgGV6dAwmbK6t_^0( z6f`1IqJESc>JlXWndV{WI_N$7klx9re;N=BBu`&c5f$K zw!B(mF6`Rl>24{A>ep@N<^Jy?l>4NzrD4G$EcT{7DHd_nWN+%<$Brz8GkZya zgP8lH$-QnzbrXjHB-KnLz3jxVtX>&*T9*FB>yLeH9*G z&W)}QR2+vQwvQG19rDp?*4d4%V-lwcRC1u5?-9cu7gKCSG!o9wOmUrLgBZOmHf*<^nodp=7Ey5X?KX8C<+Mnuvu-P(i^LF7MMmG{=U9y%)|@KBp3R9H zd*U^=F1XHpkx228r7X}|qvqCJTh{Vwyjt?e96{CQl|#<6mPw$4k*SzeYvp0iGb)wc zq$XPXYz$0lxX5 zGS2ut^J%kcMw8U!7B3fwI`kRGHak7zb^T)4q=m7PXU#1T=snjp?Bam4gHop=_d6Ce ziY8fCyL+IvOO4A+a8T%7q-{F8wb6qvFU;blE`^qSEV-*q4nu9ViLxC>PagWF=mV4^ z5z)mBB4uquVY{6SYD8}F&tZPaQ|&ry@fuC{&+cfEXu!kE*N+SSr?)wI-yPxZH~1P$ z4!V7%=RS&rxi-;a1O}HNq{zkhX*;N;8lHZ~08QXy_V~a`Ik)`bRqYuq6Fkh|4^WQv zhvxBLzkL6DB~RwJdnFU=PyOdVq1?~VA6?V`BPCDfA69&CmW{u{#J|xJ`Xj@iq1?~J z|IJ$eKFYEF{pQC10m{`d{x<8U^AE+$KR@SRKTW?cc`~!T6`bFUB>yYQ{S5ulzWqm0 zj`i;&^q-^L&t$(W@pmZqlf`~S>6z&M(Ny|2Y3jGX*S|V0Ol-eyQC=^ z1|+)8!263NgOuds`s9<@DQ>+dfauJi6e4th+U%DbS8CqLn2$^54)kJX!2r~wTzAX1 z5B68_ZBJ+oiu>vD0c}{5W@C~IHI8n55e7B79Q8D}l+yh9pd~CFW**v_7uE@ed_iyr zL*@yBd?|U?J!hqBeqpcsTQy(C!8u&k1D8HFLfxJ!-3g?CTBjRmf2}sd6TF}ngK}6h zU!7Sp3N@TJZ+o!$RJ>~9j9w;l~soMCd>d3%5ufSs*kwFCXL$Q-tyL3v!SuwAm~~U z?aZh=06p#b&uylh^*W=$CC z%9jX-6G09+VO7#tGQ$O(A$dXc-)W}`0TA_1a)wtcM8&A1dp&5U*Dyx<(n2~4U83iG zVO?o!db!eXjIHdPu&$EOTs(l|EjslUvxk7aUG1dlbi>S>cvI2JJ{$ATUi$e zUpRw;^n0xFmF>TX>E9pfD1&&nDbe>pAYn|L2@37r53SxZmkt%SkURSuP9-sqpL zczPif?UMZxmq12%jG%WwV5b%!iVPQEq>eZlGSSL+;gd5ZFVhHAj}Y!DLLIj#8Zr~*gg z9w%-N`~&QBh5j6GMFEJQfsa5bhHknGqSiD9hWoRf4D3vS+#@<>shA7}t$4gI-!k1Y zc!C-F6=U+A;)w6*Laz3ULV?w9D<+9YuK$ z-Ltg^-@m?_6F#41CZ4E7U5;>JK9@qNsvDXSM5vgtMOr!2>9`SDi0yD5$|cT5pMjwN z0B@!rmiwu7S=xmoPi{mb^dEc~RLkw=RMVqli(vDfgT#RQiaFAyPl zgcBdF{m^4_&K^P|KIV2)N@7oN$%{(Fxl38|mdvhs9*qieX1!sXih~IeMUadhhJ3on zx~PBG7swYmkL4~jo#WfxCOo)oJJi;;b<^jI5tl4J#9{;mV?B~Agdsm6vTlt6;kv1a z`{YOLVJbQcleT3?26_45+OCOJxkyZc_bxLH*>Ztd@fP3&H}SmZfg4d&wwXJ-Z~bph zUhzrC)kM2o$#8Sh?bw(D-r26Ve%Fk+K6T;6ue(<0x~#2oclxZY53F)*boXP5MJ-AX zSCxi--n1WUbj+#0w?#{XwC;#t-eNMkpGkSF8)~nU6g1lli|mggfZY0S-z<7G7G3xg zoKF^MMILSK#P7$%&q=yb_!s7D1fZ<+AW;@PAe%&Xj;fOT-{^86UcYwDJ{BG5Oyd2{?h7oR1NwmP3&-Mq(xs4SgIfOnnB5I8Z-M zgM!=8p?^=Mo@b(v{TB*`t^C z3bbT>AI9>W(k&$kbG0 zR(m|H1HSd)=5-MS>#8o0JU#+MV7Bi;`9|>T><)4pz!?pis!HN{Z$I@#Zq}TyJo|Dy zO!eJ;?@f_4UBk_|n47y-H>KN&;>WZm!$egu>F15{B!8}ne3M({tkM6Qp=AhJv|%CkH6d>KQpoYpsioGid>mho(m>&oV;tEx%Bqf? z3@3ma=m8Hm>-%L^nE=ylT{$GgTx}ehLZ2j5D#5Hu_g;l5&qYx`QY58sfYL3>z2t!T zXf}qadg<^MVoXJvi}}iV(^)pI(mB{)sXCf!)uL>{!U2)$Ok%Z5+87e=`I-Nc$ z#DHx&`1J-O9GrF>&uE9nm?65;^3Ww^Ncrc-=X1~|vgCO)@|}9`AdFylW}F&~CPxz4 zlq=HySjx8MSlXthSW;;xTv;KF8b+Lx=iD!qNyOR!p-RB%t9A^x6MV)At^?n{jtDq@ zQ0{*-d$`;WOI-L!Qxl{+6IHxfoyCy`a;fiacxCr(V7}!b_v{-d^cLrEx$6D=YNcvPG;6sGqoMLj$4+JR4d7aIg z)PZFuKz0^V>QTD@T`F3r7<)m_^4a1RHJCz>GvEUoW?0Pw=->1L-PaHeO(K`6h<*@m z#uouBaGJ489hqyZI+9wbq0fW{pLt)G8fbjflJiBk90R-XTLe>zi5`VRfVTMzrM;?Z z;yps|3*c+`#HvR^Z^go$CLZ_jurR7gaA`PQ;;!L5i`085Cf`L*1;&hk!bMR6v=tfV zYSkk|vHVWB5>=^Vzhb-4q|5{rs@ahHW)ZOahvCt$Q*~fxD9zU|(EJq{!3dKuwW>^!BMZAJx6wQivB}_TU*_vhGv-XFY`X&rq844m-rWcga4Dv>n z*dGbLRx`!Ss>oL}uTyb?o77(d_~;2_$&S3ARaTBDiH4lv#q-Ml$aFiuwSBA+UMk9L zW@jp~a)U54oZq0MQxuBph~+>eS(tR;#JA{k(4Hg0!|~+!&YLn2zETSloM4RejFRTt ztNG)#+&jRo9n*xsO>#*+d@rSjim%dcUNz1TB2 zy$D%dpB`VJWVo)$iWdPc%F<_ML0e`$eev;8G)aMVYg9Oup)KqQ!n0-_eAepd+pW`F z>UBqHq^_3M3bS=z6*KhDH&B34DiS>c*)997A$Z}`jQZ9o}iWr3g~Xw zZ6+rFHFxMUl>`>exJSVf&Jtm{#c9?03eZA2LGruyWQoJ3YtA_FQcUwsFcO3gX-)5$ z=Vg<7(qZQm*^-Q4M>kchU2bVE{$7d%QBA;2{h@6~ke1pqP}FKQ5h3dITna~7!a0wYoD` zXaa>2??{Pyg%y{K_5xu`3gz-`Q0CXnKIJ<>m~qA>v7?xmX5w%DPv}x(d=f=v)Qg|- zMgh8y{15O(qysHspCw z>ZbZw2SP~ojec(*iV8B{eHdD7#sfH>3gQBd|KU{HfGdl5rFtJ}`K$rVAo51e6K zAteux*;O@dYE(`CanrPz(3kAwRE~f_@#2iHw?4h%syO@3UDdS?g-|53k2Qp zEK6)GZx66HYQcmh4>=Qcwnft)M^ei=YH@u-__Tn{>CqnZwe$1d!*b45*mDT6xUcfj zN$2(@2mbu1xgznIVGO^bmQB$78qdfYocbe3YG&hNvN=eaR{>u{))gbqnhRrxKq_b( zA%`g}Rk*b1_pttaekA6S^y$he*rXDCH6cF2w8w#KW*Wf{^(wi^Dv{$Ej&m>C5fE}} zF3N7V@74`R$*8faKiMf9lv$%2dHZ#>f1Ta}*a3=Y>^#~Pb^M&=Ghz6SOFCvaatp8p zA5DFG4TyqnJpw;e93^iRy7E-&y#}`Maxkv47a&6}6}ar^iI|+#nlZ?ROgea6RA>UF zs!4dSFhRDPd4B-lF6BP#IUtGeeJyieaLm2`s4-oP0;W&5a<}8b(syN#S)84NA4wUnut=6fV zl*+{9^ZZQNZC32+@?Gd0_ANzn=&Yuo;0TE5CeMO?)r;$3~0GpHnwBjBMC>vqLtF!C z_@pE9KB(gC8)JiJ^_9;kxR@tMXhcUKSQZ}!O5w@{vVX#2q2Dy8Bz${@Q<0~znRgt= z63Xehyqs9GYtBtuZz3vg;092L=|qdoX+&g3YJy=}F)>Jg0IdAZ=TSV8@4!No!0-M5 z+Sq<*lYYGf{tMUae-ql?Mu;%6{nTmwGqnBg+5OK((fr?u4w-*g>%GY%{|A+-w+w#@ zZNH=AKU?cx-NOG6+TOgwe|+x$RdXIg&2P{)|407rzkJTWewzLX+TI%TnBPR3{{`B9 z3H_S#pZ4uP32n@@e;=0r6xx1C_RA9g0d2of*N-SYBP-kg4{gk}KWOpS?f9$X!u-|` z`*SP*0c~&Ah!Jn1L-#wS^C~POhGxxMDrHG8WunmJTtl?^Ts))TdbVg%k+{VRx0h5x z;-wX;C1ny(pZRKeq7tdJV)jS&r$LRE56x2bDJTJLi$qmpl35EB8&&bi&cnLaxHXiQ z$`lJWw@=58w zk#Fg+@-9*-&27^C9$Or9EikY;=wQ$BC6J{}JHrC+^gGkjj>ql=euNE;)_fpdKA^+5 zA9!SEQxKha!3b+h!&Wo%g%;ZK?0fC5bUGU(A}*0L_TwHU7j4yoym&&XQSH}_7B=Pd zK!m#~&T9m)43hM{JcJgSja(7go0yp{H@Z^O&qWCm3*swuc`WXI%-IyPlUBF5u!5f< zpXEsA>EvU-ghz+q#HdLJ6bh*+1!czs@rTbDgu5o9D+*Es>UwlNe zgCCjQ3{MH&JE4%~|U-4#!)~3YspRVFMAtA(}Yz_?PhWL6;xIY|}}UAL&1M9BM+F!xwM{lVy{f zxHUUB0tajysW=?vGxo>gI9OuN9w9n20il%xY?%p8d|kTLvgO(GSl`jO<)oA#gJ<^3 zRD}|c7a7M;N7yDqmpie>HM(R?c9d$%5hRBnWm@+ES%MWSR$fs`Ta+i^pBzNw0CTT= zxdakx{0dGBV0Z_UqYCL5_x>EvsRzuIOnf)Sh7EAln4T5?5%w5bg8=n4l2S;a_-O0Q zmjxISTf!%`&l+`aLeQcw5fDBAn`wBgnjrxI&>#hp&u8ij%x2lE4Vh1kHU)z8dPfH4Zhw2Es1HyA`DkKP^2GIs=RR38y zOS{q#yd_diN6En`LE4LseI0|`f^z*^7A+ZHJSeujD(+4C)E*Y?E&>CL9^3X7&WTlv z;_cOw8A z$~NM&3dvy3@KOB$nlK)0`fawJ<>jcBB}D`exV};-=|*yy^wyk3zK!(x`Uy04O)z}g z5pKiEf{8Gsc%R}P2EcuE_9-@^f2HCeDZpHzK{#v#jMw?lujCWUoppw}v(DI7>@ZHi zz_%Wlwp3T`)9 zH5T>tos>j5dcvOI_>$dvya30ujfNIVUlnB-nh{uCVTsbe(-r!WW@x7dp6i43IVrCL zGR?&~%LApSvDmU;KJeV3>!9}gVw*oQnl~^SX7k2 zh|>z}9yF_vivR6)i&Huy@EoozYA3^zyWi0>X{;eEFx|sU;tf#wE_8wAgZMHI1ImmT9ttQ)D z=O$cOsSI|+=r+%o27J32YkD5mvE2JaR!FKW@U<|lAnzvQR3A>Qe)zq zxhGIfVL_VYv^#V(GF7odc=?>ypEhjavk(<#wBQpI#F*Tgayl*Z?z17N<|57cl5A87 z;NLU4tapmBhqs;?oxM+g)QA8)uO(huM^ZIza^Gz7&jpV6a+@(Lq_OOk$ImZW7+(UE ziALM7?d`e6^xwBdY+&6@Np&cKahs_!K!U^;^ww8DHwCqU0(=xJt$?Q8k zbIsro6?R6eK{d{(%GeegqaN%HF!Zfmy6*dNFHMa)Jn=FVzZ|cWYIZ=^$bPvr4(n_3 zv~UxxLExm&OzdvGG~gIqTc>9cNEU;&z)4nj?dZeJ3ysn6BP0%VlAg9cUXD>Ju3(fU zqdY|nm(euup?goQgdW4bbVlmBwLfW`&~nX3p&-KliQ4$ZM!dp1$0-gXpGL%|9olm$G>t=gs@T_s1&1~q7^~i^PI%QA z$+@PQJy`bFZh*U0U#wY8cj7RznurO6Royn8hOGkfoQcWSW@}5>_M7M{IRJx?e#+*~ zT!rQbMj;O=SxpQ)GYSNjk-{yArJ6X>eutF@W)m21AWA{iXAa$s?(*6>?~IXCZJ%EK zs!HBzvu{$EL5`f)N$dSdrIGv2uU1FCjC?{Wede1;rA~N+dZaC8_|o72YxNENw4nrKbipEq*z`Wdu8|sk}VTO$O2E$`w@V6gOUiYh$c=sz+ z_^o;ovvM36?p!B4jB%qUfFz&N<@{bg)uL3ls$|JI9J%Di0h4m_T;`ww-=wTgWzOE* zs3Bf5b13?Jzw0`ShISI&}zL_*I)Be_T_3u;+-!l9OeEyR7zgg>F(Zrv? z=TD;lKZ8%E-zLMf{*g?Qnf7#v94uYTFgbiZ!IKhQDnvwG}O zBf^h}Vv8;CAU`5KGA}4EilemxaD3vEkjquJo$8_N zP7FzD>9MeM1L@-x>18%;c1(hmCP6I5CQ03te{jaFbU13^B3w;YBw|;#y z*&&FWxiotU>B9O7lfjzFDa2u~Nh1pHukrJN*H*@2SjnG7zGqa2R0}B+ODHM`#Ti&@ zP%pe#+*)u+@3D5D&ea5dNuj7mEN#fvmSVUVUP|Zqx}DNgrp2)mlrXc1F57b339S^F z&|c*eOyf3J*R|>1fVSztJZ6D>R2whF4b`jA9XD_FX!Z2)ir09^89c}uQIfjCQuF2F zW&ifIqvq_)&^oUA;tXB}vrBBuSM16Nqd++w?rTqh7PnE&k*d>l$N|$95K5L~Ja`gF{9Jnzq=Cj>d zZ&L%R7Niml;ptDtF(nHp50sY%;-ZJU@m&?7eh-`wX_(pRG`4d_7=CwG|4roCYFKQ; zPLG)idspYZC~B0lJ$!qUVC}l0-I$MbqbR$~O?^fS;@GDs*WhCNd^hKVd3-oLMRZ;B{~1OakU#2DGLv&;3~#OYi=0*4ToTkXe=oYS`ov9QzSo1 z_In}PkoiNV1V3F0P?K+6Uy_b)ut|rTtZ4+p1_S1Ckp%$Uw`(Ris8OtXvE)@MnZl2H z_qm+q2@78$ctT=_eT^5DcF}8thLsxoIomYCh0Sj0pi!VP4!PdM z`Q_ckGh$VSP?>_QW5jKh#!mSZ7>gC4#*}IxXKCd4Mt_u_$Qa&)jJ72BnrOT1q8y6& z-6U(BtNxqApSRYcs<|K))jA{_b2d!el?^-5qLP6T&eZ5!^XUnGzYK4ba(Oh^=5bPW zP+OoWBSaiJENnPoYE+gz(0<(+OaG!HfyLPa&GNOH^{tiSl~BY2fRHAUa(p&pPO(~? z@r`iPt~F}b*DZgg8JAf7i!b^etgu8%C!EV}?Ygd-X(F#!agwA0xp)|MxJcRB6b&jY z?bV=;OZc{Fh%ZtUY(QGZ>zz;gTl=)aiz0Bq`k87(Le{dNDdNJ9-g<8X{+gCfS$mp9 z_>og)Al*~h{pKE~l$&b>dT~23(M|~>nY=PR3?+4)Sz5g_ zUs_VJ8=YhJ+8Vh!;q2})q0B|&+Gibeqqw`YL7+;o4z^W}7E+Gy84A?^yd}1*rqPTu zVzNYklmr#X-^IvZXr!FFwofUX1QU(UWE!B_p3}J&Nx~%WH-guA9T@WF!rQarJ$_dN ztAaRRR9m|_4{nqrhN^&o-PZ6P>YdKuNgNpWO%nS!Dr2B%HcIB%yM`HE6Tiz(_L&)YO2+HkIM3Ps!4(lWNF0(X zl-U6NM=-l>G)D}?sY|f@tqC`>jgiU^?^PvJ^kpGU^g<Bkz7NEz=IZZpgTMY|%-wgcz+oEzhLoR`#qERrz(q;go@E$n_05>Ng{Habe3j$_v8 zE5S5Ner{#ni8r}?HsUe2*uH3D|DGiS33WF402+EhxhmoQi?rgr_>NSJ%x4jwHa0)+ zUTN0)c#o*g(W(759I(nNk>M)vO@1-8HmQ-I@&=$D(8Gs#IrELj#aqo3@Vot@pCh1rewf1wCP zcXfAmRo%MPdFq_ZM6K9b{49lHm3M>9rmOa%&m?I_`zT18 zuz^7sP)wEk$q>gR@TN;Zd-s%)x35`xsY;QU;ndAM-kaIr0oml>8cxCT67DPivuJtxp3TW@L^RXJAkiiO+`=IUZJ;xJ1Z`8;i z&|GY9SrdQref|UA@V}(FnBORjUvl*S4bu48^l!ZHzjqmB^c?KWT;BXeO$NH3`6}ML z#$O)La}%j=n0IsBz&y1br)gOQ!pU!MNn<9w?;@Hf8hTa`FA#@|AM{u`w6 zX5pV9jh}}9%~=1H^ZCa}gYhSP`+tTs7=O#Hk@H8|M7E#Z`E`*0!D{;3*=K#5{oioj zU!C*c@vwh_=KA|(^ZV=0#`x#)*}p&pkpLchA{Z(BhV&YuSB`;~OiTP6(}!hLlGiP>JXZ z?cK`k4;nrZUqtcgpg}xvq>Zj8j6;|;D0zfkxEl@hq{)1uy1QR^ROR57$P>XC9BENQ zul7=`cAq_DL8~0P6Xdd;ooBBV{#LHoQ0c61QBGl5haBkC!b$an6e{O_pa-1Y6M-Oh~SZFSHai~Qrg&s;S#Y(6&%%#9O9xe!|oLGry z^}+%-Z|A!Pw$+^CFWzv;b?n^zDNsoYd)Gvw843~0>Hso zM5hx(%rAb_KqCc&gF~P80T`bmfc4$1vS@V<$sNkE_eL~#z#2}qiluy)A8;4kzAXy| zPoumaMt#PzCdjJWWNV3(4as(lmqzrqDOIaC?M;!d%M9KL+7*X}(S?3mV|_S%{nG)e z?hr%N#`x@GAhxeqhxXDi8sMGT9sL-!FFd#QK>d?y*Z$iwH`HA0ycww3Y9H$ovq@1y?kd zGl>NIXte^g0Khb@rh1;%1a=)xdGla1>97_UUM&*yF_x7;#CZjCh-ZR_SUd@Ql~_}7 zIaKyDc6aSIZQ|TjLINY98&;?e2;q5^;V7F!=oH5%o4z6zQ!MU^C?4lR8!r>7s2X2P4FU@R{RMA?8G=bxETtU~!{isuG#?3T3Pdd{%BJ~V#7u*{Z2P|9A zd>BrSjyQ)VqA<1dXssa)%c@R0vq|L)ce2>BswV8gz`w2B3-}26Zk0Iwr7eFXNhBZ!>}I{vLoOHEC>cW5ZKS zQ{thzV@DD~zz@u>_;g6i`BKfN<*;RHadZ{$&e=Xr6O!zy0^2M4_*h0AWY`xy;Q)cs zF_RcW@&Pj{8A7s{V$V^8Kn(Tia;46>n|>86Mtc(&xL+;f7HbhiZIL6lY`XG&t*MwF zx0Yv03$7XQMb_oHh)OV$qysUGjm--lP1%qr1GHBufclM5G@DP*l<}VIq$XevkRIgg zPPrZAr(QG{nJ<88M}bxZDOn>py5u#%H?1cR|-wO3t^Hy{bI4gcu9>49~g(_5@*UWuQGPpP3x2oA-C?7 z$2oMi22Qhb`x41uQCf!%X9{Q3Z#XMDe#5n`6h?=>Hmc0cW`d39@WS{k&u$cOqAI*= zJSwcbhi6}itlG{!m4GbGT-X~k06f+~*;BffTC4YE*+wTI)HVkExr2gTMyeK_6?EVk z40`d3AzjPCm~98(^b4vSFu6e}tAJ0)C-RvDPziY)qb2>JUSy`pd0+Mhi=3rg{AtrA z)wC@a`b*%ZX5^Xk$ypvW1qb7u9v-U4bu5_lJ#~hAv{K|sJgo@z&q40nydHP^%3(xt zRg*KNx9o6_Glr;$p~htk{;yA`TxgFYn-xczK~E&{&hrI98(wgWYZo7D7GThLvMXa#`NjpE7!gv=# z2Cc&80tEYMUC3oiv6_{jTTtqtY+R?6>bUaAd7Aj z0??R({3+0`%xa7{nhsqE+4?*4VlJWUgFz0#3*zUqkux{EY_uRq^R7BpeJXLv;a&MbfM2Pgkb|4$Mmytx-_Oh1)6 zf8&q;Z2C7o>fbw7|3CWUZ}rC5-s;l*53;j23;)a?|7rN&jP=j^<4k`(X61i~6=G)l zEkQ!SALY7Z``MlU(jRAj<6YjW<^8?v>}S&-&D;NQ^3U|=QOrO0$A238rGvlo$A3E1 zKU$d>7}$OT7jJr&|I8nMD>DCgqOn5t6}umzvA0@nMrB`Vp|@Tx_m5qe>sWWvNGdbI zx#1BYCDh_Ugkqdnbh?TG72}MC)-P1H!+Jv^J2Ut^u^#xI#`&J<9<`ldsbQWb$-G9T zr>fz54J??iQU(NoRi4xV|!N{e`9`}tC+y{ z4gC?`Q%eBj8WTh#$pa^aC$}>cG04pJavUjhMhbX09WqA)rl(VhE1Yadw*Y~V? zW$xHYJGAgL0Z+QYsFu=!Qi{yGzV>KIC_kG@y^_?V3416kvV>}>oOFh|F)8v`xQY&K zt6jZLL$wxz%4|^-V?HdZQv-9}V&3}e!12mS*XDWeUg(|zN)u^RmgRmBP1?tq@H|ye z?KP&UgwZb=EHnOs&Y3iB^CS4YZEXbp3r!Sy2x2YhIYDnhJ%GJ?W^tqflBuq-aSUHv zv%tNQYFNfyP`*O);q4mYfo%<(c$Ys=4${+#P)2vKT=_51cS*)dN z0N?V1Ts^tc4SCyC+r5R4g(^_{ORg>Qu+LsjPAH_O$C9jefc|S}f$SaSi0ca*UJo~q zM=t}U$fMWqeD(qSI)(Jnr1W&%_JZG%=jSdiUIDfM(J+6rx*Jf6oe-(AupdflrMv(a zh#Y6|nNz#%3Zmi{atDN*wd+>ZA9(^&no3WBYWFEVy{QX2^`UggXMzcwl7nc97yn#` zCJc`C;Q5YyMh8KQ6dy2v)e|}yQAD&F_*35 z8PAk}hG+W=g$1dpp;ATZ+w+KMgxW-bH4ys-t!uX}zD&^q47U^U65_+^kG{#!uvn&G zF(_ria+T!>*8o`T6%+-ar+6vObgjW(oNnuSb^x&Z~_N%r>%+ z`C6Vczz68Cu=2gFD;65JsBQUP@80nv;?&$G7kVR0BYZ+lIS5305*S!-9}BSX;YB^G zIwdL+Lv7fp?BKJ4rLVs4!R&3RiubALT<08x4J+!DfW1Q$8JYg)x|{hTwUs_>9&D(i z62jpG6G1+TWjZ~nmONdN6%|bj-FLdtVSJ2@F+a|J~N`@egna5vsq2K;0W#g+RT9V9mMPW%+##Ek#ICo z$aj>s={5(r?M?t;vf4RVjnP0O90G>$a8*pf z_p_$Ma4$+>>vFc&P!?NhcQYFNbi<&kd!xNKZ)XJEw24a##Wle5Bm-coM3L+5~8%JJH6hG{KkZL zi}_^(_B?Xa(wt^8k&>;eX4$dho$_ZJs~CqmHe}asxv^jM^`Dr{J;cnDowA1Ud#90~ z2ERQvMR#>}Z4kC}5JdDwh%83+GK}{r#c{i~Z81t3NP3@#k=8$x78)0F0e5QNCJ=5F zR8zr`)Wk1ykF}M|`r~l(?{f6P*Amm9wOUPq?mjsZ&GVW9XZI3pyG4s$3pi!r$JcXM zn9hczT+nX^I$%J35v;>1k;v>O;!EM^mZd`9qd=+=`9@gkQjnW1W$W|V+=u6JrL8Vr z3_ub~6fr<;d?x;h$}XW37G)zFtd}~eGZw%s2H-2<#QKR-OyP{=L+OfjGTZgGSs{=0 zLDchl#fF{!qFJ79iW(}zI5PM?!D2=B#NLp$Dy}S=BAq-;M=#8Sojo&DDHv^FVg)|d z;7UBW3XHn8bJ&-0v16bJET~Tds>vp|a#&~pMQZtCQ{`%p1dG)|XemlL6(Vhw^?*Xy9f_abkwV-xY>+`@N^eq%umV~8gr^ibQG z`3e8jyn1iCLFzV*)aIKQueO=Ad~t-(m2@(PKnjSS2h0w0Vwxxql`@1zz8U|tTK`Fh z2CxiCnO&}#8a#kEW85`IfEIS8Sqp#uv})pWZc?dY#~OW9lD%_+eZlq!(_!u-~Lcu9iAvGJ+NYIM3?yK6sv2cXDmtbFnMIP#Kq3P3_ z+l|@TfpTa7rhvp@NVHW^Tj|q7aT&JI5lLf{66Eqx>Jv2e(DNuE5%>NPK`%An$ulud zd*%~=u7-bZ#bBpL5C#zcS9w?R`KPuAxlMzDFJK|dGldi{A6F#X>{n=qwe8XB!!_15 zEt>pSoNu`^8Q@RjBGmz=0*8_C$IGKWuiw6ZdeSg>aDik=A~3Yfq^Ql2N~r06ivJv+ z{N>KF(gAi2Pn1lne38*q8CEiPBUs(GTMMcZ?Qk@cW3QgjpnC*N8;P}*0{Z3 zo)n~qKMIZ6b;9zz zn9UEOf{pp7V&-ok+|Q;za?k&7#9MC%`fSX<#jgA}l**fhe+J=x8vZw9{qqox`OinN z{4WrW>9@!sQv~rpSYAK7^Xnl0ET!^xn85Z1;Mi?nv|~v}y0+k0Icp)-c?jZ(8PHeXhD~NVvqrm~jBgODkKI9m~#*ZUQ@=)HTW; zlim{4aTodO7FjNFV^(*lgVN|q#8&gbsaGvp-RRx;diLznOt^%EJJ`n$_!X;5z7^YE zR{pZD)8(xUp#||)Z2MG2uN(QU_$kJiXLyE&MlFf-gRnpb+Hs|N$I8cku!-unwZk2m zs;Ca8(uV27N~4B2&$1O>iv(VWL_K_K{c7>GH^_PWE?;9KMy?#Y~=s_B5R(W%HmZjDZ zi^s8TS(qLEokN8lYk5WMB{MI?)RQ)wORGH@xPi?!$4w8W+!wgD&>>E=keV9H;tIWL z=o%B1TX&=H!~p;R4w|tx`4##|a_P1HQ`Z>%oSNcfG@owp?$}>PUQv~ zy@O$a9UK=MGQBiDLmqO+1r%UqK!fmWm^VWfpwb*HtaqwM=s;5CDB`q*t^2yR-F`( zSoqy5w_B8IrRdUYE3vtbOlV#&o@&5>)X8sti$_6eAGG)_ix8Y&2I4MxUY1f0L-^)k zs5oTDFx3M#_g~(Fb!ZGM;nIB{aEhc`y5oOwV0DNi*fNWl{7N$<$U!4}<&pc|&{~vC zz}ksb>_go3kO*0n3yz^e8sA-9V)hm!PSY{|r0J3U4LhH!m)G|-XNvj-W{6UKGm=TC zIFirK=;LW!V|bS#$dKx2zHVPgZbEi+7o;1E_aW6bApzjGkGYUoYXW7f5dr1;%0v3? zq&DgH`2jf=Zi5PRB@!+;=!BcQGduIh4H+2iGoKsFTC*yM40izC+ zf4r6e*J^Wci5`7!@_(-S9=x3rW4*%>sx1!(@^W&RG68eomm@cRs`It;Nr)*!V*dG1 z+^7Y73A9e++^HFgpwq_j#Zdx7o)my!i~=_gzJ+q+W`jp?e`W?d^@jXUISt_m-(MXGsI|ER4^ zjh1U|V^}M+ZY}t|SZJ=xh#ef|uA@iMygTGzfd=r)129-b7YC>T^vMq=vpXdVMFKdV zag^WysYHRoDbVRZ-Kp*h-~?ir^^0&Sfk(NfiP^?P)27sOcjdetQv`syAC5Dl-c00Y z2X3a?BRdFRaz67E8r$nVo0NUF?pxU>96?;VYt^VzY$TTG8~vJbQp(DWd`7(j{dss* zHC>HlmD9x6wM)xT1SwM1OKUXogER?cv_1QUcI^eVWcU{%Z5OuNRuXpHXP*zElAl?Rh_dH$8hZ-F zoJmg44AfejwYi;}ITgORqy&dr1fidJr?WxX46uwrkdumTCSZkfj zxr(|*0VW&lw_1e>SU_7o%4@Y}?Oy6F3>D5X!?f&8w1jc+ti-Hmm7%5$#>1G%2_g2B zv5s=^km(b^W;<%%I1gIuOndV9;)i@y?zPfGrv{tJbV&`fR}c|+rhjCqIJ@f*Nal zB@w_u*&cwJm0Pw5Ymj1FUh&z^jBi&BJLjPd&7j17WIVP0-0{gIfVc+;Mk-j>ELq;A zj@)3ybuoq@pSBc&fAElczOX)I-1FsjJ>pw)yK4E#DL2xcL*(JDT!MkyhSx|D3yobC z2N?e$l+F}sk|Sf@?ATOV9sXbtG>}#_{KrutDVU5&ILZ2&`a>FgN$HREiB8k?lta$_ z0;=9lXwdevZ!(4wh%P;}HiG?SE#;ueR^B}k;*E9XKtK+K1QZ1?fw+-QeA<%)bE=gR zoBm)eV`T)}wxPc3pvbK$@8%AHg?u_<@?urMo#l(?B7oF@FiX}*S~_szte_f*zCF^^ z&fv?FSp$l`U?)@4<$NF)R%p2X*)M%4u|(uF)&85*rVL~%vE0X9bVDuYaLmYhN$Ww) z15d8@^M&rCu7o6|ljp>vyV!E4!V!81H6*7#86=FOq%+3Se!we-V1-;@v%=Q#7 z-jKGk3`$Jx-t62Vjs|br0?lYFnVh@qO{G#G7sIfvjc19!JkuGp^dzx%DCdEx(>WQPFJ&Ft zca?P<)s)O2+ogR3qaxvD2X&{x2YkBasKO5wQ|`k5r%m7Qk4Jt!w|8=?^Y3Q| zlf(>~zPQE|ZHx3}!ow%)u7Y9APe6^dz6 zF?##RPon0R*M2eh<9iJqOzq#^sL90sqx=8k-GipLSSTf8dwD{ri!k z;FmN7e{FwzpVsfO(C-c%{zet|e*^R1;?>w#esosx$6>^O1M}Z3{4<#U)9@d`)!&Wv zUm?FgR!+W2HQ&1XyVWFWZTImn>*iNs;@gDVJJ=cNeS~t!p#BQ=zrg(J-(Y^mA1Wt* zcIVeY{s%Yx-zX>F;3{9E zfW3jy+y27&7Qsg^q-P^;WM*Re>(d|m#E;i#ncs#_FKVf0Vo$*G%N_sYhWK%}(=xv0 z3jO&yBja28nl~$N;@}^9%-d@Du~q!kSpL!?tM}39?IZtkBm1$PnHdOJnY?|N;cW^W zj6RAJywy7U`AFsGqaQoc4>Nzek^O#fKene|2lubmEE~&DO8oZ=P)okWs-FPif9gfEdYWNAQru|gH@&A) z<3Ll~QkA!mU>fe$#pQ*$>e`;o5v2D1<^`yd-ktvsejg z&)+-(`d_3hj0B7?+WoKex{ryEaQFY~b{}Hc&hMQ^+I|ZQ%21!!SO_sSw~&lLN@U)Ho>JM(#0) z=;g5S0*+r}Faba#7!D7!7=i&`*nsj+VQUr{kO{pa8;k@D;SQ^b?I%$t-qBTmn+BW& z1_93NM4alz*x$PNtOJ}%LzGjK0*K+LX>0=mT?B3yvbjP$d9EQgJv%@Da7vhxmgeUZ zQUy`%zsP_KNbN=Rt|J%09}@?J5QkYeycH8B1!zAaWb+*xfzl@g<&B&-wxfyrZcvJP_WRJWi$IVp)S;SChjx1+nAk5ypuM5@aDP z7%4g+LI4l8E3FbNX&fccW3_}4O|b;8BY;}=06BG5+K6E2iz(R)7n00NZLivkDXv(Mo7Fdo%PxNPDMo7(5s=Zmk6 zj!WGQJz*_}El(?-3PHT>-b~F20mPxnG^#Ia@g6_?JKXXjL!SUjlCC98^`nc@yLU!_ z(y%vxXXHfZZNUEa5jb10Z4+EhK>88`pu!08{Tyn|~jC3#3 zl?_NA09Av*0Mx>~RDB-rM}`RT9nKA`H(?B--x!#9#tcTVgnz+TGf+O|Jw`4tuMqJ& zB29d`8zn1px(4fx+*- zi5afzR0(ICJHcVWa@F-SezG!_+L!(66G49X@pc5}aDgGow)8*~Y7P9N7NbSDyPdBI z^jy}LN;r~kcrhY_qxIShUQR9otH3IB9U>h$bPmKFPNl6Zdt)^xCl9%1Yu>9L)CS`d zu_WmBdF-IN^5>CO$Z8tvdOLDswK37|>ZcZ-mj(e3&;+J{53?IRl>wjJN`sseDzIA* zbC{nd>ok#u`+~(?1)Rr-_GHIyq8M4`Hd}RY&h~F(zoc@6$H$Ul*#?<#KBV4rT!>U|}wMIK27W%v?jh;s@@wFZmv4SH4YJXHLV3 zrqRVM{4+gMZf9_Cypaan;&sD$;6v6SQ0BAg;rE!Dl1Yo)d%xzZuN8drV7dVN1Rf)+rImZX(=8k1@cx;7vmMtBDRac zJH0gXN$ul2;n*pMWQoM&_^DdIYDm`tU#HuVLDKLpH(%`y?i{&o!b1Pj^8jDKq4Lxm znWj#sHIxU->_w+IgVeUdTSh7r>`%Dowo1#5PUrJxWur!|l+gLnx6WruSQ>^4UsEWN zEqJaz$=nu=VC|$r+fwt-$+B5x=kAhJJ=og2ho8mgwZi8$SD1+^!99l?@PauBsJo7i zJfyYgO*wu_1|igtI?W7sdF7Jla~pLESCkV3u<|S@5>(mqDk$K3enA9KOIsltoP%BV8&3W%%GdXCuD;l7u z?lEdknxb={cj(S26h>s=Lj8$NAlXEbI4@ms$J9KhpUjEq0CG3XQ+^0XgnFZCaVu1LKbC0574Z$A+Dcty=LIVk3cxRejC4rDj)h0)@a{ z^iJVue4m!bn#6VVT;8|3wts7|%8A;L9`e(8;iz$qm0P}>*e<`3I6dhk=X8Jg*qm#& zJv?6M8Z$KyUnP3@e5pc5L(#pt6uq*D2i0~|RMTRVa&g^#I|<6H?RtjBR+~>5wgrw&M4O6xSs$re#Hr=rHU#Xzf!ihYp!nLP`d{ zxf7I+EiZP8&Zjlsw7{>AzGJo>INpV<7jIQVrapF}F9qt|dc$RSEu-Vk9k$-HB=J*K zsa@NuYBlz3wWjQ(7QI*)WjJVkk2}yfHQr>qaG1EMxg|hFtm)|s*9h~o`=m|fKGT+b z;(4}&WBaL;@?h#XG*B|JF&%DjKKlZfGRy7cJ4dD}vNNaS%_BORw<=#4FgTDbEMEm$ zxpJFi@=Q%1WzEFb%n#CTfuyT+%KAy&1NabwAf>4O!y-W;dKr+sEOzk|u7RUa!-bXu zkY4g8TS(cMvrCi#yRx1pqFSV6D(=~fIxNfuP6bb$i}xW+fUVO~XW|sA;-OgFwj88W z9*uD11~o&YmWA=(>n9?n=c&9Cv*y27?&=K!quuM;7xt<@#Aa{QthFx(p|ZE9)qa3C z-_%d9>OeK6z-q>!lRAzyF)z@%yIN_MwSq}5FMe=2FkTB#DBRWfo=_UHt?x_`-+-en zS0p|5Nr#=gdPd6i;7eQG9U8@O!ZFuSl&Y60xu<1PeRr~@HK67RJGT+zO zWCE4=5e>{wGj`rnOxekGlE!up6}w6mXY{-W~L{rF&uA zAGk61Zpdu~(wZga%`_`MmTCJ?l6yQ#1p0|ma+)epT~`l{1Z`GrHog`rIe3V*R|3R$ zU0#R1$c>C?3GEYk$}-L`5Fho=J)EjI^TzQQhJQMqHj=p=zcOOhQf?+JSaF9x-|w3#|5t)*VTJs)&QC9NXJYY}_tSIEefx-zWxf6lf*^$9+($OVI)hO-D${uj*h4vzxx=Txl%kHsbP#uYUMqpXmgx>HPG`JW@c{Rk3R2R3^o)P`{vL`^>(zGG zF+B65$E7}FFw956<^0J2Lhm;VTF{N>b(nStf6>fxV^@=p}P#`)J6Kh_`T zApZ#rc}uPG^T6bon6VhTTr4$>5#vg1Tc;FwotM z-=#i0W}E_wOA-~9Hy$f05G~Fm&lJnSchPo+I)9HHr<@5N4s&i(#)Q(m+0KL+n5Pp& z3{!4F62XGZ?>D}%y-@Jz%f2FlCwh)u|M1NW$Iw+&?wRUQgk(Q^Nt;#B8{)th{}f#T zqx@dgD|S~v(a`mM7>^F9EmDy>`z#A`)r4+nt}cWS6Ua?Wb z5u5QdSA*TJbSC_e3|-8eOQ*55Ak{fLRP(S}*BEY_Y#P4-WG@-m-*xfVe#?9-EH>6Y z)9P(m)rnlGkcVwP2NhXft|_~uax?rz<$JC?rCBhZHBP_3`H-(lxiYC(S%6cXz_JD@ zEGX5}$^fOR9>%u=k7Hwo%&+mvO)8Eu66+DX83IY{#`}x#`5FhhA&R=d7m&W&xE?C% zk|#p>)Rn5lR7-dMef~OgCl9fM;zk|G9r2o^L1P9h_K^=UZ3c-B6Vd8RTfufwc!V(w zah6B6wI6Z3G^r0!c<6j%R>Bdn&)nuDEfmCzfG+ZDxla=KjmoKvXayp+?8%*mZK58- zqEPt1-=TvM;#qlOytB&r3c9#z>x%wP!aGPzavOJ=O zcWw@Ocp$76RS;Lx`xXlD`RU8L<~WlZaC8y{t6?{QMwX$*v3kG^%xHS3slqC@c}(0p zPxWHu*$**Y8}z*d+J#0K(9%g1Lk6R%1QTI2i30*S-}I1parxLK z)R0!<9?FzxBEt#7^{73ErmZ)X3o4R7!@!Hfmea+vi1(3*>Sv*t!j9_I6QGNJRDhSF zz@xoNjcp(UA;L~%n-eEUnd!yc@)eC_5U}$HB&G%`6T@C7S1d@=qC?;|7ekQ%?g#nG zoU@#ZyU@A2A8BmU=VdCOlE^Pn*2x93O+Hb#iFFI)7|=Bz$=Y4dYPaw~@EdFj#AneB z-R!;5@795laTIE(WlWU)P^{x45!f$gdincvy`pB2A;J`g&`Esxwa6Kvtdkwn$dqaV zVQ`rR$r-2GoHj=bpCaaSB?>>`Cy(FR>Db#PxyGQ|#u>>S8LU7f5A#~beJ|tJf7e10 zj$$S~!a^|;0q~mi8WPSRp3fqTRi&B70>=mS?VQ-$q~!6IGicmkxR_GZP$X7(V1|zC z8z^u!uCG)>0FS(TR_Bwi;vQzCW)b@L=Y!;(5*PdRIAzR!1&xAF;#Wi=l3~y4>)TB3 ztkx1)pspE+&y|{j>p&O4Uy!e`4HXc5s}RkBFP|l^kCononTZ zhJMuyUUzmvQZ^aYDa*`TVo4r02?v+mm2mZ0UlLN9=OLp<9F0VWS?NpX&hXNQ?pcp1 zI#@lP;O24d=o45Y%;oXyK~j_kw!vtEHSDq!i-Kt(3=85DU5U!(Ye(0yVn8qKKK_>V z{sSDzOb8$dAuhZW-5^vXF5xiwfR7KeJdHT7j1k~TwQBu=i^LbCjB8X=V>I5IeO3H8 zZ5}4q9H_O5!sJ?*8~T_e$>)-ftd}GYcG#;~(7DPfF5q6I+1&fGF2^<~Z3547;K!2; z+RFf_4T^2d%E6ZtWia88@PT4}2iNtd^kqn=PgvT5N|+Hba#)Iv7$?L$$d~gV&V2Rr z+6P+%>#5_Pu@8!qoG_*{u+oNoY}XRUK%=Xl(=i>snVXe@L1qV_GrFBD`OkF_FL9sX zPD^X~Kn4L&dMJvKl*?j0*sWd_?8ZvkxUy_Z1Z5pKbIe>nu9W0Vy!xhm&blF`uBSYT zK#>FGA@-TJDo?XOa?thNLMNsHZul&AU{lrRC7ip_2R`}|=}xe9XRHr10`MUST_OTL zp-jq-I!vc{fCFA|nJXx9@HjbRQ%=@B3G7`*MxY_a9T8= z(THJi<2&u+kIp`Bx=-JF|0%kjMRj`^7x0;ndWBNp>*FYe@4cpsL68__V)fh2bgtB{ z1e^i;!4v@)s=W>A6cWebIb-drFBBbLj;Yx{La@?e<-W-jKyE?K#XGA`j(mN$x~k`a z2i2wm7mv)h`ZH6>!Ogp)`io?YfGV>NUf`t`|O9KU$EM!|I?$_b`(DKr9e1@Y=;KjEQ zJ3e?8FsPnQW!rvY@YDrAlT*kz=$!u5nEa{FL+0t5x?^ifR2+?XycD$I{bLDL^+|v& z5MR6bw@CX#;Z}N%d-QATW~jITkuz5Ac3lqw$DX4$m5I{J!@}$?vc1zh(UcJPd9$qg zc!F?-nS2w~%%FL8Zdq06_lFN*cNGJXVTO2vZ{^L^k!E7NI&*ictQ+-EKt;vRjgb?%a$+6jKKVD{Vr8$V~5Z*mjy{bO5u?P9lS zl{tM0ck4w#mGL);tp$j-cr|JlqpkFtoy5S6I@BcOF6}cetkwOxn!N!%c&Dm8+56^u z3-_kw)=9vBm6w?n5z`NH{ttv@1B;T)&>j?sAb{;8 zg3=$YrXSt;b&&tTYWh84`J=!Z8|y!Ymi}z|qj~!u2+N=J`k$kD|2bj#)7UQ^{2izF z6UY5%WnyFgt%w{O>t7(_+YR)uF7}V&8vh6bt+0m>-(cXi6aXZ_pnVsQTb*bOfvA3f z0A9P7={Uycs;I@o5Xb@r8_IpR=OfJXL2ZIYCkNe_duDN+4_U%4G zNkzR&4Ud@0#neFR4~Ryw;zDxul9uQ?XgdAy?U@CkJ6kf3J5-RK#h4aO8}-r>joOzo+)VDP2whA(;wFZ-yZ9$zeToy37#T-T30tmOt@@0U~(J$kE-Z)V)K zxV%t|E9vE3=*2y{yCt?+ZsW@)A(YSuBOnk$(K!&`?P=-m^i0GJ-gqxP%ftk#AG&nz z8E^4QZQwA~L-7vYyZ~TumyLI&nM185<;Os%JBp@wUPXw@z^fe=c?QZBR38ok0Tg89 z2qud199hw)wesY3FHqZSI0AuBYb@4gVer)41Q3ZLz+c6F-MH-rCiFYwsMtXnyVVn-%>%+|On@hO9UugzTamYpg1O#ta)IlL#2|Csh zi~tSX4(IY{U~}^va6dYcK_N-}?gU527rP_$J$(euLb(3-timdv(tDb#j>@^!I%0$r zNQ#axoajit`Lkbm?17OONt`IzJjGJBcp5aeCJ;myZXI&{`&5AXrjkWllk(u8>dh9r{JVj>BMIg&IEBu-M%xo{iRE!!Z`% zkix!6@(xZULf0AuUMuC;G()>VD7K!=YGfoHRFVTZL53&B~Gs3IZoyvo~Wq& z6fsw0moK@0IdBsTK<<=hiA6LRPaZ&(bCO)nA}kXe{M`fsIil~NnWyiV01L!(;g%B= ziqEVNRlKQ4z>&{2*LNArdRyGoYD^K>A8Z_JRLB6d0f}2!67Wm`8Amj#1+;iChMHOR z3$}vwO*oaxL@Pj>z!uCzC@oIO?PTv^TLT-`Q0OxOj3S~Trd#5`rBQT|^=%1$e-yT% z)$v9R3j#ZY2Gac=0E|y)O%vhEBz7Asrg==>i8D2K1`WNt4f0b~c2=@jymTR%C}Q{d z6WBdP{gAuR?N=~eZ|^jQu-g9Ud^Hd=AK?D}c677h^>M`n4%-yFPHU`f8qad|dH)~= zm7E4bLpwC9h*oqhy}@4mt0K}yBsmS<=sf-1spTCvTbHPU)KOWbs*_b@C^cgrnHT*a znKZ9lOeoZvtz;Gb_-~uM3M-vnLNYh$pVmZCoqHks!DE{C&l~xWLsdXoH(T@?_<_WrZ!2;=ciVL+NgG8% z-LDK}!+F^Wb>sw)_UGduytf_i1l_??qh~&x=m#|<3I`Nv=wd3NqGH{rc);OjzL!_h zOeCyFRaL_n*>Xtwx?FSsasa)@CH&e817v$Uo6r#mL}dZ&ircrJPM(&xi>(_V=WZd^ zQk%>PRYtIZxQ_7@Kz`ihbi@m7$ypY&FvfK3R$fe2B4~HbDiCW>wTG=}u9f6ragUm8mzO6~eu{GvL>ukop+bFd&O{_`^;0y}!J`PSDK0dT|Wu|pu|p_)ze{`qqrynM9X#GCh}mB$4d0@@2Z^|UBx@Ve1S z>45R(Mz^Nj2QfQ0lkN&E-d&q|)uK+x-g3UvlfDZCFlQJ6z}Ccg^|!Wz|}#?7gWZbv;#e7oki$MN3nA zx^dyd` z)v$L^;X8T1=N6xQzacG|R%2ed+nm}@?aZ37JkM$xPexLW8iM_7V5A^%b$g5mdADxm zIBky3D9fH^KV+n2lM23HRT@zXpGbo zs+l5^(JYxC9IvS#BBR1;Ac#D~b?HJ`;Jy(Rl0WyfsQ2#iwYN8kWAs|y<3~O1>!Xnh z^zU)rM72T0GSTORqN?M?>!&15OwH@A**2{G3-VM z1doaUM6F7Teq;AW@tr{RKBJ{~S47F>0Xyv&Wm=3gQ`IBO;WAh?A}Cw>7-*F4YKFQz z<(Q@s-C%=O!xs75r%gMX>m~we`O4zzgE@J@w)I(q!#s4OI_y*z1kzAnTVebt=t>#9 zy_Z~Z6p?ttHIxzrP)KbtJ*&nyA362l>-8+|QnUOa$h2{0C(FK$IMoXWOAV(6)|Vib zW+Mi^8F4j45o|)l3hBe<V5KLnJopQtn-&k z15)-&;aXFV+E2a|@Rov5fI-Z|S1R-FCV{P(HKKRSx3{TBKD7_6wVZ-7<yVN(Na%ZwNA#bXyoJn295 zEJvWW6JACzsJDK=B5pHnKN2D4wvEz*j(;B%wF;K{4$xzCrwqRi6X6&(GMtvwrZFB? zBrtku#*UIc4S+tlVrN-$yaOmk!)ziLeMOxrMD)G3fkt(=&}7|?$t9UQR&U4veV3ew zdGbKvZTVy=fJ`jKtqm&pN~hDLBu;zctO(TFL3HS>Z!BR2t&e^{>j=^4ha!qe8h7|; z*fkop%J4j_Elq;%;Cn3YX;J+RTL!^->QpZIoY0gZFogA zP<<5><@*WAG2|wjg!VAx*j1R|a6jbZn&2>k@NXo1<;(frPj5~#a(rXDZmGQHjSeWwOK&sIdry@#-&uS@7Ah(Hr@>~NXu79UB9 z$dWp{3KO#BW>L8wNk$`PQE4_UHQ_HeFOk_2gQw;;Gl&ZSgu!a04^4H?N!^1ul?NG} z9Wfct1CasJxN9AIv>Qtc1W=36?Nbv=iPYiO@EA5WrNQPphmx^p+?jm83KNZ+0|ZX# zuj@{b-u{3y=Q4mAg^c$sD?Nlnz(*x7tF8s+!YDD7TEaxtP0(30YQJ`-1Zu1x3aS}x zYU|#b>;?D$KO2B_zC`&nG60;{qC!ia5?_wK$@(?$>(Qqx?7a`YvX-C-L##jpO}pNi z!+tqOdF3;DV?q*bvp5i+u;r3->CLtj>;)0l3;LO&SvFm@VFRNCK|xp*ILS`FHi83y z9wkjIJ1kvraEY%9469h@`b@1|uX5ENijU~zHS}oiGL=Mc?_ghS`xPeMMyk^m1&-<7 zfF^Z=ic3onslu}NLzY!4&^7?%eiW9wIKdVHrVy>J7cO`4kL&o#Y#xK7+XMz=Oz52G zIG-B1BCqfC06n?c4dK++U{G2|eKUqjnzS)7F4kZb!@&W=1uSz+lR= z{jD8^LPzY~ZlhQXS8<gZ~I{;-suoF=5{Xl{~qs?Q0Jm#{Ij`iKK*BN*U9R`TX= ziQG!zjP#KZ^f#MXJ^;&ZFmM(E-_3M?*+8gNB|SLzOX#1J5&$>G>+b0E`x4c+P{PU? zM{93Y2^LD>^+v72KIf2Qj4+ck5*&wfp8+$TQIaoor(C}~S3qN%P?!$32Y|f?yHK90 zLX47z&<>FmISJkEno*EFuF&>l<{%-0d}(^FNmQ@|m^b*;s#_mN$p<2+($_)ak07>b zrml87k=rxe{SxcCQG6Qa!a!L_sOmW6#L~dg&xw2v)@|U_HdV>Clw_Okd7XtEkqcm2 zYjy8Q^n!5dT|o#=7vW{oD`yIuA-Wvn-e#g`@hpQL4fkdRMQC)0UFF)jA8c_csEgU# z5GP8e!qg=}nC|Nnt7u%S>_hYRz;q;GM6l~DKwiEsU1(CKr7+#2aSh1UjM%Ogsa=RW zF4}jRlJH1fdp9VTIoS}`l-9{c+iBzOg4X@6{s^9K-6exoV{);zG*Nu42(QKmErDA( z2O=vBP_}&4rIb(3K8x%O5|ByioCLtJVz1;-B#3X|Qx`b3^fU^%EfyKabnBysm6=O- zyW1-Mx>%x*55GEqG%78sLiNVXTUsev5?}u_k5~^y zu*PEUGB1F1**S;`(hl`d(PN-%S%FbxDD$B;j1+nwzHE9!XmM9Xx0jW>M-D>__RLkY z`?YB#zKKuo1Y(UT%bp~jyG!?}1^n|5_juKIpBtDH_l=17xD4&C!uLFCx`+0*9r22x z?aCqb5aVv2Xf6-1AMWJzGRd8<^8+c70pJjqSMQv{WDMgOlWBqA58TYf@3Z|3?J-e| zF-Z4|_l2=Z%xnUkv?K7gin8+(SVt<{y3hrQf=h%?vv=wmPqkg6K#q1#v%$_66w4s&l9W-#U@>$RGPjk^m)$YC7JY>s1nliQ{wa=)T^qwDUDP%1> zYGC&ipcM9H`p?-WtL!-_&Ut95^}*NjL;yURf5e@B5Wjc>g+|0rmk6O~Z!m6tCT0$# z`0x=goai8yiG+YM&)-WbSA_dIc9cQrB{uzRJ3{R2b3k3Revm9#t6b?-@97&jALyDl zNLC&nvRFq_$S}cJe33sH0do*I;N%-eooJnaOs!`WQ(gf%HN^M>y*Cc>D`0nyW{|1q zC<45^NT0$Hom(2<2wIiaccP=;+TnT;bqzTMR*n!_nbHbFr*!J>8yQ-~s(8!haKTra z@=5!QtjUu=aowvqHC?Mi-{a{6<xG(f)WzNpme z3bJLp){Qw*kx!Z z5=2PA@oDV4+G(oC;O<_Dr3eRYhI5}IuA4*NEr|zBG?p-OzBZ05!h&4~`w@nnn1pSB zrtJcua(90&fh_*S8A}qUI%RD$uWGjD=hptQ!iSaJE_7T)Swe7gVoL3qezSsz!XwGW zmDHL>#(0U6%_b-K`bo(b482>D&nLTzcRV>B>}(6PP&Y?!7eULxfHciEgaBQ)E6lH8 zfodg+czj6J>TaUp);Ho3)0dzblVi=*X8a3vGBQz*mSN@B%F(Y?dy^E#I1w(jP~=Rc z=IkvzR6W?12x496J6&Lw6Blbkg`On`=1U^4tK@tparS zg5pOpaOQi`yYKSqVIf)WXc{~5cYr#rqOYxdSSZPlq5YX}sL^SADOSiZvM&ygIuoAd zxqn6!|IcPxe@>|1XI5F+Uh7J1--mntK@;h_>1XWmf5WG*$6HQzn9$!8DA)o$0 z{(n?w{p)7yB$~b-=eHl|U*%Kw*UB&Z?_URgN4oE( zpT)>O=Tk8|Ye$=3KaU7063`19JDD3AD~Jga&`S_F*g3uqDZg5k?DVbdZS?JotqfiN zHevexH~vyQ|MkPjs|&_o7o-r-|Gui?^`L+KyTBh4?ssRuUhxOQ{f?-=See*48UFqp zKU8=BS-7xq{PvVT67Ev$>q@*|gmG;PEW|Ioih0WCBuI;0s8^&c@GbfHNk|CEAX7lZ zW?!q9>=Yv+jl!CANdqX;AT6?)_E;C5=h|%#g>@>h=|!UF6|z&p;mYJIL;Pg}A_PD& ziNzGLaxju*Xqjl4z0N$NVPE0Tqj@jCk$kJ=$wlo9u|kBpXD%B9gM1V}dNEjd7!bjD zD9&>_^K+kIWDbvN`iP<|a;M4cpL@bg#MT#k^i z#-aAGZ{I8ux=H$GyII2MMBJKeQzfdST+CBNY&=udcff5!J#~UF?@mhTy?_X))Ts&Q zc6`I`YxlWCswSMsc|YSx6qY=`41~l@M)?uYcMlIhR^pNO+lDf{{qT06`GJb922COS z-kCHh0;%QL*IW0o#HU<-;vEkGA?yW1xM?(~>qgeHq~`k~LI6=x-}Cx|csag%fe2#O zv7)*3x~`kOGa3=%pn9a-nrV1HjbZ^um_x~I>ctf)Gblahd~DVXD~bgclQ1{|B~oxb zk2PN5+y+Iz71xD%gW5w_Di#HMb%yL-%tHAK>T|~Qw&U|M=JdH?frig;EpAsEJ+>3s zgI_uwg4&5pZhY(s-qs-2c*VMCTp*HBH)~@1>R~HJ4Km&u6{FndpK2q&qw%UYkN&j3 z$vY%ESC_-PQ;lBOifSv5b;r1vhD8DQ$kB@)d5)&trx~B%g$ZOeamS&XL(y73lFM^- z!9oW)vg`a&YtL=_jm1D^tf#J*758(X zN`IJ>%u~ON1Ef9T$hSzLX6ZGW%pR@UeqH+Jn7hT|_F2?}GAeJ{WsXHQf&@~GwgdPZ z{g`PGm?Z!U!XCa<<}wdHkw8mcZrt(fdXgDqlyX2KSxyqyqM;2UB|Xv?#orT#%;5Ek z&CF!sj+a|LE^!h;5JpT>n`7OJeAe*t=-Rjk6cUI0s_R9GgoS`ou9nCqbqz2n%)-4H z!Uc(>cnB3|Eu@-@x24IIoXeN(m`Bs=fY*hs zcTyd<;wA{y6VcA|X0`4Je5_b7S6VB5@2{jMyQ6Ry6_hB5V93NkBP)S9hJg77=N+^l zwmuJ2M4oTN+h35q-6G+Uk*7a4p?1~*E>CZ#9}Op>!8QJb=WNKj=Y_Vsz$2F^I$*jr z{+ylUhl4@f$${cZrzuADqJwnpCAv@e1$gUOPv7c&W|h<#SV1O;l<+>M1j1ao1B$2@ zMnL@ZMNE;m_&YGjGjTvvUnix}>@EA~_7oZtV*}Fw zrhPcw-O%t3m(J8Bh)Qi~OHe`v29*ng5>oaNg1Ne~K87j1+lf+6u%&}3Iw=_j0_h_pM#p`uA=q^F?bgpRz+#qA-tBPF}&(nq4JS99O?=VS*JpG^llhrZm2!I zo7RINE@VMRa%%0vz#tLHV*$aP-M}@n&j9IpXn7VIdX`Zdc1YQ^TtS@fiI+Yl4RBm zipV4$xG<7(hN9U)XS?N-(%l$iDxuF!F&d4KZ=7Egf_sX8AZDuR#r>N)N$pn0eIQDq?;w#Y2MF znWz0dL#p~+#3cs)q$=hVnkO|b4rG}Ehh=87DWOZO>nvJ-(8HwTpm_i5nCDp=H%tH0 z_b(6_hllvKer~-NdMqjw0tetN(}9gJ z(EezpG|a`as9!SJmo}l~ye~L~i*t1&F@ZX;a?p`j(UWE?*g~Mn0j!&svK_}ZC2pb8 zdI;?t19iGS>Uv72H*eb6Otj6N!+avukj!)U?RE7|yGA1mK-Kx?z6!6z@|pB3jkO2q z6o94`e-`;Frv&mx5FgKCm%~kV<%hzL+m$A7Z+PIhxY=v)mlG;6X}4i!13FgUDW3y2 z*+z?kbI=L~FYG^0{Xq1LtUeWkL7b zgeLq=`bunJ@RN&kl`6aJ+W2-f{e%dF5{WSi8u>}U@GTeG2blM1lduqZK? zqsCdQp;cf_JhAzOy4pw9bZ;08b>bz`I%tuylw@gO@c*iC@Vv}Tvf}T9vTvp~I%l4&27Z4&bxKM_s;Q2BeU470KS&b5C>i9aIJ|5LJhE%CmF$zLMcKPId1 z&i)r!eMh73WW~(M@%wXd{DO49h0#9?m)8a5zXtLTWTo~>R&3u_j@TAResD#mua_OQ zDsiesO-e7Kg2#MCD`CQ(R6skF?L&Jk0@8@Ul#pY)1w3={tO?Kiw5i*g+wK)l6i<4` z9=o0u$3qG|GV6Kn_3B9)o}FsP1{3owX4(a-bMx~M&rfp_DK_}Jlk}<*C}$iY`ys1p zwP%=56Wb#qd(csMxB9hbZ#o+AkKt`!7bD0#hLaAZyeSngA)%^N2 zeIjBgwZyQ%EN^@L@U#=v7nK??;O7|c3fQCIl?K6GT_^{TBYJzIH<+PhG9 zJCfxod~B+IW-;Bl>S^K^Qx;ug7=cq)bf)WVc9pB|Ic!HOLLZYU^OiVO=MoCMY)Mch z@K|Hy+qDJDzJE`jGj8E}bWxOB!@-K7w`WC8d%W zLFVSAp*D6Y7EtDg>#lGr?0Us!Gq+iZ2v3WNt=Qt0spGgdUAlubBd*@MsOwLeu*=fUf(dOHCWt@lHo6!BW!P*K|w^kQcVRGJIK)bq$8U9$izp z_?e?GL~0%YvVL2){HBv+vnDwAq|gRcwDjg=9cUN&S2@ziIY z3J@FftyoM|&{o$gCZ4u?il{MmzJwk2nbR0yODX9O8h)BptFgryIyBR8v`k8GNVTxa z)G2FC+wlNkMf4i4c8}gm=XxSd&V=)95k)Qp#q*LN!GR z)If0u^qCGWEA8?;Vi0|NOoPK3@Mf%Q5Y|Z~PCK}UT(ES$XgvkLisD_}Z)m8L z9@32%s$6Wy6&x%cRob9rwK5G&Bcft;gzj@G+4k7rru7NcvTHEM)ISLsgvBN}BuB_; zaKK3Gx)c)7t`(cWELK^*s!5)o>f_JeH@z*}dTgexSmsQWqAEdT#+8xwP$Rqnd2UBh z?i>S6>?H@3t5slQ!>Uu@o&BhGZCuLp6_auo&g%x>_}eErC;XE;^|tv-=dIGNYGoV3 zutS3FVq(4lm6ea>yph_0<=hAm+4*xD@Bqc6HJlU2d~&)rK(e_EAJ==4-}*>V8GSyO zhLL5d2p7?We&m@1ODLN$I8yLK`&=%ZeK__p!+bB6D>92QB#c0&s9v^VLsn)BC~yj1F93V?Lx_QjSFay|AAO6J3v+ya z9oh!zaBc&>QN&f}5=c1WlrZE9?Edb@wDQ$aOe9p339d_Bb#n)&fOi12%6?XZ z`NoHe+yr6o!M6)>LMJF)czQl_jIa_8!vW1@IhN%((=L9gMy%E_L^&|TG^YYHeMyJ# zrj|IM72^g~2*|_X(|q0b!-bbJQ1FIjn7IeI$+tP)S|x7pz5WIPr60jPY0o%ZZ(}L$ zo=4X4(Mk?FffJE$Tq*0tGswx_qASiI>ic>s$y8ntrBFER8*rqz8q#|nvUq-$=a|x_ z5_juNsI|MatQ&MHdW00H-8|{u8hEqau}bRQ4Y9XTK~6)}uGQq}v4#@xpeq|?&27a^ zDrp(faBad1f;n1z=`yxMR70_CWCMs%T$+1K#stvMABcBCLf>q!2~3u`zpfCD zjAE4WBnf7HDb41;q*KMutol9(q3F;(`>jN}iXt}GE}*KSxBI9c%jWnsgVx1#W_sp^ zHv=CjPq3#QbmJ@_GnZDT6!KoB-kx=MuK!^_+aDQ5xR~LW zU%19l^4{TTw|s;iyF8sz3T$r40aV{!n|{1ywW16@UwEtFEh@XJ3$>gtP%{;oJ|+9~ zF1-}fPH@Jmh-_-h#Ak{|W-TRowd5kLFT*nK4cf@od@**%O}7!Q{c2AqAaqVNxef<> z#@73#c>lDjxPr%^CmsPc%UbkD^Y3zw<1VJ$!;09)YYME-I#h%ej0nr#Z$8QL*QvTXCEuiZz&8|?7 z4D#*-QF++_R9sXwct^)n4+5&a4-GY@x#sMSc{Bd%5I-CtNbvDP6WGey% zYH_{70t6S{%^AmZ6z2&f=7W|)VThj28N53v>oK5&39 z=?PW3`h`@0Ogc3;ErpaxL$(M{cn=V*1CSA3NEZ~)5Y+U986D-K4B;x_I@U6EX?LhM<|Gi%l94iGc8@AX31j_j*%?y3KCAHIXzlp(ck8fw8o4 zzLg$M>A{~*;em{aylt-$74#~B;65*dWk*S~${cJe*q9}qc4>XtG$Il$_1df-wQJLL zOU#&2QJcg#ZJGBixTb9u4q6W5mQ72i8~fTvUbf#R?XV@JK(kL*MbveQ5)%-3Vaw}# z8F3<4mT4!7MmMXLDMEg9^if;5>w*J(S(3GAlqszh4C{hP`}5*sX28Kh#DbdlXI3-* z7I#eZN<<QA2@Az)}w)>#K)90;2DZ|HWPZS%&*_AmaE=nSUuU|4)FZ`gb6z`e`fV_v8F_$v+WI{{TcR zuZ2F2-*NQs1n@&G^dF%4{}zb8*G>P@yZ;j)`a4Gc@U`GO(fsn3Ow3HIzdy(8!04X? zBF^8Q@^9hkl9sMEWVOYC)V`~F_WF-y!ganib3_I#a~H}YKW>Rf7AaMnc5D$*v4g4i zQyW3P){)kDpgu=Air=STbb6h$fdlY|OFpaeao|$e2Nn~FL8&)YvOGrbYN6BzYp4v3 zu^4I9TJPKLzj&?L%A*PR2VHI(#`Fv5Y+d%Fvy~Bv^0|2~KuB$gM8|^;=TZfS`RBW0 zbSYq?4RVIEe#!l=|Jxasjw~Qg#1-r0~Y44l#Qy*pw?Z_jPHD=om zoxXfKHPE|OUgqO_!tUS|6~GJBkEtmHYsI1#y+D4K&vxw#X+<(S=ye3*N`R#{16xX7 zx;V?wx}F%J(cO~Z4PhcGKddr4d6K*c1)B-xSDA*c$>Z6DWJ&LZAJ9OKmuG(RhJ%+u zhSlP#Th}Ovc`)Uw1TkH|a?)yTXq;!%x`&Z7JdX8dKG?HeH)dZq!zYj9Z7Bm-d<_Mq zfe|6aHzqNtm@May5>WhfvKveWB1mA61b1Zg&bD~Vd`~-;^sZjk1vJZxZ`m?{uSFPm z*M#l1RBkzlm_(zG#b|~@8A3K{7e4K`Mz%vJ4bPuriFq#ok{KAKuetF*6vgT@cqS5HD##Zqf41PJ>61!C(22ffMgW!vI?TH!7elB@)gbU~I% zlR}LMt}{v=lpVN zy7(Yq4vnbi`%xc8Ps60wx&y6;?AO`wZKCxDg75SpNXte|Q=nsN zD*_nlbm!C8r*XtmxuOI1L>sATQGvbp5%97juw207S$AbS5Zh|8(cd88HB}^n0h+X% z$uaoxdwQ3(ZF0aO>8_}2fQg8UI2CknrlR5%B8qpGbw8%$2lb6+IxtK70bK#@G#!}O zEsgNN@e|J-d62T*2OP>ERp^E&Hz>3LUGV#;w1H7K_cNQW1U$zq>Sd7KaeD!zmQTPw zuPGq`K9Ha7mW8gm$`)UM8WbD3yl!KRdnNAg$yg8aVU;Y{c zXtE-%0Sq6$jX+9W#rr-2e1HJ7*>|dlPz%UmbGB$`V{5vDJRZznyHk;1Fi<8GY|K%* z)iVvFl*cB5!NfcnByxU(IpmAgTl_3;S?%izH8(PY(}aevkWuhC^pw%nERh5>6>jOP z*DXMsCVdeXe8`!;?OA5H028MI%FAU7IRqbkOWdF{8~GiU=Y4edD6Jc!eLZNk9zZ+* z-ZDUDRx_8VBGvVN@xKvm;J=ynt@--!px)JfDst)$;mgS#%r$%vAXC)1=|@Z>4*pXA zFqa9%A7C?`L+%H zk$Kg?L>@VPD?Aot>msf7IZ3n#)Ckyg^>S-Z?1#lOqO};{KW&m5KL;6 znUzEz$3^{HM}O#&2FcJT7~_=qBv{0rG09KNk02geyN;ZPf-57RL8Q=vQ3S8*2Q1HFWFkq=jq(UpNK%Zw9g7Q6;vT`f%3V9^@)%YwDlV( zu9E_F6r{#cKqH^7pc^2~$}C=?-)?ZlV{Z^H7hkY{a7qqO4U%!-^bH7Ix~XOU0N7RK z^92_9527Flgo}q{E;KN0Fihv$s2rYrcN!og9JC}5XV=!KyalhWCWAdc zLxGeSenM97k)dx5ajuw3Ctm9**`~s?> zIax|Sr!7x16N%Kj?ituexH`^oJ9WhuB@%@Ig7aN#%nk0eDVO&a7Vot;s32kY#3upP zzHj=r+}`PSK4%GEmN08Rb+LUo`-R<(!$%R3g&j^fXW&irNzfxI==;vuXBlwEBce6B z_{28iH?<;BNzl&zt$@-wyuHdWn#D6~bR>zSg^%~DDpRD}5|P2Okc^WzAKv8nSlVP_ zVA00}&oQ{b)rMO;M*_qY44qk7A-TLK7L#0cjRFPRXLka=u5Dbi+ROR?@;*-kW}s}w z>@K2J0kOroHUeD&t`JLMSVpZmoDBcu#9Vp22?IR4mCR*z&U#7$%~$}#*wv$N^nwY^ zTjvR%iTowtCsf4w?~8x@g)-6W_CdB+VT}JtMc+;Tf)xIXRP?%Pf{pX{5v6~Jie5eZ z5fy!R{4ehM*Qto}ua7$YCsf4tyZA@^PaE*QALrksqSu9eY@EL{>EAKnhdSo}f2inH zspS7%ndrN-UtjG9D*BE!zgU@=ng8B^$N6`H`ls?C?5~S6zW;Umk&3jVmRb-!Zz|V@ zL7f0KsfTKe+EE5qz|1vB*d-eA`D4K~a^ADuLrE;ydU=k)ke((vP0@Xe7ETl&{qoU% z^z7lNwdtN9U9kW$09AFFN&{MxWkWL1z|=GwqK`^9#1ZLLMbhco_FCR!VAC=k6cKMS zDvTz?kLB%tgm;Ju8sxp9lPsMeyrA3qy$M>tr_h(-7x7xXW|>+FboBUiO-a*(COt^}bADK^$^{4h_+Bduq>)TP}3+Z8_)nBKcc@`7t@MCWXRN zbJ)|!^?MpYB=7Pz97M$&HTGJBzSP(zF^4)LmVaQ7O0s1#qLNU^q|CJlsg#>9FDvVI z^U2t0k2ZTLljp9$uyLtXbzw+!h!TQa+F9bS_XGQFWeH_ zzq2x4*VvRjjeDlhusjT3xr6{S$27RjgR<04w|O?Cb}Be-C9JMF`HmDr){9Q6l(y29 zqx*ojWwga?(^Wp%=A1hjOtPHxjuJKim=GbEg}$f*6Hyc}eWI-zPF@$y93MaufKs4 zM(~_covd|*t^j>FRi%?C;7Q%#Ou@lSLRm|@f3Qok<9$#`0prk_!feooiY~XxosFAj z7iM&JH6I32$@cB0kRVAMAnq@HH##bdzA>SR{xK_?if5<1fO1Br$;27LWa&;?F)LdT z&6uMA+~nzr+2A;7FN1SoDb~^b)xaW4trgBrTox4Ghcf9%ru~js5?TG1$7dXy^SdBB zxpM5N%!!wFN5&qxgBp-kJZQzFGTgZIp-aK+l~&K9d4On}v%L60)1gh+)NIok&&syK zYurRE;&153u3`r%A)!!ALq)DO+VR(F|0rONgmvTaq2G+FW`Lxt9Q=5q8S1MVwepv z3kc;8nTLrpS#Uaxv}yHgRo8DtCEXNY4Ro9NVp2tkkPFIYLONa)aqNu1&dxBdO^8y@ zBS43%X|lPlx}SN_F{d5ywsRsY*);A8nwy_&BNkZl-iOrI0JqtD@^)KY-fX+R40eIE z05Jr1IBzz8frA&Gy}?=lZ4k?DWHaZoKOL9Jt%e$~krZ{4fg~VV^*{ zK#)-;J|<)M5qwjqQh*3O7ZZs57;Ilb6y4r>A017fBD1*IhMppQ5AI=uENX)wDwZLl z4{55)tv)50kA8#Y(J_v`halv_q20@b6i;_lCGxf-um_*I($m6 zS!=uMgKT9l$w)3LWi5^gPRa(3?+uTNBIEAL5<3fKj0BE3a{d-ww3porgBrNqa}uy| zA!7f^p1!;G;j98`@l$ET!ZPMk^(gmp+tX&dnbDL}Z^<)6#urx;E^)q6y{s0)4UQLo zxbaDdTrmfrvEf|$b`WtZM^>0mUT0cOPV2jc22s8!8_0D2Vhu4KD(NzKrKoe8* z^%JC*WdyTLG zHs_uPUKCQ{K$DR4FH>eu?Q>c*?m}Z*-=iI+KOM{2elrju7vAB@O~@rQs~ko`J4t6* zT?E$aO#n>1)f{vOZU0nql~?FmxToii^3>;_)ID=Xs~I04ARf=^s2leT@1F!BfhR`L zR;Zd%$7a%`Nhsk)Am}&T+&&eAvoJ zaM~$?YlW0ve5J#@)T!`}st`4RO+**}v8wRNk#MV-P}QOc8imo=PebNiw&RBHR%`KP?%*7*lx)?(98% z0|EI)@&gj?A=%TX`!ADxYi}rKZd{gEp+|L`7?X`KNdi%0c1|2HD+?jGBD^5$NpyV^ zBnmP$_o=Cuh(2YeQN`LiicU{m@VKCn^#@*0=tGjN|K~J$8owCH3*E>9@rHAF}ixn&AG6B**l+ zhL7R*VV-}7t4mbN)-PI$ua#d_2KffhQD8r^V=o=L^S|g&^cd6sXEfOC3;^+sSq>Oxgjy17)xv8#{GQ2+L|-{0cDpf zPl9N6f0?5i`o^;@k3*t9bdow$x=oX!57VkE(;>C1&>&TxJ`vUt3r)5NT56$Ru0i#9 zpVz{Q{KZQ#b>VYxV0{gHPQTbq-Uz!Bb6BCS6`Aho3$v>`AEZk>nK1%m8GgU3|MD3R zbqxhya}UR`OLCIBd|OqYd3+xGjd$z*SOzPdK}Lc;P_Hi6k?JV#=Mr|&hUHHZN3iMo zXy82Iu@ocoG+XKjAgs!RtIP+0YVWS+S^6Pd%%d7nE>iS1=-3cR-#G z(JnITuoR*B_2SD2#!;;Wm&ruZVHTWUY?25P@~r-sI1ZWNEsVpWreH^m?Z-Ze zXin>F$YV(?GHx;puj7K;tTlE2a&meruI9FViVq&WI_smO{&_dqZtV!EGH)a)d}hjr zS03u8O@_H(`wnUSQ`H0NQ2Pfoly#81zK9=^B9@O^BmGfXx264%B`jzmKjLR3YROI>&=S$F zfO(2$Ibq07Fx~th?|0<*aHC}&$0Mm*OZMwJ+OP62p0BKH8@)SPdCAO2{i>w0E!AVs zA<}}L@%u<%=>|w-<#;LJcFA+1zmSGtP!?DLK6xcviW5AkF0Q%2+L;DR4qX&MOr9}N zaeD|yEP#bKAV_!sc}(nOAVF|Ju7nYu&I>wxsGSqGpP?+XZe1}RG8nehRi0clmPHO7 zDNq5gUMTmJTc?~D3kYLFFv(25bGb>`BRfXul5C^^uX9+Q>Jl2T5r*LK5vwPF>C?Kw z?Ro)6MwJRG!ijO`E{(ZhRfL62AR&Os4D@Vk&BfZ7>0QHLMu5*hrYNkuRFgpbhGP-t z;cLBBsK^CNuew>Y70Ent=NhPy=6@#VibfK2er5rleK`ABK^#%~j^bb?Bso_ThD2Uj zJ}bsKF$y7$9Bzu>`r?S3W4qK5lDj6XYqd;(RP{p=7zfdY=vatEf)(2$6}ketD5eWGz%vp9 z1=u5g=k0ATb`5<-kXb<=NSIPHJ;RU)=CXGDjZysEcpX*&F8J*FE(Ax^I!~li@5YR$ z8!(;O%@BeDXhPY0GH1jsY(*&6iU4;_4J_P1Eiq4OD%Z}Kgi}jHw4TB`u;gUc)`C1v z6wb28`7LG33d|iV+IYQ2Ic^Xl0UY`a@L9b^*4(*brc4Hywd0pz;srB{lp-2E;#tpt zI@ef7>CAGEMP0Xrsj&{xru-nm^>fa40^=42C@XkKZB6Mi7ah4Q-gjjp9p(9+te+#7 zKX`(Hze;etqpG%GWsIR^KuI{WaC$lzS=ZpioVyi{izh=LQqDSs%;AvGNiiEY&Wb}N zwCx@$&Lv#GT*9hX?_N~g!_I~vQB}hQMpv+cb2nCK$ksx>pg>AQ@ZV!i#2{>xQOWMI zq~@0WoGS^5r6=>@CP@unX}h@$WT9GAD3j7C^qWv4?r}|AIWydv4|J3)$_9BaVsC%a z4hyqhm3vss$rbJxUV$GF-&dRK;>uZ=3-Q|M!FthpJHQ~T7rz*KR0a+0bM!Xf= zC8tm`k41bX;>8R#@`-%3?kotZoHnzVOfc{Qvq5};mRlm~b5BBR(i4zn5&_q}aKy04opY7p|~ zO$H~=Ktenh6p{JTsZ`!>7(h`df`G4m3E-yCBg$>)L?;@#6$}@!gqh?i`6+d@3bC=F zmMKi%onR2t_0mEFHqT3RZTq>GJFKNvT#5&(u8aWRIA`fp9Ia!x)$^gQR^QO6S#aNt z&>5S<#bOa+!Bcd0d6)q3Vh;Xs>neIVM8&!&ro^!|MP7LHwm`qJKgscMW9$9nRXU=D zR{X#;gs6GQwpPfSoz3mTDvEwV5zMZZw9F8V+#pJa&VzEta;lQ`9$1p>1T$3+$oyg6 ztZI2|bs+|;H*e(8>+Qfy+GKp*Lgf;_0JSA~YQk#w>vJ#Y!8%lD4W-ou)A%@rEx}zn zfQ7tAP_MmAR~UJIba**&^N3?v6OkKFkc<~jquY?t-pYUJ5z$dOn_mTMV?c%(@64V% zVX9oykMTEmNKTb`vRJ!sQw5Q#`A7`vzSK%)FDKL5S|Vho@tNDA;;Wn-zMI^BhYru! z$rIOASwtRoMkrF9iqpsr7S4(PlPKQj06n+O6e{U$e}N-`4eg00ZnrUd8t-t;?J>1a z{hdN*8;maJ8H$ioX+i2bCv>xZ+DE)klUgO6Q8wB@8xq|Db)CY|rtf+QC52G%E;ffR zFErMrWP8dZL~0;7z;m(60*hWGrMo}Ni2%@nGx&kQL6Xd0lfQBGTDLhA?L+`|^QDUe zDWSW3XT>cbKspLL!gCNVa9U_F)9fzt=0?W&csNeeT`RIUvLcEz#$=vH$a0i|f-{a3 zwL4w?VK^=GgOd*$>$p?3RL4}sZw%Lc2cpfD%&7gOmHbQ{ zz%m;2_Nc`;>Er!nX-&C%<=`~KRM{_e8R%}G(W3F@Ho1F-1g7BQUVvtImyG{6KeB&U zcl`4i|Bw9mKPrKL3Mc=j9r>CDvon0}G5b-Q<5$zqVB!A;C|^&={<=^4dmjF8Eiz|) z_3(#4`2)HC5j_7hLilOJBs=3@pJe(^fRg2RZ4NC2v7dz3_v8He@cIX!WMm>>XZ#)K ze+SC%rk};g|0z(u9`>(-{2v45cW1v|@du#%PK3W$nb??qpCDuZt`qXxbNsV#VQ2b@ zHpfuxk|Is(3tx6+;-T4+Fh7&OkkT6|?1*TE z~~`8ZgPK!G)q)VdR>t8$pnm12t#_K z2j3B)kl1Fvi%s|8J+sF-Hg`uepR|3;OTo&T@KpIo`mLACbo!MF2E7AyU(iy}Dw?;a zny`YgIVz>ES-(U!f~NYGaYbnEdE~Itj&fqJBnst8_XYSwhdk9vrYF;TDg^}*k@()i z$UB=0%HFvO8o_GDmaEkCacyFY&+`NpQw_x5d~G(0t!QlqRp2v67(%yYq#A?KJ-u49 z#=j0oCYqy=22tkRS}pI17!J5qX)c*ttNsmGP(8A zYkHJ{HzZB*TtxH@B8L3pB*{n54}jikFO@b8HR7Di$L-x?KWi?}$=T{MvIq9bU=*&5rSL9~ zMQ$m3{G04K2&PFuU92X+xG#GH+#J@25ocf~UFC(%1bmyRy`nfJNp`#^Or%GVxSy7t zARs*i!N8~NCJ}>Y-+U@6-ybxmipMA8BRpt2V^%gdwc+~zn0v?g%(HC$JGN~b72CGW zN>XtuNySdZwr$(CZKGmUY*mscT{HilK4-dTrf1IQoagO&vF^S1yZ@)%iAqlZ+iTsqjN2QwIR0xM`Qsflz0~ply6M_#N zG6H@K$r>6)S3EXlEfBe?VDfd(n$9WS?6P&C?pp#D(v(?3$pBF9I~?Fr?oLYRa)51 z*vO)r*5Bh;pd97m(J&&kX*HvPqF2nHuD~9SXmtd)BUju5IEOw|ucDx|YI=iG9&GS? zI6i?TzJh(|G23_b0}jJvp&(@h4Jv>V=kdANW$Ynz%wa5*r!JJsaSYrffT2c!ArB$| zSf94WIb^r*ppEBq2aiF>nc(NDd)Nw28&ZNAcsn;w$yLikWZAN-5M35go6U)Gaooz1 z%T|t+6cx!C`HsNChg~tm>HIX7P`PPSo=YWRh_fR*Bu_6Ak1{wX<*Lzch$xky6jRiFfll6Sc8D0Zn5Ti0~IYs3!U8}yg%;p;7;^&oPVb3#~FMuzj^C_S{mp{B9 z9h-p~x`JPvRRn$W>7X#}2t}_~Hm9Y^DX)P_AdgAMD*`plFc!gOvyD zAN%@OH!Buy%W@p7$zqV?zai9H#2M7H>e3ZL@9JaZxZ{WT8HVr~n z0*;*GefI2V<$W@{>L|6bRC}Bulfa6Cjt$%et0!(AH)!VX{LS5eexI^V?y-l~;LQX~ zx&~bO%2-k&SF%6_Pv0CWD>x(iK9`Vi5m7Dzo$$fdK#{CQnWCQ9Zt|G;9?G1a8hOYa z4IQ>o5G$HO#?%i`jk~7rkJ0SEFh>4uGtk^GUAw+2@nZ0(<(>ES}U9K`RiZ2J6KF60IT0J0{L*6Siy)-(6BW0~| zD>Sf*T*tg*iegqAE^vp|6?IlB=i~=XIw`fup%_R2B2K~%F6H>gm1q?P>A9mqN}eJj za-7d(dOb)I^B(+3$n(Itf~umNFbuQYhHIE=%XChKkwgrZH+H~Zgbf{-A|-}sjdv>P z--4(=gMNe<1(ieCZc^=+Hx_4?cCbhb{4j82XsO0z3KZds&}Glk6is0he(CrFcpT6` zwO|d*Ur-`Ft}d}v&&N}{ozcBp6WDqTL9_?j`s z74Bm0M4XOKJvnVOsHNIuBGT(yHxlybZ5mxcK}f*f^VqsB7|He2DA^+=4k} zA|F@e)Z><#Cp6-_=z&-lXv{^nn2})S#6Z$%`$*T* zBOEu%-t8QNhqgO1KQ+(He^_Zn$B0#(!nyEBQRT6H0aNp`${a91Jp%WDZC)ofS~yw+ zErezx_>9=%r%K^_yzXpn(r_TUKrL{MU7hVgNs@hpce=@mH|I(p1MMbB2Ho=3Ox+pt zgJk=@^n;blqCQn967zBye{nx*Ned@$9a-Y4LxDlE?E#%L47V{$uILNm;?1ODSSb% zV-jjH*pedqA!y|$X&%I{d@Q73@i4@*P=)7sS99=1i6PMu8nn7 zk*Q=!U`a$A&+;^6HkWHgRN}_D>^~@s^hEYO1xktK*x{KTfl7BS*IRW;uouz&aNVeC z!frd&oMV)2D_ogXch;)?3d(wnm5_Z+wIHAks(cc3_67nk9- zk?XtL?v#ZV-#3h4>Q|Pmy?WJ;lQ>4QuE1}F33g%fv!fvt0{Cfr14q8RHk7aQhqoJ> zqjY^fixDXbgtd6_qu~7>?EI!^@n2BS{sBw> z1M?N;_i9TvhM$Fq{{@|YhW^R{e}t3&ds-~>dvzll!>`?L|ArR(ONKu}=bwrH$zJ~v z3;h8)|M{M`{{)@sf9*BoM-cwK+w_~7&wt@I{TFn8*Xr3Ae*JLqH+24^PY(a5==?6a z|IeZG&t!kO8oxp3pX~cfl#!k7*TD^JjQ^tf3LE2JkMcKFwHh*ZKharjOrZiv-yN9S zHT{H)BQHiL|8DbE9|C4r0}#X9m>}fw_PGTx20=toBYxBu~y!XNXsxsnpi(xecUtirFgW) zwzHL#N|%>K$$WB@Erany(XvReEyg}=tMEIET3n!_YiCWTr&cH~K&&YhSyUS@T;^_8 znh9B0dLIE~+rW~a2l{yCoO5fzd%ANgB_nLlgfxr*qlrv>jqX%U=;7KWV99Uh z+##)rYe{^qe9p7wIqcG{&L*YOj;&(tTOAhX&!!!mLRJXS^~Ik|oO7`D&OkNsyUXud zc6_&q5eH}7%oamJ4>j9tn^iPti%Lu^44({b!UOCiIa!>vRE|(Jom&OGUpzzUwJUih z$wahkomMNo{Zxx1bYBO~Jl&K0-zQ>((xZ8mzBgWasmbpe;i01g(vH^<&|(f;QNV@J zX!t#HvCghQ z44O;AM$HWmfzUNG%m|Id&TgB>S9x=4tWBoHuWVx=PJ9H=;adRK=Hz$|Y&TV|%T%R- z-Z@l3gA*qYB-1liU8RqTNI1C?YAFI;OBR;)Nw@@)yU(B185IqJEb2wSM%iho@C1Nd z0F#(%Jy9w20yNC^Q@dA_RS@(n-ISkH5R{*Du}o|nF9TnRSDGgIO3;P$Zd?+?^a_wv zE-{A+EF#Jm<4h8AXY_U=e9qm6p{&2TTHi@mhs`CB$-Ox}@^mx0^J1RKSb{`Vq>I0U z@wVsZ=PzmOy;Q*tF2lU*+U$!|?lC!P19+sc7=k!Fv;}p-LIHsIegj!=wkuYrLDktS z&s5O_%)$xDc(Kxq_XdF22O&@xI8b3!sPk+?LV+&OOrbQUrdc2mQL+`^c#YcW2B4u1 zqJ8_2kl*Vr(oIAY!_8-X*$I4Pf6(S<4rp;_`dCdy6QyjOO(9I$Ei+&Z@ZuP3okgf} zAg;weD~SX&Y3n7$Q#Cl~%?K-g7Vj@Utxt^9PAhkX?!^c-5>#S4I-{PECu}c*((awf z$5|v*A}dq=y+Fw4q@0+=?*lNRfCz{ClA=zLVHo0Eb`f0SzHU)D9v^>x_D&bI>v1-B z$@Ay~aNlJ}W+)^IY+8U>jMQw_%?YIWfdU~KGh@&)Jb-vCSdw1=hPo1L8d;ZT+`i zae6d^y4N19v0d9fDMSCZE-HAqlVSgyGgr$vpCi$cCyL)Laph2xn4)ah^zw8MxGav1| zu4qZwRyP>0Qhc2(!Gk|D+x4nhfI+8(N@BC>7)y5d%3)}2*1pq`i4ntL>%S(|tKnbl zv+G@l>av-v3JtR07ZMq-T2mq&(p)dydJRaFjG~{~y9E+tNc{%B4Txf%azI_xyNGTd zIbU?A>n;#%xk^mw)ky>g#|uNYSwet0!TVnL0Q)Irxe->SZB?E@*hnL;ZeFMH=Jbqf zJh8w*v?im^OaI1kfETV&*4?g;nU(&t2$j|l>D!RRgg3B@vLJd!CM;OgZr`i$=@p4| zht-<^Nmdgz`8Y+uZDax^bIg&(Dy$9%%l=JDm)Fz&HsUP_EVk7xzNi0{yq+4`bj^zC z?HDnCuV89!jlC49Mv!U8F(Hzs%=y;>9Jf7HQ2PE+Vg~%>ta;?sOLZBiGA1c5Lw?Y% zM7Zc|IA@0?Q4SF#lk$S`pcC!p1MaeK;L8I!rV^#GYgX7ira-&Xxl_!V$nzKR#dilA zL4fdkyW>to2uWO(yRBp_{z4@Xg5dL^tjm#XpCmcMxgj~tn{R~(&OV3tCynunw~(rm zk!4cA!Iz#rCJ8S;e~O_&6X2Y>hn2$5U#I7V`3AdVqx=eLQw4#hAkgO4ixhN!8$Kr5 z;_VbmKt01eHh~><4|Zmkaz7%z8&I4z0bieta5@{~|Cpo9pHWPi-<<;4-b*b1szv_# zr2iTECn)s4h<)#lH*Ac*w)6cv*!P~{kFf7&;(xN&KaYKkf4-^jKfu1~s$a2h{P!lQ zv;ADoUpM(Dv5)!Px`&PNSLpuR9r^#G#S9zcpU3S#R!n~;`^%~Q7W@7m?=rG5{Y^1_ zZ%h7{EN0l4{?1~iUgN!)lJ&g=w{mj_MvwaGbt#cZCUqSJBBst)#+wAbPI!4uJWo)P z>GJi7Cou-cO>`npi6(>ySxe|!q}?Qk+GEXAPe%Km2VIy!zR_e0`*kucs@s4^LoIdY zBMl`|!XO}%&wQy_^W+?r-c9DR4!0lg#gTWWwW{|y z>DhY`F7tH`_qdObk=ZV4GkzgxtC`y5cbTScrjr6)kESP`IKkU-{qb*h^`CM~+Ah^i z>8?L)hvIIV8~%FJ4ctSswzZw+ zerzN?rk;jlPcG2<;#maiNn(uEJMpO_v;U+(Fy1gUWDROmvRu0$9WkE9tr)fUAfO$M z=j-!eDZ2IRcvFcwTNX$Dw;-)`n6qM4?VTUgCFwd$`dkW<-giwKMSS+`?nQ%Q(nG{}SdD6o)k-TwQaW`Izb6XCr!C!0l zL$%bjw|CRighcyc>v+tdL|j$|_H8q(yNR#hItukg_W`=~B52Ulio|=XL{PK6ipMH; zM4)Kv#_4YR5#PE}4SHEOzU5SyPN$exS3=JDY>NP-prGN6MI02AM~y_GYlf&l_G}P5 z>*_^J0rjc*KujVkZ-e8gcfK%k@p!iQ3q6CjhbXGKSrm`?B_jZ_DKoIyf@qLTDfRS} z`t90@u=U{g&ptz;A$^mf&!<~nzNT0Ke8m+`ZtD_sF#c4RG}1{7R7X}X%H=jo^IJc6srjw>FZzQ^%o;%8v$ChiHR57gIU z=db(lCnnF1IGVb~eS}Bzo`CXR2Aga(&Ub? zDXgI{ZhmNLFUOAv%tY5i92cAZRNwed0+Xjbh<8c8_dX>ufs4M`eo3Eog43HCE(j}_ zm6AUcZ!0)^N4n)|K})kM0pz_hs223ilz#+Eop7_C!!?g5_2ZbM#`$eC?&M{*SehF|Sj?g1N@nJt- z*^qVtetk9R!E}qJR%q3h_)W^TN(5(ABhr1@AzW&}OniK}jV3ueHiLfVM%3NkVukbG zR!_@`@5HO>E>AYeFNVa1!zLo;I$fRJF|x8CgRiUy^1Ax*EjTMDIV*=L6-UfF5!4fL z4_WAOkio57rm@g&3k`e0JKEvvp`S2X1IKc-I@-{7Er)SR- zX-HXxoJZ`Cl!1)Qm{0l&G0tX5z);X3SUeR)E}NvuvvZ^595Nn4%j)gfF*)p_C-PsV z+8vdoBi6zZ9PE|nrUK7te7M~Sik09+SZXsX+~$U2K(ndL>;`2-V5S9ii#Ev*vmF$Z zbVH7=Y*1|x%MOe`>};1Ypw!pIOaPYktAU5sBjZA&0e^p^EVhZb+kfo7Bt0~4Yr$%) zs>N&p2zJ{X%w56`E?t?}swK+@WFVsTnu@3ztdgX!wUJ^uqQbYMC3kSU^1(6EMHDg2 zAWAx-Coplu=N7no2*5$t#S*R~{K#wa!kmato}I{^`t|V;zuPJ8F$2HX1MMvkqZR?% zGi`9A^wZ&BBzW*GkjLTp6D)EWvA~B_R9i)A>VRn+==t%zXh-&umc39FJ8Q=8%KQrU%-utbS-#d z($`Cfy^C$YSdXZsz&T>N;GXUle^nmzlzVm{)yjbhdG)zS!d25$&2>7>uUDs!SnS$j zt@Ey0IMtYgILOjk*%IKlzdKDkE%x*Vp~R(G8hsS3u38V13cN1Wo3qVQ$+aFTCBm3G zlJmn~+vx$E9`D<3qbY&(TZ)!~EkDam-WU`PHrVqZH`muNtS;w^K65`320`vBG*pX(M?Wrxy%6YEMfOX|7bFR?EM2wgL0cdnE0 z4UOb$w<{>5>Lj}kh5~tPB~F+Y6*K{u1Y#Xb*|g!R93$K9f=>^%f*{NwQrI$A-+~YR zrBjJd;;xZTV=u!B87#?E57DacsmO8l0c`ut>Z!)^CN%o;=T(JM1#w-pe_Q3G+MTptLGYvY<2o; zQUtj^2yUzURN?L6CvewU`i>1*9#^A;HS5S!OX+rMm?szQ7rAfoX&n|h4enY{8Jgh` z3G5NWj8!zrWxc%y+*am97~4b+Zma!v@}ZNG8xJH`>d2sS}NS9*1aoVVMh zzxwyJR}sTL%bdrHAQpV3blIk!MrbvW9W@?{SsTsS`+9Y8J=h3HQ;s~Y!wo^h7TlkQ z?B;Qh#cS!A=IV*!;)-oZ!)6;Jt^H5&+tbq_%rc8+3TNYL8d?77Wf~qAOqX4#yZu%XE)0B zb2)$A_k1^$NQ zexlefQAQS)zX^CuzmZ7)(Vxx6{CBoE^%|>ItKz>*Pq@j2QOEs8_guxFGiG31ZIc#W z4O|u*h_)QVPPhz1clU69pv*=&jV9s-N<-GQRHLpluW~R~ygk`3cXy^*S6(KSz! z_lRf_6Vr~2w#ot>sxcPnf#$+mO1k=?gQlY@^MO4~&}v8MX@}|t2hR`CW5-GL!)MXA z`PN#|L*USki!CdCt+mC@@Ay7ZtqKS$4A4TETorqu46W2?=MB^*{^wP#r`w**DlEul z2w#ghEi#i5+$qfSecA5p_~y~`w!p^L zp7~_aR8tz?s?J-~c3i765~0{WZkWNUborr^mT!CccAPK8TJ7ZZ`&nc^ofC8@R|mQl z95C5uHhV7wJTQe#zu2N`9dNxmpg@3K8@ab1NUTzecx9RXnWxR%HNf1wCWlXgbnRY& zR=dJ?X*OITx=u_N5AJ0f9GDvYYc}qZ=OGXyceINnPil|=+&&@7Gl0Zx9 z#5qP_tIXwX0JeY$O=^LP?1KUKMXuT20sH9t*`=9YoZ>J{m);L3eG79VHahSz(r!4 z3-E02SaM1VVIKf?X_iEpH3V@)*ox1Er@Z}vlzdbGR5uHa5;$+NF4svV5d1v16Bxx1F2=NgwDo=?OY@u5%2gZzwfDK-(Fc;_U$OPOrWO$|+Wnrlbu*)Zg+)L4l7>bskim!u zuZgamn$N4iM%6bVXzGsva|yE<>iQLY7QU2_YSgXtTM~72crt;}fhxxcUurO5OJlMc zoG7(&Z5S5<$8dOsw-O||=(=`a%y63ZSeH@M^fvO9GN$n`X^)v}>Cqe8$u?>Jav>6q zFC2(=elX2a!v38yGEc8XF%r{RF-gH7Q(}gExnisVGio&gYSodLZ6T)iz+(nb8a5Zw zC79r3CoWzPKiiGMa_131)R`E3R+p-JwX$>l^*J#z(FrL^S^PGF>Xz23)r4`k-8)z3 z%#Tp~>%@%HJFJAaf}IFsRc2>Y;yKi$1+7N=SKRp^G$NkEKY(-UhrAr~xXlO7V z^k5*|DBdIndowkNb-*#EfepsPqkT}C*6B1-0xIVB*TtWmHIERBifb>38*>+{TLD?%@W;!~l3Spra~ z#qD-FtlV|-dx)n#?s+>sBC2(4Ta`1_DFrSUt5l)c%#3W&*mZvIIx=c172*?BF*YRQ z$$%N`N4*9k7WZ+5ojOA0j%5np{+0jLl9cDB3RvPu=w;8kwL!I1*)B1v68Pf2%e(1f> zb6D)*`gc^jt$wvO@WC#JRR!Xntz)+?wQa9>Y8Yb&TiaT^{9bv~Dr)_GIbznaodBm| zI_##;y6HLr$Xvn&Kfr6CPr-0Sn-ag<8bekFPl4B07efNFS^%wvnrR~AFnF(fW137Y z61l_za*%wnL&?0(mFGlb(VP@~YBNLlbuJ0Fs*gS95KJ77m#c?BJ&C0{V_ahyH5Gkz zSSa2@>vesgk~!GxGO&C^7nMq_bu@Urn)nQ3M0jzh12iLbDZE|(%1)btGWY9*Rtq_x zFrzKCDXk`@3a}Bmlkf4O+mTR~hZanee-y%6XdTRA*D|;wOF!A{2rFixQJdxra;~0M zT$FimS_WTv5HeW%6B#%1y-%p$j!N4`YzK{4CUJkqNIW0wT$ zXQa>|bFsiU`o;_zdxxZu|6WPWK2y^PFB81$SaH(bv+3J}W4@|#q9v~q2_weoi=>y!{bpOKp`nLs}{}QkLoI}pc z@~+A-|Gjz6&(QC^#Q#rv9p5GF_f`JdZTIic+IxmSqP3ri|H)qe(F6YX3pU@UZv0v9 z_)ln!=~wfd_}{ZZW&63Dzi#qhxJ|z;*kopX*G-v!W#wOK?Puus&h0;-wf7kwfA-Y= zSa@N%a4XypetzV*yEX;q?-I#yD#J_IlAN|>EEPro(qdpq5+JLlvOkI^$9Anzi zH^>-~PF48*X{FlANZ$=GI?4~`-HTDa=IZvInV*NdH(8=UyMYwymm%W3>*3GKJZ{&) zTGaM~RK4GB%u*PX98O^2mlwta(Q1sE4MpB(J1iOBx!fhM@7U<>{xaKPhe&RRV3bMo zlKj{hF_+Kvd6(>lAP>}9t(F2N;jHvIYEMAX&_%h1`y-eQQh~;>>gmb2F(O@6Ksjq| zC0kpyIrWL`p;tLdY5m%#Re@RL1&jLm9*lPUcCUaf8X90mZDXN_3@!Ko{5n!Au^RmVR2FgI5yj(}N=$s&t|Sn$ zk?1d`B(ZQS)WDQPj)nf}Km0x@Ok`_~s@j8KStlL2kS+rE@ap zd?VVItj#Xf58GR(l`S1cDRnWaYH9SOqiqn{If1t_iM4sTI(i9o`b?h#2I=|o5dGt< zzR1Yv5&{C_5_yh(>@4Yf3Nh2RCjcsl_|A`Ve82^Gw9I>hl_RFKK2Ap<_zf&jHFh|} zPz${hT-tXZ)m^%0*ZvA4sBJIZQHtSDemq09M4{IbR3t7wnWIKX)iku1BaFS&@J(Nk zQ7=!wR~|^j$0;D7*LB-*dB87$kGj7C%=q%_0oLTi~?86Jq5>2XDBH$cd zEWtR(-kE{$SA$$=mY9=k4u+>_l>rK$EUpL2K?@HWF^LS|4+zsW<|wK2;znI|dA&IK z+JRa2t@+=eh~6pp;dsCS5qa&EUyjV2_ScnCUlT^iLdg;?js@hMAg^HAQEo zqIn1<3y3+R#aNnG@svA?S*>Yx`FnCu`_IgtWz0J4az)?yaeORocb4spK}L+poHI(^ z2hZCLW#Bt#!-Ct~u zWYjO*+xKz}5poGXP1`GBs7TU17>p+xTOQ$gQnYaw^#ElQFUYnL(}H9$hRv{%St~q) zrsE;3gc2iG;Y7;^s3}DU@Qy{z?VTf29^a&Du68ZDKuX1U4nRC#8R=5ZS2?B~AqUIHeOdk<5fxyx~Sjf5+~GMZT(g z2&*ooGFT{pDXf8cFr9l>c$!aU1ByGy7%aOr<2@nAVVnNdc6P(bQV& zkTOcrqN509%6Z*b52H~Lhd{E)C{1z~ryJri3NfNOKv3rDk=v$QCh0&S5tUR!mH?D% zytBQh4NvX4mbm%ABWQM!{~_Iqu!uHR9vpL!y_9^aEgu2r2x(}5riR)sB6_}-8qY2= zrmN)hOOfGc*nEsK^JZLN(iYkn-07Zaa@)*Wbh5g47n#Y}&SC~*^OwX6E~|iwtMrFF z`ghZ1C`A{UlRyXJ@iQ8E(<*gZ%9}oL8tmk0NU^@^xep6#&_t+eu=Hg`6A9t3e1r(H zKPcYtA+xOO!_4~}M$c1ZWEpbKj_1qSInx-!qY`N#LkqPo6D1N+MhSHG2~(a-99@lg zWJ8FuB;U%yA4h``KPVsyxxOT)69|^3V>NWNi(CYNy`a|hA4uW!Nkkh*27rQ;3Ap4< z4EIl~V-YN-HSvSSow^WP@Fin{VE6|rj^@BU9^orD^r+Pn_h^Rd+B5GmhAu?u+JNq| zkK->sw~l&`J~&&6RF2NE^PTM`$l}kBLMAe4hN0pDH5CRDIyZ>cHP+YabC`|x zy_)vz9#RQjFzqdK3HqJE1NVk0VtEuzRyp6Go9?kbv|`Js1S759+c8P{FqWb>})5j0`+J>DM#nWu3MMt?o?Q4JV9@WByo!8E>lb zVEjyzH}DTZ{VBf#Bh0_4()t&kkH3YH|K`NXiFO?^EW#l-Rv1A9TOUAOvhIf4+<9KY$Uy61(98 zC=fva8<*cT^8C4)zy6>9!e#m|EW+}x+p+vgg1@rJ&(QB(k^g{2-hb+UjzxYZ`^y4< z!y-Rn;+H5R<8PFs_X+m@8jG<0ooh>qtPKW2^Lvq*svB^ zMj(@O{adjjV>i~L1^tO9dG$vH*a6-(tEF4MbuNqfIBT@h3rQ9&V`VznQlnyFx{WpM zH*CNP_YCKZMA4I$z6c|?jU-(uh1&AZBr{8J@rNp_G|5N)XJ+Yc^XKZ7BA=kjIiI3y zkJAlT8>o(o3s4;8bZR<`AnK0hjU%0tC|n(!Ivh|)t&Xjv6Y8ImsP5i4D8h@RG#XKo zs}PIuj>=W?pLi?@S+O;4ObC|&ury0gz2Ue>Su>l)(Fo&&rm-h@zYPZre{BB@0|XXk zSshzXB@AjxTY@4=>DuksQ{@>*_1SL#%5f(_S1lZq6e*DBv$v#(OO`%BOtiFW z&45%uPd9K}3x6-ll5nyv>d7PlM5I5}isFj>rKPa^Y~|q0bma}O$$5Y$?F=3{Xd*pG+na(*P=V>EDHT}Q+$`$D(_%;|_!&&ee*hzc|65;3rdDQ39(@uZ4 zV@QH~&d*6c;bQmlv2}A}vx@z;8C^^wfoH!8e#U+i*H1S{Q8S5RJOA{&J3l~JGy$WO z6?7ILP&14qVB1SS;I?JHL1MvF{q9qYJI)@S5&pA27J+&OnFdncBq@TJ^R51cnkN=_ zs34ln35Mvjzlf*y`jZm$U8;tnbiF0 zb+%Aw;45Gj1OwJ(QM0qBfx%P+Z(rO;I9%Kg=l0KFz9#&8ab`WZa zR4nyJFj=3v&goqNk`XhOzG*NLF(SLWHYL(6ijU5`Uc29~yV}toMmIG|@SEcTZ zfDwn3S?^#0jOwUSCMN8PR=E(=lF*IM$Xi)1w0KBj0VyEv-KJBZcp6<;2PLOCyX*mT zsAhRbxt=a5ldhr3nI`I*`gX8A`HvhHG7HP}S{-UqF^N97({l{avtiEyqAgd>$} zltIn-T2{^m`KgF3q3X*u1*hez{)qQ81oyLn;|u(yQ$w2F8PTW2AJnOhSi~HR<~-MjOoZc6TI~% zgUDRg4KIU_>vFR1Bf6DiHT3X@SlCttFj1Vi_{_y?9eK8?coDO9g1x1&9%~h+r-5Bs zCm1ZlcaVt$pb@oYQzcuAiHTgd@rbz+pHP<#VjM~C%&Ui-_mw_v5mTk|@tWpV*K`u1 z>RkggP<}NaE#2F2Dv2R}wv^vj<$AIk40LJdab0eiw+-0>WL(57O(LgPa`{munDG6C z>|)T8Wr^if({`pi`jo2Pq=W(x3(3)|G-~JI|H)oy~(Ft6fBuqNS`Yy-{uwC}i-T zS9}YcX|WW#+Q*)<2w>PoKb-&ThS^ zvP@u&%P+?yXD>}NShQ8xzLM{lq4`PJ4hvw5Ig7cu3ww}{%nF|udQa0YFQrjMHMG!@ zfB$+(rC&3BD|;r^IJ+hgI;R>HfV^SPj&3S>^Cd>`R)lXV9?RRX(iw^zde$PAbhU}A z0@_Cy&4Tux7Rhtt1sY40P?n%7`u-&21M#$ZYIX}h=ebRjcS6i~E<(O2(?T2_C>8!J zWT{c7#nEKm?c6#Fi64p3RPlw{yZO$Zz z_nxk?U_sf9DmrxA`ru?PG2rp=|VBYRsMSd=gNZC zdkk)0!0KVvA>S}T0AI^n(S#ewpcf)c^t1Lv|N51mKGnaK!2u(3EUeo7Q39;b!u1=Fx?|0Y;}qGX4%3vHY+2YP=U)v9bKDWcn|@ z8b3q7_w4<@A*1*5C^nW~TYUZr!U#5B2zy9na|34rj28Lf7 ztaN_YlJe(r{<_IOi$?F79UIH9T>3XO`Xi$J576jcP5ZN__Qx{K&t!ku^Ka1TC)@lI zWn`vj_;oX^|Eenn>t7GzZ>|{S>RQ&T14y2Wl}lpSQ5fGyBqJ@AfO*&xK@g(?m*6dI zNzWjMjhdvmzE_^J)sm1dD$jt4==p6MRcws8*M3C%*q7=w!!gf7)VM8^I#6BZLi}w= z9_N$@C0qS`!5~#MM`^ptxzm}hW7Dx=+y{c~Xp2f}M=#d$Fg!9xL)holVX##2P*j}m zN83kX{i^*;_v&FE!`9)P^r0FD1akeBNeuL!2ZJ2#^2cUIB*()F6*#tu0AXbETpYD6NzZs=l35gZMQHLg-~ql^`LL z@2i0?oq8KoP^|hG#>GqM@`s!nriC)Z)Cv^yA(F!t)8jUAMcDc+f(|{Hgx6~~;mF{# zkgErKh7uL(*|2@Mc*T`9R@C)=5QXrnb5}QlE-6-Rw#yAuqEvAK$Ouq&DS}D9QQM3t zQo@UO@x-kc4Q{Ll1W_#Yr~$O>Fx_6dHeY?oCu^wpdB{c5T@}++h}aD2K_6aKQQC=2 zChrZMkRp*P>lK~o0rlB(`tBP`4Q-URP8Ese7w+ii8mJa)W;#m^z8x)c+AMP?m*CIP z=f^N^jKBqhhtMwDPtms+*Ynmv4CLCx1_Mue%qgX0|RKq3fQ$k*#7f89h z9%hC$Q+1AF=z{+C&W-A_Di>5mjo&UgB!N0e(sm&I9%+aGiZ^xTHyF+g9hh;y zXgy=m#?6aW!lDKm)~8dt#$D!Gd1bjEqEsaRKy5=o49H%-a{+;P zN9Y^%sccTc2d|c`$tm@wF!y^v)glHRQ48Kfv`9j4X2m8WxYI&W2Qf=TO{;3{H|p2} z4>zP0n~!!L#OWonQr3MTE7$U6Wtj?7!>Z`>@t%bidfR676Xr<7*dm))KA@G9B=w(V zcQ<*yU;6Vkc?eTsZ?VJx^J$+kK+#Ff7*a1R!y#KsC035`KzC}LnL*93I`|D-!b-q1 zm)Wb)Aj8NX;LTYc5?rMO~)2UEbhgO^aan6+0}Z$=9m3KJlh{ zF5Z=6UX7#{LkwoV<0QR!FUai{b;~CioD^4+P!4O;Gt;=cby&uaH37d`u zr$iIsqAIF)l1q=k!Kb2f6Xiokmt`B#wpHO4l&c>_#c#C1Xyf4NPiJN)X+hXiIZ<+c z-U>)TkpRT|;f-7IPoyoOmn37#Wls)q@=Vts&;!7#3R6FKGlMqyXJS&R5yyxF_~dj5 zB@Ng|rlH!Qg_Ip)uQryr&8X^RfAKvfHmAG2`f-c#Nr%8jY;R&pkgespeBdM>uH~T` zE#-4#w^h170UPAdENf_w&QptjMeTXx)EDGDOuI^JaaRnT8l`oD!*JKTab~z}otpJi zU73gJ1Jo9jMr8TCjM4j=x$k(r&zB;ryAfvb<_B=;8Yx_; z%+EG#&5`yBx(9lI>Aqt7A=BXdi|Zh8K`C@>4uXGBx2b_!rEUEP)d~mia=oZD>0D{`1t2dz{q*0iG=bY$9hH|)E9{w3;Zcr|!ZQ4E&{{-kHuqGyC0aWxw9Y88#n=)T8Ty&#`%AzK>Be{V-oAfh?+_ z%_MQ=0(&Rb@wBqEH%HO)tb>|3 z?JSBNNcE!={0T5Oy};qWg`HPsS>b|Le35wfO-TKv)l`)I=|n`cCI+u8x~kEB2cPB17M_LOwbFqPuM zDkL-`1sk-m*)YO#oSAP^zum#s>EF`&*cK;8JD!IRxFfr%p7{|o*5I#%F1{3`aA*g| z7M`ba-raAO+rX z$zKxt8+Sg?Zsc5u5?9r-sqwtOkZSl~aAz~*b}@&8eBn_ZV$y&Mb2S;S=I7tU$#Gae z5kMJ$K{F+bPlHOFy!PD5yQ|zs#jncrM4GN~<%bD(&H3Pv0KKg?4F~KUF|l`BWr0hJJtZf23gi zeZ_{PuDz{^^Q&l~PWz7KJ5as;{PqhqDrBG@1N87|=^Q|9x?$9_!bpFd51yZf(N8S_sf{OdLT z4|3w)oRR;8N&a<``x7SlYw^w>ndC1hMNiMn{Btp{1#dq;ZGZN#Sy+DGh`&C$Je9?0 zBsRp?u4#K!9G&kDw62MO-yfMnQ&EK*5P5+&iPzUn^@zT8a_jJv&M!V{pId%!1>j9% zAeOYdH4t;pc1NM*5j#RL89WiJ+`LAOoB<*?#2b#42CA z7_|cp7+mVodl)#0;tvOAOK8OV7syp{1Xq;?Rmc{_&0Da>jX7#s)DoJP11+g@E1eZ7 zt7OX%U#;q@8xCCKHG;&y_qFc4-c}N&G+i$37EUQvnvFRf>ex)JA@A>^rIM>Et+bh3 ztggsInEt{(d0ZdBOVyy_OnInQhee`>&9A=Bo?3czJ@~Dv(rvbIJ}f_slAX&ygS|~z zkMrb-dXn)*-PWtFmu{$he)_;Mm}_C)&2Q8!%nR3#z^%Z70W?(oX*Jw&$s)|OE;V>t zw;=NMEA2SjK`GF4M`i^pIss$rx+k&fm8Xtvh1%=SJT&e&_(pC1|@U@8E=ozuY1AH7R zr=A{>?CEd-S0WF%1g8lj1)Sbhjpz38Cm%`=BOU0%e|>#HmV==5bc8}!*qyw$F%D{` zu>XE_jJc)}0v-_b%!_kP*yuA}r>Y=b9#CzWUz{MxS_LtyT@Fs*B;=cBacguoiA{<| z{7}KRlkF=IVtV|Pi&%taxw1*KFag;f%?#@J4>$)(`1HlRhc~8*Rr0aLn?T&5AO6PU41KjRS_u=8Q4DP3g&>K+y^d!E69U zY+j~3ycCw(@^)Uu5Td8+nU5uTGDh&ml!MK;WcQ6_il5=Yv&{oIY%jS%&R|b-qz)Hy zhlV4-CW-p+<-(0P&Su|rW%`W1j9)FXGmW0r)6VFSJc$en^$qz%uul&{6lpeLBNk2* z)Y1*6J63xm(Hj4`gpL42J2sl)Wpy*hTEGr#&` z35Ft2p^6lZaecF&MmSs=z|vhgxb^#4neGK>dXG}pd{>b|8?DdMCT=;mQ}*23CKRJx z6UA6n^wpqMfS`I5b)Z|s?J4!Zpkr3cO3|3}QkiL@<>V4%ldYf@Ru99yvjF2OVl}ee z%IwIEWqa7WE-H>->do?><6b|58&=C<=b?ungW8m&uyUR`=m`^dc46CrefEd`#Q3nI za5|BtrmD}Zcu&PGJxCOOS`>9y677m7`3yYn=CudC@OuELnyGJBywf$U#N&gTQ7Kmx zzpy+WG2Oy{bJS|+xwVR%)sLOsN{{KsZ;*{bxiFTQ{U+E=0#Kbd&QUB}6;IN51a~tE zS7C(j)T2R{Cf{7w*Y@#CNA3mkDV}OVM|~Fl(^D4g(gQ(IH)JeOO%zU;EY>Z8RtRdfmU+`h26i_UP8`tCQF8e~TlB)` zfnUhO{G-&3s(Hk{_%?1Jj~We+5*?bT|BU9mUCo9@l@b1R>euJbpTo$=D87=6NKDFs zG&4uaWT3k!~%OD@%f#Sqk~V+CVj&>usj)@NFGi zM!)kBiq;RgteXDXwHwA@SmZv!^HDGF6fk7~w3|LSTJ)=NVMsOYS&=(~0uWM9VX_C3 zG6_`hIUA1nX1DuW6YKIBo41mGFA04lGXgVNO((d=r$-4aT(YK^G~1_>IIO z%chO%d8cZc$G^2pfgJ*j6|93WZJb1%&{Yy--b;?0_f5K*(?nXrgr9#0>!>Yfz^beR zydCbVrGe+|5UO`U|F&QI1o?yNWbXB6?x5=YYtRPcRpb7Z!sPnIRj&srO}r7O0jW6^Yzyh~B9?s&4%Qx418#&@i(U+g!ruvh&f zLsxqaX9OISv$aE|qCtu}=z=#P9HI+!J;Bnp(8k{Q_7LWeKc0(2k7o4arPdxhwT!S< zDJzgd1QE)SnRh}&>cdstIbPjqpEk+4vV^LxEYvhG;QG^1QWhhLPe!Qva+b&0Ckm0P z$Drk#(ZTJgf-r}$zW;Jq$m$dmRWO=Q8Fqd~x)~IgPjl#FQVXHSy*t_G{@h6-d9H&S zP0v#|aYHeQeSuDfMU-7^w1gf*6}Yp9ekKqvFHsCqiF~b-X6jB>y8f@MDX5v&;CdL+-Yd`F^w6;}S*LAt1IOTU4`VZDVQjP` zbVgjs9fmgCEXOK2c(!h3K*{oIf8)rn%@NQDfqs>+ndX`X4>*ZOJADny8@ydpE6A1F z1X&rGGrNOmKOlBxb}~yUCR6mBbpUty-jzq$(L)o5%<#}(*1slTL4@wQu0t;?q7bU{ zdJ!)ad{OeNM&E$?OfSXZCldVv>ITH~<>%GHErj=^TQG&6>_5~e}O!j`S_t?`4 zdrWUEBEhm9P7ycm%dU0brfL}{$cDvfg{r@47ky<#&EQ%QRF1(AB&Xb3;0-bl$tYlo z?>$YMc!psghwezL^|yMO9-1)vK+J~`La1n=#0;j&3vHP`Vo|+z-}ctA8yguiqkSQ| zJW?OLv|wL8{uhgfG*wNj#l~0it60kc zo-1xA;nB3OYWb|SgnDe@j)z}j1!wKC*XwblP9q=SF50gJ% z!>6iCrOo1VeC+`)A&jnM)1z?HJwqD^s^+9_xlYJZdfmQC6*KS7=_Pa7rmJygXtl=I z;8JCkyyK}|^p?8OCcu)zTvFnRV1H}jxN3!gBs~#jTFODPA+8hT=thd@UF$KHlYW)8 z5o$_UVQZ8)cl0sY^TD3HWkgD@KmHB>fz zD~N{!&b||*OOMf>)`?Uc|BTVH8PA1)KEwJ|cp>2d7nW*KiJT~ME$n20!erEI5~m{a z;IrRh!=fsBxA{H8278iqTI*+r^nAXkSR0J6Mp!b9(GI{q6z0aJr=j^ro?_w%CBDSI z`9sXR?4D{N>!fLoB=MmYR}Dfuf9C8;imCV1-ayo`pw}&u1?mD66P>pEQi~?w7=|YX ztCPSCrD2s7GOmGXxAPRO)3n-50+Jb*8g7ZYB4G_s!o2mkk8_B1AO+ttYqF0yhOzYL ziLMGXhWuq%;#s^?>jXX+M7(8(!pz0aryPnTr6f+Z5!)gZ z0I#pXZ4Q$e@lIwHV8mMJKyBB85E4-r5lcq%Jmj0^_YG5lI!`p!E9GI;;pNe6fEa(%ENIz3ar1e%HF1S{OOg3K_aDM8ZP8 zBvMY2u_wSB&hob0Y`*RR)@AN4*;2p7P`&8TpFfDA^LYk;;*6R}VE8u6)aaCYkx_I~F?0oZ z0_t2bUhs@#h#6AJMJ<1yNkX3a96zF}QC}iDSr(O)n`#v#xy_a?zccSWFFC-*RLgib zSQ$1D4B^I}rHn?=2>aj#Cu2e#qm@)?G0;XUzs-|f!5AIy zvCK~&v{jOnvQ)Of)MTe#^o&R9O36v|GnO`v-i62bCKZ=};^vz74TA<%qw=rUSOj2q zc}-TzF~ahnG~j}KidQ_tTt0y#DI;d&ZNO)x=?MkSR%SHE=jut?a1vAE)Y_};_{Nn6 zA;o;so!T=5Hp*ycMJ&ccCxfNj$?b5Tj!=L@)(MJ(?Xw!s%T$Qsb=8DCbkW=?(H7+e zJ*|4$xp&3k`ji|Y2e^)pa89MVGQ{;JSs{sxc&3t7^o*nVt643XRW=u2jCimu%FZkm zH7Uo-*f15`hN2egyaUT*!!lKLCbY!;V=&3(1dj*z7t<>u6A?s2_Y^%BBobq7~= z%D;`hO8%}!%#*ZGdLjX{%U@R=8Nz+DiFqni`I%qv$6;)O+qXysh&T^{piQDs3tV#V z&RrmEt0&Lhk{vP}2HLRSC|@FO)S8XIQ-?s9vXtm16ZN#koTLq1FNo`frj><4RDXcV z710+E7{;6__`)wM-z^M&zVvO-5wrMVn+tG@yBELK;sg|aA^U0d!ibZLrw`!F5*8Ry2 z=Ns;`wj6F2o;NI5P;K^uYrJ8A51-BtaQ(Do^s6@a+xiP(xR|78VMB(DuF=px#8#SU za%h{ex1*KN?BhnAfYoN>UL>4bgtHj;qql-Xvu-t|;)A2PB^STbe`KvPtJ;i=ySGn2 zpHs!{P1(H@m$`|{XX4&mp>(43H4ke*Al`q2zZ>z9Rn}>0TQmG=_%p%j%A3staET~^ zgIe|3@Gth%?H@07XTdg47;^_>a39=vZIKPO&hAWobpit z-n0y^1i@(I@Gc#nlHp^Ga`GV0ihgFy@cf9EAjH--vgZ$^#|~gcy5BUS&^llsD+{*Q zH477F8pZutK`Xz&^o$4rED~a#kU21c#G@xQnqM+%$1ZFjZfS+p%8b)Cd>DRtwBZ9VFs>B=D^Q&Pi`gSkLY0X@zlWgS=m7#sEe`9a>w?Rz6Ls`Et-Z znPEOH1b&G1@RWL*3a-AAnJ{5S%~j`)!rqsmMEn|`(DR%pHwBgGc)mtL*CxHaKIcj0 z@$Fj1aR{!~QQ3R+Sv6Lm7l$nzy{55OerShX5yc)!L!G3j)l#}Yi^P$pfy z-a!~LUlIbuoInDq9uAK_4p+M5{fh1WA~yW%{ttgwa`jKyLHs72Su?P`DmJV?(!2iV z-u`3gSNHb+n@q#+J)~HFmI3`cZ1)=BAF(k82Z(|{W;r-*jhPQ|91QYYuNp3bwkIcGkMK29_UO{vw6`_`~lNc)xwA@EU^dN5g5kzo~tD_3Zyw z_x&I1fe=!{f?!-Z^xe<7Z$ePxAKp=;DGHSJL-2!iP{ok z{~Y3X46W(K9C5jtSEx3q@Na^R?IHnbtM3mVmY@bPspcy#-p0KR@;`AtX##go#o^4GKYt_sE*=_LtwlFLfAOSK1!aB5MZeK(ZonQudx(o-7;N1lNPC0_OV#%((tu3YO%Oef(ET^U$qnFOrdqw8h%p3F&#_{;cDB{=dXG^HA+ z3*#AuPo#91wv`msnDPo`8PyL5zu=1i!$}>*GPd&zH%u7kHl86Q7PC;-p^>+wmMca8 zXXi59yng@C<@=yc|C;}X;5|3P1?k~sy!!UF*NIYSfzWtF;>2<$YdEAq{w+LQe6~7Dwl&?BISU*nj^u1pR!WF8bOb)TM8fyc} zj~cjnDu&pg8WvC!P>rS1I||mJ(;nIE>3Krvk@f2AQNdS52Q|UWTYq%-UJ^99-ysc- zHi)FYFi39K+kY5_V}l&5PYuVuk%u2UxjV=y>? z_mc`Q+5X;Z&LZf~gU|~I-@Zt1hK`LYC97ggbEYfWQw?AUlecQk&H?3Hnatx|a|(qC zW(xp!s4&HPglOq7yk#hLm{lLx<2+_u$8Euu2-~omD$zDW5y#RO`}BcgYlO346itok zV~wI%__=^1wN@SDQ=ORjIaU8I#rn%MEt-{o692pWT)2x_syjYHmpU!1y7(#MUKK>5 zaPpuL0ws!|FMx{Z(}fw)-N^aYIgpzB{H4$lx_q24a@b0i4YD7zjbz^^>FZX0mG6Eb ze7uFLrIhkDj-rkqr}Y0!PK?KwSa7&$TyuPbVLHecAa%82ADPaz^EC^hv>Gu(&e zB5{u79p~Jxkb%EH*4fM)PtftP+RDS6F?Dp0M@;2N-==lRkUUbY}KPp^s$&d6J z0EI~_ybXK@W3#Zt1};Kkfj1pl?18J4?R5#v%iJ$Usy1vlT+XT((O`?dwLM@YY2gKf znMC&4URL&5^I4)J7;Tt|%lMXNpBRJRhHJsAN8@9ydl(q;OlBAQk5K9twtx>kC33ZJ z8%OZEN5ZZsC{j$^2=Oo?&wg$%V~-}!s|QVLI>JOr#c9EtG)$uwHwHQqAldIFOu;#J zQm>&T!`q9_7vrv!mcvr{1dwy496Ch*ZIWqTu;nN!PGvpem4h7~if$g^$SDvJM z$W(h}iVx%nog3riAKVKu2w@9Y!`Ft@*rd)xZTo8$Nmuh}xyE!DkedXCY{uR?H5Gg# zP!7}*Jj&stw5f1GZ+mqSs|HK@K)>>!N*HNh)Z*^{xs?V3)H*R|uOtLe%8?e7)nsdWITka3vXbvCXI54o4USf!W%_=TSP9F(S$C=aQvB?($bYL_FEsC^+0Ds-jh4V z7!I3iyWxJ{ZTi^IodG$nx6v=^G}VE(JS-fLU%e+q?IY5$9(#cPVB?IP=UFZ8wXm1v9=3T{KA>bT{> zuf!P278WF)UZSGipBex%afW9{_x(?zjzb&d9LpB(C#-!$nV5O4F_6NZH7*j(mvaq2FD zrl(>H!zFi^k-Z;1x&=o&Dt%s;NAA|UarpL}(NF_1`VH>oOyq*_txy!T4;2-vv-9Gx zR+(idRx5#}a*ecdMPJo)oW1K_8M;!ok|}aobpGs*}d!P)UWY14v= zMa8M*3xk{8&>h^s8}_t`ue+0QBDln2ktDLI7C3=$u&o>o()B&2#*6D;ASrrvQ`N4e z`v;+CAPYE{`mlrxt*WuqK_d*6?dUcta*9*im5+r>zm5gvF_$Aa$9R4S5U@c4eWfjv zca~a_Cy?KsA5hVPUs7P(hpl;Hu3QUDC{B8KyN^L&xKyw`ZD0g zU1|wZj1JM0Z$c~bjNU26>d!WOTC$q`UX8%VZ_U258$c_BkL_8kLbzSS-j& zgl`Ub>8p2R%wPr78{NA>zHHA~VKqj`_G(2rMa}VE~(-QwOUGLg%EiBe&@Ko!&ylTD#1WY5Y2S6K9 zovHsxo7drtsyK%p=G=Jjj819K5D*&48ov|4R2?;yg(we9)07x-e;h#;xn zY1x(UgHL;RrwN|{lYwI%lD>V);u?-~!K69AaLfo~-`WCO16Bd#fq5!oWCe^+G(70m zFB~<<~(+kCZmc(YBy5`i}nV_*%o{goQAu|APT5s}R2u-Blykx?y!awPV?(Fz=)>-vxtgH@&Qe%*};9F&j(uS*ok^`LI?8?7F| zl_@>1bp^IAD0z#}TMum-2Z;)xGJxgElP_Qi;L);W%>@G=g9KGW2B~W<lO6xRG?SE4V2`E>eTBFQncz1&WW{jFEh)G;nj`ugC$leW}%rXvHRG}Z*S?y z@G!&1KT?=x-Ri^Z9vWHJew$Xs-oIs}^r7c^J3LN$c5%0(qxw>npLDM#x4PMyhr6IF z$?(h=Wm0Y7#(?iaoPI`NumIz9Jw*_hrlog^JI^S$MlOBZcY6P6(9|g#*`^*ECWCP> zGlAiDeJzKzv8zaM!1XHt7kdSUzvoGYOMdbPtY#yFp!u5vldvm1^T_N7G%~2mJ^-zN|_!hx+`09w&sD11#idoSB5Bf7Twp6FCPe5Q^o4`hSL z3=$P8p+w--ImVcOETEL{3WW;VnAar{fkUXKBHUZ#1ewuu+8Vt*Jy?c-f$EA4=cF@F zGrSi?xI1{Ilp_g)_{q?)>@q6i)63GBXdzkyBDwbKU%_ zfk9IgIJ9=SsJhkT9ejL_EUOcy`PX)MrT64k#eH(j?@>Pt%ldoi<3DD@euzFVVUStg z47g4#W+TNdh<&X*%+x^hu1d@)ileL^i*+9M8s?<;m%dLUTzdfP>Xp8^+8QL4j@&@-2UO@yXkqG+_`$fXJXOXw49(2eg-yu}o(QZV@hf-ECj9lphtskV^j*C_wE&3ggBvghmj;~gzIsNidt)+8xLH} z!mFa@hS26^b6W0)Uk5#@b#_q?tn}gpL+ED} z=vr-B_SL&|bvxVv$Bsn`(;TX~wSTKI(_`N9A$)tFjfCHR-kVu zZ2udM8w{_xZ!B#8)p6s;(66rLf54f)zk~n(t8QcX-bjh<_p+jY58YlP{3CSx5%|AZ z>%W0+|9YL)f1rca)Bdb(bNywGVI+ zU@alUypiU5I(WSoV}$B*eZ(n~l4^%f_o>CC2CDG}rKrHh`oX2;heh%|rzNKwLmk_! zZx=%rf@lPy`#PuFZV3gd97t^QQ86 z1tH(d(X{}9JwYzk^L=G-C;leS-ETRLvLwNTP#eVWFT^?ROK4IaZgFn{1BMR{ha2Vf z=2Z7%okKUf4~{L7a}cW04jQuJ+OX^P&*U)!1Jb@&JftBYJ)i2gO|~JHMc2nxWxuXXB8x z`*>XhS~1wwh$e@#P~`dv-dTU#etdj6x@}V`{0P=Dmj`bpIL&yAe8A|y(P8FX>)0<% zZM$2GAU4d_b@3rb*rdS}xegnJ;*z6}_cN5zbpD~j>Ngbf%4URXox4y1@ELk$|b znmmZ|+v6UFr@)4iV!K~+x}Lw{wX&pQ;uU%)fAmE$b$muG=t2wB zx9F);S`NLc?mbW^t&rD$wUR>2}4`oixljEZVNk zxEEFpNWp?#wqa)pjvxasMIfE9ETE9cs9VU8wyYk(rz=G&^W4Bzl?zF$6sW8FrlTp1 zT6JA+4;iknn2|SaXsfs8OU9A;`06b;dpsr6ev*YT@#1b&Vw_{D$sc7du5WEl;-2|KLvA6gXNjYXu@}t!>Z^IRl#(Alz z`dhcc0;;CuX{N5J5zv;i8A_ICKTCJE3=*P)B)#gLd%}x#bh(i@aKcavRj zp*N_?I#mGHb8-W{Z}_4TUaK090c}#jg<|7P-d0mVb{6;7Jl4Kh610&DCpg1%jA4~b zz_sO@Xy9nXSvz-#x&fc2&cUY~gt*aEXt>o_E{5!^ZZ^MFo=C7gCY|b7;);nalN2ANPn>Bt;~Lcf6jRfIcNcp?R1 zs_2R*f+HLz`^+_Alk^LGf9;z4C$Ld<91@VZ>6j&YR02t_d9#uNDbCI(NV>)KP0=^8 zknI3kBpJiAk)(q0pU260c690?u$k)2h!WNr2=qmVh{?ICJn$M0CL7x(HjuTuqM@$1 zXgOE1DYL+3)OmODH4e35!;A%v-e`upfjpE>3L<3~a8NMY$$&|I^2}*Jo9tdmlHHLO z_!~Ft`44}E=S;rFZOk?vcT%{pNSdS_En5@`$ zNqKXGQR;}ed0VZ>Q&)Dh*pG&lXbwC!!Ue-%-o&u8!c~I!BNYDv4gu;t@lNmwdQD{L znvl$zS_i33Xz~s}DYDU?WI2BTagR}=g0837#Il10y%+(VA&{Ny@Ms9ZKv$;tyCJ>^ z;$-OSfM(LxA``6*Z%MHN+4c7p5K9|`h1G=N!A|Dzs?sq!L;*Iv-oizIQysF$XYN(d ze6V(CIpPxMCbyVDeUQ6QCS2L|$oOTwf>l-dAKrR=#9hJ+yUY!etiO=CHq6w>th#=R zfWTQ_`OG$)jKC)!mSM#u4J>w}B-3KMi$ROODrxE6(FCT(v(}`0KIIAc2%O2e0)b*q z7gbbePvoybP5VY%x(Ap?Rjp?Pzp$+t4mC++62usD#|1Mzi(Ks11+8==GjvFrm@fIb zZ5KPMGjNc#T&k;kTXKcJ)MO9?4#rZ_VwCOp2BNV~5B7Qz`+O4xF`P(D{xTfV^l(=p zZ-YH#iHldH13j~9zGrv|d9Fpg1T(Z^C z$U(pP+;QNO+vNKqezHV)jVFNCK4w3h+W&qgqn;YiH*%0#MfnW;>DVXJs&$t&TfyoE2d)->>#5Kh2avq(ri{sLFb*&D8-AB^ z{OHj%M4ts=A<)8B{QbB?)oHQi7M4X>p3izm!_~xYeP@Q8Y_*ji!bhOkOz31+=xnxN za+j_REz3P&e6?YTp}0b1qm=WF`2v`B?nqtxP10rH;hXBj>HIEq_7XOwGgUdaaIc#| z0xy;}MjcxWD62|o`JP7B)`!KWDl}Zy<1^2pg2;glZFEKiS)Uj?3or4npn@VuJo3W2 zBL-nabylvGk;sPgdHtlYsfI;5-jCuSG*FSJ{T}i?WG7ikyI%h_DN* zrK5$f5)Dt8WW$Vq{R~UPP-yj9eTgDq9JcbmEvL6;m7ve#b`0Uf(*x7>N&+au*AUhD6Nom;J(MuO>>w*yLZ*QFLL za0AyL#P+mXzIjwFsK88&`w}zr{xKU=>Lto=k`bLM@^R?kg>lc%@4pI)thC>Awtl}B z{}b5oT~PeqaLbVT*Z%i^$-aE`w_|Pc| z`+Y0_hFgD1jz!V%D>QY|>0*wg>UD}@z49)ij{vV7yoN&I@<}F5as?9r<(jOf&O92}ehqs{ss;8)a28xfTpp1^ZiY_zEFwB?( zUKt5V4RDU(*4Ps3uHnofYp=9KrN5u7$YA2P0=fK5<5WYtR`f94)^Cf8rivR+H7bwW z!xtTQRG!wRNu3RvuIp4XKZ`ZJ{y9WxMqDM2V|XH zR`zNE{e_ndvwp(k!LUBi3M3sxbJ$eZuc|rWWR~T|>7|nojbPLVzaPm_xr+fB&ckR; zqO%QnK0J{F_-$_AmEdxnRDw#!ErDCbyG19&lw7DXSVi@?;@W8ZL_?vFNYk)w~%GZs}OAxE7Ua*IBVYbWAiU^hW z@4u0wr&C!5qC**}GO9wzKAFPG=2Lv?y<>S~$4f_n0Nb*^$9XXTh}7%ZF)uY+Azen|K{c zKHpJW`^QgpLktsdb|6kMjjB^-#&M%~`Ug z=btV;+zi^W^4%2VwxOG8?^A(xPcY9b%9wd+{dgUi@U`Kr7yCtAtWR(v_*xEyLMGx+ z9dAoyQ@1drtAPN%Rwo9iBI=KWgx8d?N)09^2qAWu@24M8MFYstY~(h6%9-E8;g?>A zmzc&+|7t1&1o@Jfpfp=ecO?6!9w$DmIC0MMEMk{os&B#X&A16g2kbPZ_r1}or6rFi z?dr<6BI4!N#r@|N-%BG47KlCP%@^u)T{(C(5fW+nN&$L=p+{4yE~-twL_{1;tDwXz zFli^SD83#okl`<4+(OX@fhr!AN^O>Jq6K*L77`A#w`?#(WbZx+bJWB=DNy?>@IzI_Y5j1RH z>E$nx#*U*=WhK6R34DYHU7a4YVZjtqD8Wn{rw^e-Lb5FKB=<31o*`P)otv&cyk%HD z|5C-)E%a>XYS~M;Wq%?NK)lp6wJdw1@8dZ_?_s;!~i+Lt~_5O5>mSxYR z`U09N%uKEwIfy>J$gEE7O9`+>4G(0Jl05*p3z$b#xju|>cQ#-t_!fHUyEb!&4dY{q zo;qtd`)`m_2n!yQkzP@7z)2aFqha^&u=B+JrlN`C%+!aTgaE=<#7dY(Krz|VMYfAg?L6&Lmj^Zv_Ev<77VNGA zOp8Ei*-U+3B%lNVBQdjWYKVKBE;yw~F;$Q?c57I{gSPpw+x7RK``3)xhUBrp5)rph zxyMVA=R(=hqXI0q^#ahnu|vyygL>z(l*$)huD7q@Y9DL{8@=WO6BtT|H7{PRQ`IDG zhh$CKf@K0{LI~f&N6~ViF2~II&)yM-vJqQ|sk+V}1_2|=Clr6K>&=K`EW!K~;ZIXJ zJT^UXGj!pMu%yunVW&l?V@yGWy-m+Y2}DQ7Z1fUaX9jRJFkNCqt-)xo8P#Jj5W)$a zpR0(hiPx}^IN7kdK_YSES!;ttf^^7)0OUQ?%KTu1`i-8idoY0pUb~Gq6%GDq_8slV zZMu-iK9C7Y8>w=<8Y*X5xy2U5-4(d%Fmn^CgT8w>mt+y$XTb#x2zc7M$(m{TR#(?Y zzuZT1Q8nIG#K)%s*fF<_?HsM#=0r<5o)C@A77J?1+e4uoPtguwU12z`FwChyv1;di zKcqTKooNY7d$YU3y(}B#v=hk6MU16>n4;|JuBJ+^ejSEjmPbwgxhC(vun0vOzePau zj>7o^&#B>W`^vO>CF%21rEYCq^0ZU;z*e*$WwE90P_bV;%ay{(1YC`$eTbaHP6O_0 zYAvvmd-6{;?sSEtpr^H_GGKNr;dmXy>_91aq2XtadAbX|zgFfp0sv`o`RuAo0$>@u zN#8yaVDT{1A2C6Pma4ttuf%US)&LI8Yltnu zFlJ{GU?YdB@0@C$UpYLI82Yfc=tsNwuJt$>eRWlHwFBynwO^EPlXXa)i$WrH|1!x}7tb zr$$P#4;sp49K_AS4-=iNY_7nf7L*tn%bn9%Bt&IQEznN@9<%6!F#k{ zfRcG!!;T-(k=bk2A@aiaG%6p4wM1CIpMrQ{;}%yTa@--$d>(lLM#g3#7(Nrk@!XrE zrk$=NhhUIiKLvlkjcECDwB&8tmZx{T6@CQ_`5`qJ)=jfmQ|)Ao*4Eh8s3ornPaxwQ zi}Pux_u0QKN5S(lhotF|YIg1I&G|Gx-LNygC4m60K`sW_<L>E|WcOp}SGV^68$I#0O%*HM&&sZUqpptOHNrn4yB~r7o3;MgRsGjw z_m`5d{~fnXW~QH>Dbc@51!es)oqzr`{SDc@wsT;m`^nXRMVUVc-hV^?e@=FP|AoLG zQ$bl@?OFck^u!;*ep~Y&xaUHW@faQ87-E@%*@P8)>WK*o$j1Y(n(J4*^l#~_Nuw& z!~EB%8YNp^jk!t}xw4-HS1)qV4y z^u)yLp^~@8xYq9n!P=_df-qjDMixHpheqMuFO3jE`3}GA^K1)7#kn9VmQqv`NOdt+ z%u6;nH#CJ*oWry%P+KOd8-Sa*n9V2fes9+`Yg&M$aUBAuiBC|1>OM|TS1PH~0qI9? zV+p%6BXcX6aV-Tq%40weEzv)SSWnn;3ain&QO1%kKHdDWo7s4*_)|*!(4tD0fNfbX zcXNEZx!chDl=|j*ud3$H(EakJ?QBT?w6q}S4(lR<3}x)H6I22>(bzF~nC>+|A-lRT z(nIYviBc53nB;31dh+(TK6OLqY}cdeY!B&V)G{-GVcO&~Uaf8cNQIflcLu(d|Ft+= z&_ha7%`2VHS|%@GQZ?X1XwN_zW3lsBQ4;@8>mDc#I( zP&P&#(g6nuVUbHrmmCqvvG7a2hh53|d(fCy&rFokud_K7EeAA}WzZ{Y0Zw zag73MKm(hHdy}s&_f^DyYCy+{*@DDpFXMg0p^DDVzZZgiL2j2YPF_`ccunu7!1Jdhs0qZ*{F0+oN1tI& zrg9qkT~2XXs}{G<&4N`}t@vf6Kfp~+- zQhHD7HZXx7B6*QTv_ps4=bV>R?v{S_p?F^Vd!l$*VNwU{=yA%%jnr+iA}nyrmbQNr zAXv%Imeh?>*h>NeDZVc;J@iT5lykGTfuC}qh9Hc{v{h9$yC%`ZwwP+C!iVz17C~qz ziKI5ifv0x={>na&P8BXO8{(S`z2^XfwkahlM=D)&IGXqg|8CTjb#o|zgEhwo2*9xJH+5IPFMfTVCpZhl;2wF{ z`#!s;KF$vmf^e`_qw-i^`4Gtzs#R0KFdbgxCBHhi4R{ND9+Y~V6utX0CV+&Zh-9#! zkfNG&U-?|yPiZO>njKGp@;EVXBMI>GQEU-3Fx>#G9gFMJbrr_kJww@gCIRjx^6!T) z6SPR1_(RH5w>z_OiloI%N&9d)2|-+i3AF3-tNC|@oq|(VOJ|f!NU$UiVYuJs4acTh z&)si@nY%W1_P2v`!=_|y;1QLWS?h7LjR|*tlbVIKb6krEL}-OAr&d)X<2$Ikvg3-Y z*dnCN)*Pf9e%rm(MC2HV1*Ngo5e#6C?n3tzQu<4%9pG=Gu#S*^8NE$b1WI-bEsLKL z4+!R$a%n4s?PICde40WPm!i7X9Ai0D-rIaNgkTAynX{)2apwS0M%S2I60gU~rV|_q zp3sGUMYbu8u2+H`#R^;S1xnyXvPdV$A#kr?xgAeRw{h-iAqfErX?u5`cv%dke7Vpjao$})$|>k7C`jAYD8apZmKikF+Y@qS5ez!Kb%lCh5y15T!qZ=h z2%@XQOpNQpW-Yes8dk2(+#V%?9LI7y2dd3Z;s6wIG^kmNBLf+C&ind#XLt}uJ)&2v zU~o90JwVWIoPNYG3%@Lm5Or%d!+~Zww6FH^#$6Z(^+9_;&2T}}fM)2Rm@BW5lSGiw z#5B)4Ke0Wb)T?kTiW#lU!vEynfCQ4aC*oU|^hG2$d|svvH{bVj4@9lqMo=70gw6rRa2VF-{b0tRMhViS%U2pThn!eH{|4j0hCtg7r(m?rajX^4`u#dh?8 z%?=_AEdK@wNE^n95OB?fVCEJFT-RQfZQhAy*1)>&Svw#VPUUJVlmcnEBzx_9a(rxQ zijT~w$5vOn6SO5hZ{-GDeZtc=@q9<=6s!9;8|4%}wIQOIX}kjyR}h1kR^#;%ft36m z39fafAJ?Wn?qIT!X+{)sb+e5!~05%hqzAt3X`Go7r zE<~^hFga+s#SDscm0{*UP^L-(x98hsCUKS^7y(Z0`~G?x6Xu!y#_^}pJbBh?=J)xX}R7$omBu= zw(JCI^Jr{AK1Otng3wLNEg?>zIzQI=H^CFNS zuhEF8-xngl#*c_4GI9aEYB>3r%@R7T%S72laLK*PKDagKlqCu`Rom^nK@xcRDsxR9 zx3uN5P7mhYc9*Kdi!d1}71AgfomSF8$9qLER7???c{T^GCd?fq&ZgWL-PMS8CIt0O z+8bq&588)r?EEOvdJQaO^ zx&5cbZ)wSnA3y_yQb!`R(mqVmIRz=nD%Op28(%|*JVwvw#A|~+ajrgB3&yIw)^(D7 z2+g8c#sZ!akK!}RP!kcTs2F&&5L(ti{$@Yfc6H))xplMGki#)DL=bEO@Ol{JtpuX2 z@s0|c({8@ZM^{yp*$l39o+VQ!>NY&OPXNzyK}$x^oo|rcUzx};!Q1g*V)GPdd|3e9 zFEQel#4oj>!VFgf2tcR=pD1HK~nbE ziJq5^19%&^q=FHAGil!A@rkd{6clRV&6nJJsS)!OS)H4yO=;3c7_Hx<$X^xx{VxiF zOmAWj{uLsuM8FN^Kt%?2t4ONJ9Ca3=tG&vfUJv!ILY|bOY_ss z(LzN@#pcwv{JX)39GlH49&z!F6j}F}s`(~G>d&&J(6W2MZ-WjbflLdQQsW;}8to=PTWm!MQ$OJNSc?gNM~(vpK~f6eF9h8li!@ zHzsbGt6P{Cm73hEA}Z0+a(A{&7Bda)Tpr#pC7D60SP%fxAEt^q(&~Np*YJ zt7bSmsE4Z z1Z6TJ`uJetLFyq&MK5fGn=%}J8{;Dylnr+ahu}#TMGDV7mVSG~^Z0bEx$$MJZ40_CBAc)A<0^0(k^M_ln}iKrdTnl@DBxhUin-$tM(4>A)^SA5RV4pen+FO^kR z@2X|9Z<9n0J~SkvfmKf!uKUQ)G)&38Mmz5h*B;T%fvQFq&g6oC*iXJc49z7vLlr}a zHJW#up~b0U8$CcSnhG^EA|p>D?^-w*hqW}B-MJXD=+O0^z_x;=0IH*PbN{TUXD%)Y z3Z9(X$mzR8Ws>}^3=;#i8i8ZO2UFu1>U4HeMPt?P~?JXU=Tl=Wa<&CO<=mF%b`CQ=)0b9 zN1e(@+-Z{sYe(8Mc3E_%Z8Hp!Bgvbi#J`_H!1K-&kW&6^#RuYQFQPL9{9OkqjQqpW zcFbASsR0&{SAbt6nA1rq4u91|a-{qB*R@HP6kWO5X-`2d0kH5vWz$W=7f@0=mxllps04w(D<(*al~-ecQeFbKQfXJs`^&C@T~VKFNVN2q1J zRccLja1~KI^mUpRvK`V$#3?oP9fq6IvX)|?e`~Z@0C?P3soho#t zwF)9@I_qF`-MzO{PoJj3DoS)LFocp8k%~V!D&`A}z~ho~J!CE!fvm3&amGQRe9(~a zuTre%l|%h9KCGXbEDg^$azk(yy_R@9;bBH7+&5HaJ`;ODo}gu2#{|(ig^gIJ1DKV~ zBHufxb+JT<%q6)a(3)>U&dl$R+RHubaT+J>oEg)uCTdct7F9AjvVP_l8B-9Y_#D&{ z_Hl16Dq=e=0)IyL;aW8Yln3o~l3Qz-SK@<7e0K4ngGNZJM1tmyqUOxmtfHDP9@Hv$ zIksyWyFcNI4^tB3s9-JMR}*zYW|MMocK1voKWeF?<0UNH@LCP}Y>MyoS7JH#L2wx* zLd0nYnUGfQ0b-az0nB7G+0a-zDCkUfO#!SJ@kthzY9u7k$2P00(+Js=A_G}{>ln{& zlLFG}mnc(>2m9E@B0R46WX$Q#YGT6%z~fPd^h&RAhJIIotqAPAy>KT!1(jzAq2oL) zDLGTu`YR@Z4l9Mx@uZHe_vN(0Q)tx^EiT1yi*(2+y8N=ZCS{$eS43MM3}$7{tW+_z z9e<#Z+J0lYYG<|*aD+?`+S$a(xw=pG|IEfGFv!?Rk0Ifu(CTrdihngNtI@-bgvQOTmf43T7 zFUMC$0S$8%=e@0cmc`APE_=Sv^4ewbJxMBYnJ1ilXF%P_@iczpt4Ik~NNkP--Eb;I zcYFRIc&D=ty*IR z*7JU8kARxN^Cn_8{p*w|G?n`mQw3Gh6b**l)om{u8W8g5PZ6*uv2Nft)$?5<@z#(@ z!&fKJg#*Q*iNxQtKBb7XM0LVClYW1O7n8lVOh83&eNT44LC1J~TN2&+14NEatN!LF zz?tMVP~`n8F@6lIs_;#Cun)}=5|ar+HSemw_&{|21Us<;t@o?vCt@pEkD$uYOOBWN zS!oaiL43=IM}n(iHr$tIK2t!rRun6e*J;xxSH*tz^?-#~GkDV^1(fS)$n%zsC^ooN zzxb!l*KviR)kYC%3HEVDhs%NNNU^d~HDMd@{CJ;|BZ{F}hm3E7W=Q>KWN1DdlppwKfmmoEk>H+~!ylq8dD=k{zU1>qOkWyBEMM&#pZ_om?ng|bykS$BJnpcoYo>)gnxe7Ye< zO0(0)Ins+CxUO*FX^9N8O>8h8)>nCEfiA+%Lj%_y0j^0`%mDJ^%wgYwcqpd1pIuQe-j>=-V|S^U(o#*TKc*4ufyZteslPZ z=J@tk|9j~7GuZz?zn|>(6a84}f3cqdF#S>c8Roxi#IMj#Mbhrgex~h6c^yMp6^2(J zn#0nNX<@OfXmGfhGLQr!2vjgk)SK`lXNvZ>3hX$2w?bzsmgFAY4k54ByXMq2rL~@= z$VoRAELrvo$L}%W*71l|2GnqtsG%y`EgEsD0%2mzC2PvxLam%N;soQ?317CF=kq4& z=O=@uyGoJxUc&=Rg!cI)AXZn(VgQ2=70=!ELDjbEB%HD+E^&qzhAqfWFNYR2QgZ;} z)0CM#wlbgY+pMtaSme5Ex%;iGTl(sU(z%|6I(qa?8COBa(&++iSDqvoUEgJ~WU~rxoHa02%u{p8nNu0e{*y73> zSDQ=CLO`mfU?0^IyltKt%?h>k3}j6Xe^PTy`uNLu5nd!xf)`o$UCy<&vQ3H+ zwiW>DM}-*F59g?;OSY+Kp-D?9*dN(fHw8O$z1PE z2>wS-7X>MC;*6Q3igw|X$0E~%Lz(&ky8%ly%w5RN9!y~jUN8zewltmYiy<3<9E1-B zA|W6xC6cJ+hGHf)K$phsw`&~?Lmwd|@$0Pok2w&p30vL23NX2R_tPP{LpGLI?U+)5vJ|5j2RFzoae4emnx^3Mm%sPR<@%O_!M^_Y~xyjA7mqN ziYy(5KT6qW;mF*d?gdBhO&G?~XT&m7UYZBuzY2wIJAD2)8=EIGc=@(fLgb$iT`+!R zju?>JE!2=jqYK6r2-YtQ&u)sx-scN?#<33u1S@VAxLT$r9#6a2|eZ>_{sn zflbVpuE@blOsLt)Tyy_;M`$4&$sXp&c(WLZy#9`z$W@`^AY(=X6GCFOayT+?be6sC zY`Tf!RhVfwmS}PJfOM36K4s1#{KvP`29uKX?mh! zpag=&>g!SDjTU$^P%Md#C&@a(QDSznHw&}@IP`?3AOt*&9Aa>!V-kn`l3LNmvx>V* zH>nQMU5xF;a~mBSxf;W~#a4XNq!&5c9gpft?u>G}00*``Pl71jibr9C6$4h-;ZmNi zbrFAQpa$x{ZVd_G9Ste}35450ibfNf#@M*ClBak88n}8lh zLp*pBioM&}T%8xbx9hm@qKwKWurJ>FL?_>KSi*UD)2_w7R4x1>jm44q_*-3AUK3k`tk(+*gd?5Q&F_i^mJ6p z#?8J#S3ooe=}PvZnk3P&d?gvIVcM=am`Wmkb5!NP1tvKX3D3&Er2se$j=|jb`rMrV zeJhYt6W6!2qfQO$JJa*~mZRipa;HX*r2$^H_w+c~5I~4YDiDu%Advv(WN#e=I2zwF z1Fj6cTp#&|Y)0qEWKC(*q_7Y8{B%_Cttx6VD6XuqWXwRN8w5d&!kV8V`+Sjh8PhNV zrSf-dkIO*oZ(HlruyFL^(ui(yaRx*5BL}l!N;2_8}0;Pnq!}OYbhuS0UD~mL> zhDvnUp>nhI#LLtA4KV0(KVSUXsweN3(&Xv5E8?W8VAz6`PAgt73v*4I&FhBxqM9b` z;{4Q{J;oeh{Qy1d7 zc5OpVrP<63!R+*mvg<1zlXC7$a{Gw!Q!-Dy;Iv{=VlePLUnIBr+@93?bDnyvY6Jk? zvH+8e3iHVHSv1)ab-x>9=W;heb{~l2G`bv%Q$WFiuwK6ntb@)dB6U}xAiFJ|m}u%5 zxyS(H9L)&UP3z1M2ex!e$otJk|5=bmdaJT!QC@ z){jB5ISkdIuj!O2&M8H1Pyl?;yJcI`rC2)+&H^*F(7-ReOT#;#@js2+GEc64+g@zK z(-o>q=eCwCo;@_hi}j!6S{cRlJ|6~wG&s$xIczz`ge+xl9p4hHU8S-B4=w9{deH#I zaupuPJ$xP+%1CUr-gw_jBq6?-366@Fa;HmdX-u4tiY{@f{Hg95*6auMC(rJkj!ESv zFRX1u9PRyyYyR7+8Kqb|y2BrS2E0#->H7y@Ymtfa#yxCEv(^Bf+;`wAi`khIr(C_S z?T`3lLIC^m{kp{~JnrtFZ$x{nGXI??`NK z5&j+}{S5q1a{V_b=_h*r|1#-hW&EY*jt4>D_fFHVu=fv~rr$_xZxSqk=@&-*D<%Cs zy8J(t*xt0%|2@w78SH;>)=yCS$yv+{jK6FL^B?t!VgAcj{*|-xRW&2uILrNQ9^TM9 zChxNSsTjPom}QI8nDuH*9sC%saU=_d+={3;qt5edb;CGMA?ttT&0OVU}#AzudAIwM>~Q zc_CX~$8@aG_GsF?$(El~2sdm!Gy5QB$YDs?$!rBm)wE&GhM{z|nP8)2FF6)JtvPi6 zWk?Bo7T14su0+yaan78)cHue%T?mc8#-l)G4r;dO$cpW?OPB3S!;J@T_=05cN-7*5 zM!8j$n<`~C%k5gau@{XBvBXULq8?jLq1uC9M+hn})G?8oQb=xL%MVxNvhs5@i#p|Z zL&0vo%Hpc$j|G}r&f+avb?Ke_+T~X56|1YI?sO8HI5LDrd~pHavl1<9ERr%tOfy_u z)*%KG@)?wRw<6_S<9DY9(X-$0k5Lcg9#2wgm$#QKw9HvySc%_8V=vUT8d^r}MJ z?bp3z7dC-fqvg@D&c1SIQAD%?LSneuF=Tgnv1Q{(wAS}zQ=W;4B0a2w)GE~&D0)az z(ka9j@M>XokiTjj2lB;m?~2hhn!Y!lVUd{5p)WQw=%nm4lh4Kqk%}kg13*#%z87&u zC7tQhj06X*P$6{3)X|s5VZjHF+~2WRXMLgAjooA0(1$8N(Y0Srf}ZDXoDB=14EAc0A5KxOz-MO4|fk3g0WK?7H=wQ$BI ztH}#a7N=5lipJ6h<8>29wBq$K-|z$LP7`#|VjjYJo!#*V&x(jGOC@nq>Pj}Dz@gZO zjT%Cpd9}r4oTF)e2ky~#!^?GXC>5z&IXP)*Dd+NNtIw*_Y>6(z%!|%~P(x}S%#5#1 zbqb$e#}^xPV0Q}RmPab9!e&+vm^~SDVqBE|5}ev1HA9LWmYc@;KEjh~X%f$zt`T$m zK&9L2!ia&w?L5aUJZ8Vv5gk;_sG|ibue+J`OErcm2wJ9em$TUR$ydG23^t_j#0Z0F zg0CQD=(rlWpqedf!C?B&I(LpExQYUGF4@ekB$4PrR{Z8gWZyYh$ld{cx$`JkF+{tb zN}Haw=*Q|I3x*A4k>TlSsd?Yfvp+EyEYE^gY5;P+@>LDP$~F9&nc(Zv+-m%F*AG9B z4|`aoW-d&}a2?GJ2*8wY_k@n4s?EKz`URC{bt5D zKxvVaJcrK4u|NSR^{TS9p+>u1=G?alR3s0hsDaJ-SFM58zNX+&E_7!mCc6gN_+W(M zpP3UeJPv}Tn+Ch~+|*9pY})g8??e6D%Pg?%Ug$nfx%t5k>XC~YG8cn73_(l<*L&|$ z3I}*y@|nHA29o6q5JWB#H+?zLjUwoeFQ3o$HL6Z4?6s!C8aOF9nsL#2d;jSB@oF7A zwP1AxcZCN2e#CRl{MbSl3gdO!-p1LO(V&@Fs~NkFb{WCKY0zL33*e1m4sv^F=~(0+ zGY`RQ>*Ov1)Wbp0N_NS;7HR8c7K|p?Q(JHxlEDVpHf1!6rKQv79&JfzEhRyjmB6~DfQ+>8 za7u$2+}9IosJZc>7yI}_ST)7qGdU!d!kuSffkd*rw02Bg6pahKd$Ccny)c24ij7T< z7@QVvj}Qna!>*!MLQTWB*A4Uh9qYp$d_HMEa4w%ea&C9_jO^DA{$`MP{`fW3*w9)>m7uQr1mJFpK5<&ZFZ)O<*e}B(TKF?Dm;saFP+P)O}oJ1E;}4+doxqz*oz9IFo^W z+H4SwbmTEQL=7fc1z1SVbn z>;?cn?_n&LOGQ=+DB8xtkGgg_MR!em?6Z`XAFF(#{TdpW!}%FBcdr%t(rsJL{Ss@M-d!Ly7@yptZoafUuAb(qf%^{$SDkuT`4L^|ahGa%YRZ`2rRe?w`^QG^ zT9zL#$FFL*d=WrZ=~sNY5q(asZpR}lBTs%LqF3LJkI2Gfs{3LUqMwR9!n-NU%SCJ(@CEb_%%R42V*! za%z@pKA`nubXSVjrf4lXo8Y!js3MSJxK%y6BUw49IS>i~##&+3(&{&ST3dde;kpN- z+~6B=>W9}F*AiffT}z>jN-%Cjk#wo|ug{>(i^6su3ly({C}D1y=~7M(N8s6JeH(Q=ztSQpLyu?5UxHpxAwG;anxUZ9(y^~jwR=)Lr=jY^n)ga zK6xg&G~wUd+&6X)zsSqINZ!#hZ!A5b2brW-n)(LgD_a5A?d?<4KRM94q z(=LX#>&(|{)XAT@J9LzIV3-=&ceAg12 z!9FZ!=NiFvGmMn~vG$QJE{c)hcesN2zhedSX1oY^tL%Hb$^S;Y`?>T_Q0;G3nEzts z=bKp#fcY<-TK}GS_ZH#rVd>An|0LIc9ZQ-2_13rlM7*o~1xsyzxAODnrB&V247aDZ+125+0nD|#Z~ z$uvTRUxgh|_D#D%Dc(V6c-%bFhEO+bq4l z&u3^=LA|CmNCp{@B|VlGx)*%zy>o)90`J@N-lno<2pxTvly*m1Aws;)T={0|*B*56 zB@+)_6sPQ-__c~Qr&I%dKnUfCdNIzXeKn}x@xrn}BGFlh>l7{Xl*iNy;YXI=qf*hT z+0^?QiH5Lj-J$sy}C&UxQLXKbKbCk3c??Me6g{{2Wy%5wb z)D(|QnX_76Ud|$4J7g0@Ya_NAzxV3FuC>_=zW>miy+SGOS>kJsHFI)tGpdrL)Mgx4 zi)${hX>^b;k5@rp?{zrv$?o%9SKa*+-}iDRtx3MLLVfc3!{j3_ZHT$@#Muw7?QneH zxrTd+Ar;TYfi?US5nkiH{u-|1LfTtdV*V|SfZU3X0oTWAykN3Ptj`# z-KaPlK2^#qz;9-4Pl-+wvJwVzKL;TA(20$%@0)L643U>b-1~2p_n)OTh}Bt=+=Y3mPjlSE#Uz}u2iKOnV_>&m$ZmHdsN;>uc*3SeF64~R%5vecJND^lU~dO zfXA0qEgrKJbK;$N6hIqmPjX-eO z$5-PS1#<30HBvD@Om9o7&cgzHhkocBdWV~*5tiNoF;`+)UxVN3umI(+xX%OAqXfx| zJ(p^-0QyORVH|`tJP49D1ei|Zi<5WN1&aw?gXNhGE7G-<0z4XCK7Kp5ArMc+GuT`sIHUi2(SX#l&49o)HELu>SfRUtXV$w`S$pMEO^tlZeWlN&P#cF4NMoc;5@c9yldqQ5mRkfG_K<6 zZ8-dFr&V4KJmkWMNjwAG`SU!iDfy1-1#IYuI$~u$hMX_}r8QbFAlToVfKfslM_|-t z-r9AI0o5ouh*8j&$kLx*BZKB74rcQ%7pOoL)1adPNI(=b|AAw5n+{mZ1AW~dRoF(T zvR!u{86n^nxP54VYGDfh^ExAX%x7qY+-gha_n_Pzz_iFwE;S)TBQ!$lf**iZD0k9| zC*(7aI^q$eMs&E|`Q?2|#H1AI-5kX}BDydW-VqGK4}a;}WUUCZ_0}BMiL+FUBn~u^ zq}L0dt}!Ya9-odvN8D~a!w$bIvP;hJS92IVG|fOZdUj50%myRMNe&!kC_air)<+gr z^(nC37fH7H(4N)**mC`S6Es#WIbi@hfXp=wVY&mPD?g-B6gNLt$ltJ7W#m(;##fWL z9aWrfg>xtCg5BX?0)fRwl$sI^P#7=K=?6dFGkN!EYjBA1d=xb$Jx>C;&vRD>gaZ#t1fV9rJ zO!f$-x5>;La5voDAO{RHb+A#Jn?W4oNXHg+pJ|?`8()7%6f`dPU@WryYa1C*3i2bh zPXt+RtpW+@c4+wO?1>j_xS~Ge1Dcrgx;r%^Fd&VK$qvR5B;2j4cJ0A z{%hDc5tPgYZSk$d@6c~WuMkY=48V&xCaG^9eh4CeGoGmPKQP#~7PX4#HlIJ!0V5)> zs4SLbk$J_RLCnR(|1wJn?$F`Qhd*q|W6gupecQEwVuRPOLMSt;>-Jf?P=s1#TrG}> zt$+NR1a72Y+6OS^i?jJSKz3=~tIQjl@2lFF!a4#e9e%$Pu(_ZmIy zgTP^G)Q9uE3U;mrsIz63d4i*oAak!?no4AucJgVv{?$EHt_iPJxA}{V<=we3qKF`5n~yzlstgeUlB( zUM+t6U{eO|an33r&@eSS!RUJ{Lh2l z_=?Z&Q@P?$4GByqz;Rj`oYw`c-w%YQ_%!)DkBof%ZZOvUbBJlie#Zz3-CT{mvgGp6 zp}232{h_l17;4BdBr9*&LwY%w9D6d zK7O#maJTY}&ViHsnfLAB#3pI9i&3#kPGkze0$ay}?25WI3fACK3h@IsN4bennJ7QLfz+u*|Zu6{F*S3j$>*(1YYQoSd6L!I*) zx3W%I;&%4+h;fmeR3;4}tAqzK87K7}(xdK$S9TWGPvbm0ds_oti(jshEPtwo`r|j# ze@}j6e4CvHcry+8Th95p^p`*X%^mjtM1Es@n|20Z`K8C|-;m$_6yfi2&d%+Wdm6MvJt=G9IGV&n#Z5SGobf6 zC65U&^NyVJs>7$X=|D9fV$OQ1crX17A+uXg@4O^*W zls2HwDpOZBmr&V+PU98a`cz5r1bVL{cGOPv2^DBn3X<1d22zwz7|-&Q@SYUEmn}0* zFlf0PFWsz+tf6IEqP3%wUG%|YI>P5r3s7q%luyy8hSQ0G&;?I9B(x%|E2e3!Jr9-t zIOfZxzU-oxh`rADKTV=fFLfb@nvM6l0cbAS6g2}2f0b0)=!5Q}4&$$fYMs7&p*Ca# zi&21_peJqiwLF7eI!Hi`u@WS=L&=U6V{4^?aBWmx6zQ%9DamUts6{KYwD)QYK~r8a z3HLn~>r4GQ!)q?V8u+8p3a7ww4c;9=$hJ3cF*xCHqY67ZQOxtqzmowVrUUJ)#Aqj5 zbJR=OoYTc1f7nimkqDOS?T~*rXt<&f!V7 zUn45TASn^xIy=`(%swsf~Q=!vh`hA zdHtuD$}ej~a511~V03{vh7|!g>-(K#K`+-24(-`aP&wc$Sx&V`3UtNzlUkzo@gR%{ zY-l^x3E~}^vR}*(_T|+JX8OdRNv`J-E8$-V^uHk-qma*Egm_HLLl6M5P8=&on?<3w z*+GNTU#kfNZ=`=4lHx1`((yt}uY@!VlvDbQcLDH5f%nU;u4qz|g~)2*d|!ISbv6O& z8`!8)_GLur?V=L4=rZtQ{qTFB<5U9?102E>hZrk^_||mjB!vcqbk)79?iT%JrG%et zd*yigQKwgKc|fBKN&;l=_fOyXF7p`mta5(L$gq*hHzO!)M>_<0b2MR(=l7KN>3x$e z$OXb>yOT*`TGS;KjeNAOf2gIEWC**|*Dy`*QLT z+ZkFwA|K?jzRj&_TS4^eHjx z3>Jkmm)>MX*vNqpNBi9iG&Zn~KxbsXkk8mx#A%RE1vnF=Q#Qmi|kg4PC^HQ_~kw@q3>nSDDC#%SctnFcSX za1JLVG+SLWy@9b_MM>(p?RYOak>ZisqNvUc)JyJp-kg12CenZ;mhx%D01j^x3T#UX zS{6d~i;`MDlQZzeWk@?PU0p9&T%WI|M_rFBtoS{}#p~clESE2tQEQ^^YpLkvH%F9F zy;}8fl}mOdT(dxNmO1V|QX7@i;_~l4`@2U8!s~>Q8PO>bdzQU_Cvw%kSw5sZ&jT~( ziNsZ>;g-G5No{hb1N?o1U}ok4E&(1~nH?%LTnJV{lfj_DE+7^XC`etpX4LJ{g_&z;mXR>N!nLj0jC&x4V3UV)uPv)C2yWFlJ%Ogjtyf=h)d@%uZH1>Ot_d1H| zb9FLXFgiHOR7S5%g?Qjq_tjh<(~Qqy4_>N};yjoIh8%sc-#5Dj=<_68_wp*!;0wtQ zQGfzf@A%jDi1qO25zjz6?FS906E15Q+Jps3Z-E_HN!R!@&p*r6MA&?Tpm+F2r&=Tl z!Y?4s@m}EJu-!Iae_wDKBL{G(I|MIpJpTYiYiSPA|1yrW5A6CgF|n|6)F!I zk&e##QA=S~1ZLLxMZK$(40R9NYQ_*6|6t%V9d$^5K=H%*aIN3DXh1)6lWn9cP1NU) zI^=LcK@E@>7t?ak)S4JrV{lz8G)S^nw;3QxdS;7pV^6-w9UA#kb~EZrO!G2spD|RJ zV5a72l2Z$rUT1D$Kuo2dZlIxQ*2p|{y_KeZx zl#l+DdT9MtQ)!-Ng!yO)nVrFR=Wy@p`D7U|GbyNtRh6dlK9fT}Y9q~o2hb}eCZx}! zLHxW>DoNbwEI$MYeT+{YEQ`}szi^3sYWSdpUU#zYyM8C~BaW(HAxd^c=xJTchOY4- zj-!`suU1GI+z}nWZ(l=AoOz7Qh~`CvROs<${w)0aPDDm@bS=ZFerTjr2AEe0XjhXC zLO9W9bBo|zp4|ZcX zTepY2*stUqBV#34ihjkk2aoSE+=i$utsdzgwkjB$KB9o1Gg|caV^b2lKtvIMA z`0h?;?Sv@p%I^O1j%cuzIotzKgO1kR(pLW!z3lk1(H0#c82ziwChx(ZUIMvdslRn> z9a02pgvS{9Zmr&TOz1}H#dq=3WAG%40a4{8djad|r;0YVMpD!7!}O+*2>}-)RL#*V z;adIduIQz0u=PQ<>&`V?{no{^2SjzUhoqMFZgRQ%($mIUV4@hk7vj(>6Z;0o_q#Bul!!;@Z7p|0m+4wCY% zj^MP&9yY1-EDNBN%QHr^M{rHzgOe2N6FCoNfb6l}eRAwn=q?bPGtDwAr6#e;Y35hr zlrI}jms3&Qo(E~dc=5dfgmkCS(sN2Vb;!PLj9=NqyT?mfNwm35m-I3Pjc&j4d#FC6 z^LSw^{poW1w@~@3lDR)1cl-$||G?k`Mz%L4g!O0L+i%pepG*IQto~N^`0uG@jBf+u z0Ia_>Tm3r}`4-{tQRL6S|0LIc1Qx$fk#7?z{#73OPbiZ97oU{a-?i8L`8a<`@(&!J z-%uprO{--6h2ehzub)f5w{QQLBLB-iDQ^=p{?&c|`+D2YV1G{WR~-5iB>%L^$i~3( z*X{VDJ}GSf!zbl!R(s5p2$IL0^0c0lA8BdJ%lC2Qs$=lu4#yszkGygD)rtf~eqyG2 z?VeJG18i!R7(u~0%k^>b4R)9M9?2evbU$J|Dw>71m0FFvnNeA-dZ3#S)H&6I{6&<4 zRrb5J#c?vrY>M7KF3>W1w`#iT((5Vh)~P0-lTnY>M`e#2en{QXQ^ z02K;(L@b4K-*{juM0TpMfm>TVHzWVR$0&cs)Nqv&&jJPFMJX4YCQ1MW9gI6;n2bik zCcr_pSTdWEiZ#8+AH{#wT%jPvQv!ifY27SwP}yoNWTKE>vd*dZW0?$q7C|vmJ}ng- ze({G&`2iH>^JF%C{NRTevD_8|UVBriv*X@An1Zbfl1Brh<}qRNc~?TW58EAZ=4Oqw zZWdSo3RY7mn^T5eBty~1NC5%uiYtBg*TXJa2Ol42n6Rye zm9#|7ZiGrkGb8Rt+1+-@Z|+qWX-e-*ThQAO{bS-oZK~tn%UJ9oWrB!wM5ae5dnh`n zM{E15{6FH}Ix4SpN&Cj#-66QUy99T4cXxsY3m)7F?(P!Y-5r9vy9WK}bWis=GdVNe zedc@T&wcOp)KlBms^48zSDjI06;FrCFGzrxM%yfMIek<_H!m{Q{$fq0$y<=y@DY{N z{_8cmK8mPhl<0B4K#N#+0lWGO0-1g7<@78i$Q`%gz9B|FxZ>-?B>P3InXpJkRYoA6 z3-jO;eQ=`vQ)(7?+!Vx39q8OZF1l}JH#696`w^a>JJXCMzptJ8zUYk1od;zu)wk3I zr9oP{qGF$S59WGygMN4kX$kR1S~tqs6dw!HQC9poNP7F3gS(8+nWZ};r{id!jqBfc zWiVZ@H*ZrjSXGllw%J}L768ariAz;oQOn&mSl5xEhyy7ZLN3CE0st1|7i>%+fJ6j4 zE~bghyYCAdx@S|bFx((U6+QFot#qPuHZg&?b2izd;bLtF3&1G(pStU|7?JU}qQZK# zoI{AQ&Z|3Z;~_w@`}6b#r5!_24P_;SCw>Se>lI;W)R2N?&;Nuwr{aY+Je|Nz* zzh=4D)yGMVKc$n!cFSZ#hA}DRdl8h>OMIY6mVmdJNV;qJawY%>38w_G_o#Kjf~NOT z*#0Sb@jm#*LbtyHDz)0!-@p^Pf^HHK!?#3 z$tbUCq{fnb(&z#)yoSL>3O0u-Q>oq{RB~2Ylm5X3G4FL81A9D`#G&bJ5op`b2qb6% zX>X*0)WZ%SzsLJqhXh7C3k*eZh+qI_MBtxOA8w17HSRUJS}ss#I_RzlZ4EgKfqMb>_z}1 zK1}AYk!nS#7N7tLeT|t`ykZ?zC22rsU1@mz|5*N9;hMGZs zT?qJ|}zNs}8dM)8-(P3UKw)y5Dzk=U&$G#evX<-5LmB)bmt zkA7FQA_CRHsorNr{UYT&jB2iI=+`kqz#U;=n-Sc1A?~*^?%WNyW`aY}qWO;zM4rxc z)cHj_idK+nk^K2JE6H25pYT&19i+oDv&R?+H?K}zW9c=A*$c}D65ytUK7j1F4`3>r zx|fKK?g7n!96JP%D>bmDk-h^Su@-7CS49hHmr`Ml??TO$M3SXBGru-$MKV&uLQ0Vs z$MBou`|J+k5JEDctyk$8@*5$j>#bVn^6{b#0*XaOAIOBEFavZy!&MFJGvmf6EQIvK(YBX6-of{X8mh6a-l zkm8kHeF^z$73mD`W7b;=LgtvKh)NnCkE4%m=}pbp4dU=@Wva=8RZ7UL-1?jlZ1jfS zk=?!{4E*lzfI>2`R6J;`DKtlKWhgvdK%RI>g5Kfbagv| z8}#*rwQJO5{b1Y1cv-@=sdIf8_(_o>5@>H)d)g^8NvJ|<>!v*^)rSz?O+{&l1>02a zE=2|idz1L#7z2c;0tw0_M3rFm#w{t<_{D{zgvRd{KV zcc&zlq1$qK#C&)ySXYB+cihFIwKr&dPi+sy{22l+;Jb*;OguWW)qr5`Uh#6?%~rw7 z5N@G$ep?5{@akxyG`LztnzD;6)X5DOFa`uTIu>$cAF<4D9`*Q8O&jKgk$pJq&yJJ_ zJB8m%a-15Ol3vLKof7rrqM$BdY$D$`=$$LH9Y7XPQPps7 z7AMDzz+r%F6Udj_;LM~2`Huy1>yS}$d3$S#f87)2A$DCk9Kp$Il8Kb`9Yim zGSo7rT%&q!i&=hpniSFu64ccDQY_XVTD(A6)7BjnGAh2{%v-PHi;TSRGw&CDsA&hR zQY}ya3Hh=9Lsi1XfAe@}k@dBfnJ>LlO(>VJ>hei-}J!GGYkAK>-NDkBTq9|P~${;bFe`+pHR zsZ(E!S?WXe+EcyWl0t>l2Ul~KIHs1M3_X#G$0z1t_9b0J^vQ4X%}TBqZ)^Y@>l4-= z+Ho)s1YUklZ9ZQ*{^EKUd+cG%^(Jv*gKk=6r%bjwC}*`yjgmExkW4-;rbaSXlVQpi zW_$njm$?c-*ZsoLeF8e$*F6>4UxfW$YQEG6?)$&ZRoLpLhVuXXrfzvjC^pm*oRIzN zTm{|GAeUR4)?oCv8p%@AjMH3Xt`ZCFo2Dg*#SnO0Hv$i9&c$GStbVG5Q4#zsEfMX4@`%7>+pRMJ9nk@2MvgM~- za8ht#VrEqu?+h%}Zt(J#I55LMI5#uZCu-BJFJe~EP!#Y}9*K2JKsF{05N$$@uXj)4 z1ZmA2pN}EE^!NaW6I20~&7{T-Ih8j6@157qYksQaz)^gLkTx_KL<#m8k8o_W z&{LX)7kingnbSiOkH<9D-+UVsmI<>}HAr4SNsBbvCeY*%!6Vmx17+VC^jKD2UjY=> zZ)}LlMSKLzF|FB(^_9SDNcQnliz*mqri0KSo-&bKzJ+0_3N1pr4JzSLtbfl~FP2@vyZl-^-=piX67; za+=-%q*17Tt1}jZ1^cg2vH9s4w{B$C6(w2G008a4%Gl8{TD2Im$#xHYHOrNf|!ZgfLgMr3uNGQKHR6D*W8Uhyfs|4_(HAPs^D`M5( zeI`s^3HKosVQ_qzJV5iUF4!vl7<*DK&+9=2s@lxpTde#hPg7+JDV>8x**K_X)^Y`A z+gU*c>W7V^Zce&@dumE{1VzaaW29$#nDws7942o6)ra&LjCP!2?|KHO@#IkDTg3~g zd$Q7&?3r($v}ECz)eUc53m*5Q!1%q$&SElo>OM*zL~gbRi1jEUeuPex%eAm@pftb_ zg6<%*g3g*IF0Vr#NUoei@#d;cTtmA_Duv2$#messWjx(g8q}L$itK90DdcGMf1L(0 zOeK>;FqWw9H-;r5kRoE9)J|L3Z01%cvqcS;K(XDE2r^2>@tPB_juyw^dmMAl+-5_v z9+!84B>m!-Ne2k2@qUvD`^9p ziKaY$-%G(puxfx?d_>i*oHtyGvo|CG(qWHSPZqT%Wdp0s546~CTa`^pP5OkO zIh(pvui-@j4p$bzMBDn`&11-e?+vh1?fg-izi;fPMgX$mP=ZD|?T67KqBp))!1xI^(~SPP<7z)U&o@;tvn_ zns#Y;G^&;A6P81t14}GEO#y2glcXbU07OgcO3+BKCrR+4FL(4&ZHfzKM}~eGF^rEEtM5&!^MhO$N9LPYfzhRCxp@PKTo*IdqI=fbCT2m{slGJ1;X0?rA z-w|hz4Xf#mrSXm3CMTZU$)IWstH+^%X6U!V+F`v&6-F{iMidCf)MAzDEykY3v|hb5bujEjc=1>>Jjx9T&o3 z*$9%pu&=9}UobQ^ihfAmj`i;FS@E|Psr?gZKby2mL2CB(nWyUbHHNUh2`vCfJd`_s8FW|U*J4^FdrDDTrzP3WN+(Z zSmG0d6Zih)s;7$$SA;?279dazL*(r=En)AFnC6wZoc^u4WiT07RIB-j2$xQQDchJG zs9lP^X8&iz3F4SIwsjtgZlAsS`1mPALUB*jyF0l#K&}jBCXsr1IkZEg7 zVc{3|vj<_=7B-ybqoUSO_{{wVB|j%E_m9^3;eM)sRF>APDyQ|Lgc%u)yT@*4yj2so zefKNzpNdWbpcj^Y@g_LrsO^(Ltn){;pU>tlh45t@X<`W00bZ&~RcPY)|T1#a`afTq%Dv>SJ51k(n zUb;k%8^CZ4Kf&Z3L*{SgS#jY>;=ymZeO#|8JnuUM=Q_hd z7(-o6wkQ&M(7_633kzl^t6NaU&&-00oPVabz|mG#!c%9};<^<51+-cv8Gycib8w-0o_xe%BOxC6gv1uX7(dP6lfa{y` zy(hV_;l4?*mO?vjd++g#hy!--qksF_%@$_F*bTDN$Mk1(hDUygpI{u@KNQ{li8%ak zR)xM*3$nBQS5cN9OFxU)|ABP=b9ny$V?OVffo*KR_h0>;eBPUde?@ga4F3<;`ZuWV zulI5NJ5Po*p0^0t4a#`vG8?k%~?&h|S~|3P*CiiZCUs(VYj{_j!U4`cs@ z>V9C^uT;nWhjbpl#ds3~wK z#mlT)dBAD45`Ni~i{~nNeA9WPTDgO>v%V&>Ci0jH?-xZCS229rY2|$qb3xmu?iBTo zecgs;Jdk%WGhx+ny;WyreyExxZ9!sfiG79`Yr_lojxem{lJhzkJ*6i0N$!i=;?ACo zJ>?7H)uHjPlw%erB!kRjSN_oU^?YBE+}>e(M>My4B!RjHdfyu-lA2eVn5Vi`Wth0$ z%Z>9|CB7ipRtB9|y-YybSWIH%*O%^yY|46Fog3d~fNQE{WGku_XTVB%O9_fbql*e) zz(;p8To!z{HkOz@lQ5C!bX`(m6!n~y%NGnzQ1<%}Zsjsk$iTuh{urI&bx@~|T@}k# zW6W&3zXG}hP1;sR$X1H@_+=cv-5_JQL(355!!e_ zzYVr9wm~b}L+*L8ebF}=OkXea>6^xbs=3>q-ew0YiAw`;CHhAvB2*ranB5NE#aR$i zI*@WNK`Wg16ac`Cr0Pid=!j_LDgd5fO{C(R!ddt{{&$!LSLANUMfx}%i|(MoZAkmm zPc{9g+OryYBxylI)0PjfwMP8Vwoa>qbA=kaw?4xJH**o-WYUyhH$EBjIth3VBjWp7 zJ^DhSU{_sTKIaiy`|I38N@n8NBarPmrHURb)kz4!1K}DiQu$hyy`b?!Wxj)QDWJS+ z-IeaPNKdQ=bi&*7UVARbw^bX%p7=nH0Qz2gQ<1>e2Ai4+LTU)rgNTwMSi+>D=~cet zv6+v29Wky{J~OR%X$q+qXfc%tDX!!(6x%*osMl?NSV*#ru)d|4L|{AMsHcl9xYc<* zCt;6faMecq>*btzlHLuKFPdShccUa0zhy?s(>L0A+cQQ%N5$dNBVTH}k1?FGfC1NLMXL~(q3j;v6P3)lrq4*@{8xR@hEV4UltIKkzbfHykK zuQH3zY^>2d>WwbkPe5NdCB_@kL~IkS(p=5s0@TknGRN^ip%@!zzl^M45b;Qpg4{$p zw|UEJ@+Ny>bQlCdc{lhksH}g}izK3y0&z`cyG#s>_cLtSL#P7%rpSLBmaY*1PR_Z; zBH=&fVJ^{E1np-jz(lF)iaBVTFlhznyncTxXXYaN;>!2Egsjp?(NYuIsXVuU28L1+` z8ORu3T?@^?a<=Sy|ch!bk2hqs_wo2yn+hQLH*t_v*~%!Rg=bTgPt`DR82 zE>5w=ty3btds#2PTM=O`sZ_HdD&R&lm>we{hpgH2xQznDRk(zHj;fAGq*`N}#5T7) zt&cE0cg|(Z=_?L~EgSLR*hn-rpS;L^^#>+whv zG_=ZBmwP0wdJM7w6^Xal;$BmG;P17stggKTCqEMRu0}k6u)L5jBb3Tyu64Xr5>kEn zV(@u5KK}Yhp1uXomitz6ikiQlvH;V=*xc5#hM*3U!0?E^6V|a< z;j+NK$xQ&Eboz!uLbRT9@qn!vt~B#W@uUuBQAuKBQ{jeV8_vp4ate;4o2HmxXekX2 z7Rcf*yZHQ(QG_Fe0e*n zhnldbZBeo4_#7t1F=QFBlPDKCp4=Q%D7~8n@5fZ)ZiiXC7q$D=!x97Lf<3kp-`ZTE zt-OVk^bTmmiSvwo;I8hr`qrvVpEh2q;Cs?KOyUpvs#QEt;aM}W1iEwC;)!hFw~hr$ z!}OA_JFQ;`399!*jdo6y$w_je(;SIiV~ILB9ZRO?fjGLWVFVxAI%TUBS+b5QX{}Ef zsNVr`t)e=K_d72}N#nQSSMJm#R%DGMTW{NZ&aUE8MwNuw$j>C-J`Xlaw0z`N1;Z9F z5(#n@cq#=g7_avdP6;d-?UR|YbnMve?wmfv1hZ@JXnJyiGch`Oni;o&Q8;)#i#~)}t6{Y(iQagnW0TyRF zf>5<2IRijgB)(#7tEpUTcQ4*a)KQi`P*BM`8CMuHNl9Qeu|uz*_H*TVHf*v9nLKK}aTwNeR8G4g#!cmu{tb31an_B@X)+Ujg~(%M50|=@lC>x{jE;yKWCGEEd39$`2R!hcoP9(XaB7Y=!zfNoax(bl}57zpJY|;;K_{D8TMy5Zws^}K=Gy#%g)DMs%!bZ%b%eS`*CML0IFdVN?3i{P+-f_(OW3nfr z?Z?EiqH@$cR9jYiBq+`L&2X6BAsIo$P^TcgzB z{+F>Qx3gBDb|W{)OWocamun*t@GR0;XVWyPP&1D35Uv?U6pEf1n4 zibiL*Ns~i02FqtpB=%LPYl}uLs7}jyS6)mnuTF?zHmTEss(dWv8N)Wr=!e6&({Y`7 z(?rFDSgEp^R%LUx4{v4bKQcnACxM8g8M}FoO?ct#30hvrDAInWXW#qCt_}!Rnns8u~SUS`(kXIXeQ6>BEA-=GdX4L<% zzCzk2RagKqNE{`DixH0Q0WU{r$_;O9+A&aPZPuO_K8E;9#<;Yn+tVf?R2R7c3a9zL z%>xC(t0h%eY8!Hijw%R|uDda$NqKJ$(%f}P9r9RkolcphCr4lc_Gjss?1)}^DD#wQ z@?~jB^LW(A&$b{69-aziS~L7LGU`voJ;(_#nu>hn^5w?bCAUz)^<9rPG`M3y;jq40 z2iC2vDHDSm(~I4{{O`jF^wb+9Ny_>tx}3V+rk^#=-1{q{F7nF`xtE~*jF2h$8R6X zmnM%oH6dJtW-p}2R_|p*5z+HM#8quwC}MeNt7a>oPC8`-y(me>36&%utx;5mpDpI4 zZ=(J}-am=!&445?!qO_lift=v;X{KxXd&?PFd^PYWiWQ5gvB z#!T^Rdg6Uqm8d@5CtefT@r7S(Bc<&_%dWng7s{tk+aGEBrjwDxrq-50TFGy2muke$ zg9G;TBtE+vO<^!;UT3s#pH}psV6WMHB2p!Xd(h)IxfspUu^ENzDwMPiyDCDopZwNI1%x7G$aQ_lZan^scGt6@v0a}{ zjENjF`{9~|svR|B)?DsH_IF%*99d5WR?6!}f=z;5( zt{brDJ}SYMfLFMnkQ}k4FddQhS}v#`qwy|tZHAjFCT^43H&*&@d>oGK<&FV$D>K4Q z)N7kb5iSN8S#^OcsXRU-?q@$rEkn1cBV)%Ei*VhoC}cl!Di?<{;YPB`u1eH#7s9;c z!$yd{VDk}YVY+a4s^{#xkDC2iDOr|ic4#^>kY9n+U$zI)p2%z$z`n-F@N5|fW z;5&fOTY{kZaPHf!7v{E1iZwkE03wy2)zba1fCu7hnF&yBvri$xDw zJxtX;=mnaObMZ!cS$gny`qzV3Kcy&77}zC&Bc&h)rD0(4p4Ww^jvNF+SozrIeE%xf zY8Lxs@FiCt5k+360o4=+mWyut*}upGNL-3Vn~RuXIZaspvW2n?2z16{@B`m-|KlB6 zuC%yDi94_%Pe;w<7=JWV*Me^ts0v9!-bf8VOWm%K{xI-&3_b7gE)^JrerVdIfL+U#f!W9ou>{bI)F5Ws>W=vh;Y+%;^By<65mWp0or06|;5J2!X zfrq{tp-lD7nb^C?@i1Y913LS2&$mGNV}+j@RgBOBzmvsosl61f`33ZI4m_-?F*w5 zZN_7_D#G!!Ow)_D86ILU7kg+k4`6Kl{20DG)09FM*ajUoR<9RmLYn!#!VZ{VhK)LF z>%2yOR;v%=xS*xa0e=+l#bpk+&ZEB-5Vh~!D1ihP02xI;0}-BSWp}tmvLN1Tt%xc$ z%2x?5=XBYLWT8`;_u(VQ--qREYsq7D@{U0_(UDGd)U_XpCXM`?xP9m&+&Zcv@%od# z41C!lG$Hogpr^ZT0ZuhDs%FApYkvkrND&rD%5(BS>vco8WZ9HNY4T*lMz^Pzym$Cs zo7Ef*4AFPjiC=$;K}mT$V`s>pXs@m|vi1Ec0bpXkSe322y@y67p*z8f;edF*0sI3> zMg9;y>|q4#wOQF9rCYt!p+!`Xdpz7z6_GHy#y8P22RF{<-YrwV3?b2U7qA-4eDnRv>K zk%oS1KgXv|h5LTl7CW{Dzy(>)r%S{`c}pKaBnA;6HHP4>0>> zm63(x50w#)KNyhizkApmziyYj+54@3(=|0uO*{76uUxnF)go#V9O80wh;*!fPK9+< zYL=V*qggJPyXav;v!>qd{8?iI=m@<*n$hr1HK0A;IM1CHuPkp^L;Day98>n?rgdVT zQM5+!=Thik0%TIFd?c;$mn+Ww*UR-hGFC6T-R9M)CA!K>LE~`3-S?S)a!n0 zTdXWLlk1K(JkkT|Oy6D6UA}e4 zT+c0sRVwO0E?SjM>(9H(ixU~)TT%I!8%^T`Flt)IIT+*^?WbMd$G`}3-k30so_#uV zjyeL%QYX_ek02C#-Cpx_zvnVY%O5mg{SqJGTDYN=qM_6pr^F;+UICdvG1(=8)%H@RBW=;|EbLb$a_daDP98_A(R z`R-Lb*fn@wo!i!X)}*87oI1v)5he?04X5-!&C+ zB$}%bU-h;Cte0zCRkzk%#!p*<<|?&14uMPx7kXrc7aB*I&dXR|@KlsxTJ)zCx!hG= zmKttZo1@CfiLN`#%r-%FEuoR`3(W_;e=>r%s926?uzyLnUX!{STuE{NTCdMW0N5eO z!j}1n5`JCGaD@3krxn6*eAe+ z+b|t)$)Q6PITIEe`%q7t@FkD((>%^uuT0YeKjIRT;f3b^*_bG*9n1OwoHd!lrVU0TvC);If|qjYYz{5sxX|8i7z$af18!#*AGm7fv4pOV$jv2K;)a2lq){rk?4HK{tZ{xXh}b=kLi}%{Tpx?wyEYtPTf)<|H5UV zz5P}>AW}N8Uu;EP1wELJ7%qIZCSl9^8KP9)IlPD=_mPb6rr$KuAdV6a)sY_vx*j`M zZ7M0WmVvYo###>E?dt->ckU2q{d5}S<0q#^&pKok-8INsh@n{^-8O4*OJ4yGj=RxvrjRH2!VMzi^2YqWtUaN{#HQ0fXua->8oMXarY3eiO zH=iM||A1I}Z`UcC;lmvO$F@H9>s6>R8AF@g?ns zPNUP%>HCB0EcnBv{7`uL61L|vb`?(5h+CjhYhrL@&*C(|X91CWJg(4hZTZEkUOK<+LSltJaIMn0Mz@S4t*hE{D6Tl z>u=WsfQ$GJn+}#(hCo!)oY--JFu{@NEGQA%-^<~ZCrrTJ*m2;gNl&5i73L>TC4VxJ z_czwA(II8XH?Q#H9IJTG?QYcWd& zbTn5@_qJdkUCGCoL46_onp91{o=oT+tH})5CL})X^l{2#X z$=;)?ip1U3$EPgbR8G5=i8{7Tcdx2AJ6O02iiLUy&FvM;?a_VwDn6`Sol3J;Vqo@q zEEpcyWL{3k4E{ZIv)(=+pluhn_Q>97Qi{Uq`wHfa_j*u{lMb5;;uX;f$9(r^ z3e&Y(P4hcX|8G$?AhjqcE(nFea2wj#H1k@@gN$1a*w0OSthsn&!YfJD_T0 zWA2O?j<_(8$>$XYKi`}pA{AM{}0#t-$kse>yh-x^uhGe`#(s714?6i3 zFMe5NWMpIe<6{1#_5=sRPqZhbZI{G;8Pt3|sUrr{E{@BT1Vayfd<;TQA7lkbQ%!Ok zXhe0KkFzs>R$?&~L%q0agaBf1=Ax`RTXI)??|h-$@~FsJ5U8W1(z4XAPir+B_JI+F zs~JL=yxN&`h#AQrX{oorSM&lkGOQYckULkmJd_#klCXg71?ia(>DqLm)e404m&PSQ z6mulvBC#uapDcRv>O~2RID;3DY8HGd1DCZBYFfX`V{s4nP3eZsDP^h@{2CsM@*`d4;%n$b%FsWAKq8k*_Mk zfQfSh>vjKS**Rr*wGPCFa{5H`CVVuxOtE#sx*6+H19lEm5qxvJ(@JC3ez|;7<$5Zd zeQ_{nfxIfilS@0i3E?pg4bCS2W=MMM2Oj7FV=xt)VGK z=1Y2O2Aq8H_sqtTDG}_k5G%TxXOzp2aJzi=savNqIq%!b-MPMNgwkv#;(jF3WDN0) z75D8%mB^z8Z`n?56c87!h$(E&eLhSAs>C;+{L)^Lm zt`wJo!S#n;nFGAKF8LDxO(mQT78(F*kK!)tSKbJYSG@Fl-2$3BS+6Lb80h}W8u<== zGD{Mv@r_e(YcLcqd$Q7488>RAQ*8oXoO4!(#kDbZmLv{x-5lU48hshuTFkNqu@VhG ztlpzmEan!WD=Kf)5)|Jubd@uot|O&ZfLgdbmbx9DVL9&&>ov|)EA{#J2>VLJX8R$k8o8NmxHM4t9uRbbP7XPmC)~LGXFn5dvYrMP z^i7MAq&Sf>J4N@7i|xVD=YKg}Jj;Zy=2VzS2=4@`(e3RA8260 zH|+l3qZn@mvDFpjfd&h-fB=_E!E+bY#@qXt_5)N&)mjTalvNU$x&%oy6?tdi45L-* zb+u}1-8g@8TwjdnPS49|wXU)1<>7MG*jb*(%#I_d|Zn+J? z{pi{;f@XTa`I({CSYCnPd(!?TVZapYN8 zCJvzRYVMD{BYXVZ%%N6gF^mvf^rJa@)}m7r6cHXN#I>phvUjN&A;1W)IrNXMn9E{A zS)_6*LY~L<9&#kmpNPB+4#QsqvUgDehUj zoxYmys}1F1@29?q(l15iG)gn3BPL1Whj0;mH1Wm)L_i>g5JrC}PY?7Z0UijyiM9(# z=NpAbcz`7K82?bBxv{GM;#7KB%(vfpeN2_vwJv=uW~@>E{G|{*kRl4^8L9VkdZ?Q) za3bs&kQyRg&8DrdLwXsdzo09vfP4|_RjCL~ndj6y43qwxkr?5t0G)E>y~WhSdV#67 zuO{?c<%nDgKy+gO2u_Ia1tkI!biOAq!7p?y_jocjJW}>>W#FVg|{w zjAV$TKP8iymiOi6)!?Bh;Pk5t68TpgtO|a|Z2Cg)ojfzqul>z818#v_u&i169I6)K zp9hMuz>zvKNJ5Ee&hbvf3Hy)|ca7~=HZH!rk2YGcsd4t}ru*`p8xNm7RjMF3;*kc_A@gk<>o?&wYe039vF8|WLz%l#p$W*c=n2qL+5@5_w7eNwJmxv z4K0Igwm@fbL{Y&%X`npTp1L?z!)RLhXsqr@T+dxU~XNT&!H99qgu6?s|9WpNnsAJy}Y=EWWd z+&G=TzT76c71KI3QJ`n@E;LORE1G ze6#c#IVNXXec`x(koI+a3*zd&7MW?@K5`AK2KUMqx~V`-WSaL!%-`LpQTJ2hOjpRg zSS2zl-CZPU?@FcT;0jx3<2GijI zN1*TkreGeJqq;ksYkoA(ztfBIeX3T#R~?*x=owah(2p=Yry2#3oJUjfTs4I11ztPE zuQb=@<0zc_`Q3|~mE`*x{4~*3w32;dI{oy1)!a|S6ByI`O(sa3Q=d2aj;*Uq91}ig z(f}xD?V#tp^koa1>7~iK4fe}(W%Tk-avg)6T-5K%o_vNAOvbf2(!S#NQl>ghE^~Xq z9jJhP54W8bIA z$>CajV8)NY>vi{5d8U8YmVr}!g!S-3%$|C7 z=iJ&~N`M`vEO0Ui(c6>t!Or`8UjasMD5(A`Msd*pb@7LPQN73TqnPiv{quiS%=hbW z?HK7<--6~JHB|p73;65O&w}Rv8?Jiukc0jg&ipaz>+f*Yn}vVLRsR66|7+q991MTG zA?ELM6+I*K@8S?#Kb!6Rt2@6v(7#&F$G}Ly!SFlP{>~ylmVWk*{2O5P_E-OVVD-b; zuMYkLSp6WQUsf5JnOXn1m_I2F!TcZD$3N0Ob;-IPzOkZotbIpuml_K}m8>>8Vj!`O zDKkIlnmZ}=nBe!}G%To(jqrNV9^}`mguQ)tVlF{^v!ImTgp9jBe-%W4*Cvw$7|3G# zVvBq*?)eF@Z6ENa;psKJd-Ecefqe z%5XhqX(9OSzKgWuPg)0(=ErQHar$1t(&5}xH|VZ-qYCzjzynpzpT|7k_H#t#3*^*qw;GsB_yZqSzsOK!wNXq@v2 z(_G~fI;Bt8D`hO?RsEuzYRy^#9LJ55%eDTE6tEJy7A)ej9*S6gwiCeIRSk!8nuT+qHGi`bHxr;e$L zkti}t#cC}jAFHF1O4CkMgt*PA+l0nz-=BKwCFUh37ehPcNM&D{BKJQ4wR%Jz3w|g* zOOGi^&u&-U-yMDC5{--G_REXYVq{MD6LBG3_j|{-(o-7+BveWuZ*=DrY#3gCBgbPx zT9&2O@hTW2gAyiZfH~oF+=O~0Y)>(k{zN6WOz0`Za6vd2Oa?5_B)6RJg52A$*(_+E z)9C*ZC1Z6JHB|GHrs}f0!H1{-y@h}vtf(OSl%s=3X0M(ht;~{wWd&6Xe&?r0zyyr` zHXb?ms60NU%Nq)T3@Kbu<}SSo8H{L%>8L2XJ1sZ9&!z0|IKJ*u>CI4k`Xk=Bl7r6B zr#g2P3ZIC4j$m<#&nC4DOGimtURDlG3Gim3+6(l8xP+`PaHB_to6g(WyMR1Yzouv4 zOy0Mis$n|nx%)D>OUe*+9T=6dg>tBy2*tK|)==Vd>pDOmE8hXPS5-o+%iX(gI*@f( zQVtje?TdLPLa+zqDAxY@?Ed{i7(W%9)rf8$jy0kVh3GT+Ie9o|e2%pR8=`AV9Tfy? z3fE!K`oW2|K&9~42h)>)6~t{AE0Re4a}ly8DT4{wZl=3JmjrvwjjnTvN-DYYMDC8VG!27{DaxOp|5LsN=i0?;m$75Rgb;4pR6r$@PyoE8f)S%JZyP4fq z;h7T_FlO!H?%gW9Hl()oYVY}Yn^@<>1+ykch6|RaAe^5di^+W*HdZ%83e7yE&};*T z{ODni#H7hM)1Nf@@(mcK#^=)cc@0Z&h-1OR8ER}C-;}-KDZ8KzhQ;uB`PNS|6@Gi4P(fJpV^81GrRFaUI-@UDZ5&{-f?l<%;&XjLgi1*R>^sY1a zs>Tinj0v!|T?iAAq}DM8%c*XfD=2$g;Toi~sXi#T;k(U|aLO+gIuO-}E;SBW#b$Y| zDy4m!Qa1SVp04IhJj26e7d$Dx`c7;mUmMvmNkAxULXb{c$4D;Id(MsFvCEHu z^}CV&Uahn)G%lT^t>z$?+KD1)(`#230LantX3#hr$&qAy>9;Ei*sLHT125$dnUAz+ z6a&o+x2USH2})v80sEgwc93LfqVy}bXr52oCG1;vKc0i=)k9P)dkcu~ix+-t&Rl%^ zyi?ykZy?j0Mh_jIw71WjTk}(i+rg(5!~j3spq}q9zPs2{=5GEl6wGTT@`(W2E_IGe zFwIw-g{9y4fP2kjzVtM!D#k8Btep>R+PGF6i zB>SDp6MG)5Pm&5jMZ3GMxv|8$z2igLwUKCjAqtAPe6-60W7;|tqitd$J4;6UNn_yH zdf>0Xi`eV>KZ%+d{$KRkGcdnx>ofcasDJ4t`EAqr*9h8`ftlg$;AQ#2J4Jf8LJR-j06k;r=-H z7XQ9&Z~we52jKBNy zCLzG_R`&Iq)wl6M98CY<&s*a?$6K+~f1G{uhv|2JSl;|$`r*$%QNxdW^s|WY|AZPC z-rClwk3;&85emwI3uf6Y%kE-bY2Bb(wnsk@mlkEjUFAz~Y@Xk40-yKuJj6eQJ>0qX z%$#0khRhX?Z+euX{MlFYpL>W565j~Axm0Aly_U+}rA5cQx+3e-%KV#aUeuJ;gY$Q4 zUS@rzFJAg*c)>wm1#bU(a<|6gf`8u?_G{$d>-PWkIN_U%hu=JWquYlSDh?~r-0s^= zAGUC4H^c6}yR5l;3_E)&u1~ekdz4R`*yzt89XGVOF=)*j*dN#8!-^$~-kZ1cu>ZW; z!QJni(uPK&LAu4P8E>Ujv{?ph?QcI@PJ69(#M^yNe%icL-H&T^uABJ!#AQV*c6zH{ z*u^nFO!}tVqHkCKe0B(y|F)yz7yG*$`D5_weOi?`)9PHA6C3l-?%(8e<3?XJKQLzbyHE7nUptcLlO9bH z^Gy#db?2W>6UJ41TyV{ko2`z$F{E9yb~}fC?M_*^w00?hp7$h^1dozTmW+Uep!mVD zc2GUW5e0^m{x0ha*G{p%0Xjz|U6Cty3mvQ*(q(?@o8{jh85VZ7UYoN|f6qJjUgVsL z4YhB4plS1@;l>UVn*RO!jm>EfZXLSv(7$(7kMdD{9tZdrUe|8fzLqz8cUaml7uu+% zuKx1u)p`fNI{M-K`-{%!NvM4#_u6@X%=q>Gw$+Po_x~+vc+A2t1_%AUr_M+F51rgn zBhP{!Z_ZqOuHZZO>rGo-HTGnl>wk8++IYbY2O)jdGx|9 z=g|Dx;Z*74DSoEib6XtxI$+qmH4XG-C*7TwVD6HyN67T>mknQC*>`MD-Zy5g+BI&% zl)JAiTfHK9?BGMc=G}H?$Zs+8ean|A+uu;6+~(v-5k-zAx9{}xVMCjl-I)}m9{`noA*0eyUal#(?s78<)+{h; z<^3%`Prs*)$p726t-XEQht`s)qv+XrU4rmxwZA6vd zY7JTaanwieRo#5=&bOO~Pd$9=wVwjkyGz(edj0t*055~!90lN^0+5Vvwn^qz0Z4{Z zmUy5SlgHBy*<#jIvjqqX1qY1t7e>Y;%)41%M6o zQrzV!H{~b*PZfYmn*Z5`00`KNs1sVFB9e{3**r4 zmbG!h|INzxTnw@|izNlzS^1ue{Co91h0=b0zFv(JS|KZV3b!yT<6Kf4Ug4QVenAIf zmheQUV%ElGkzdgLn6>X^kzb_VLz#X`u}P;xSEoxms|xw{F?kScHyMZ&xkTbHZ)X z%hC~a(cLB&@%c|=)oGQ@R2h4S&VThr!132h?6~oybliva224x)Yg&P$H4E2Gyjg4U zv6G`xMmMbX#^eUCZx+4lepdp*iNoCHky5)@1nIW$PDiPun^u-ci11_pTSh zH?*!@c>7+*%;e9e&AsesUZTbR_91!O*hfa^-(38C|I=qu%oPHxmpZn6OZ(jqjyd;7 z-pbwPdW+}&{#*Q!GN7g4{cLv5bx;44tgh~5e zuRm&W$(=z(uRokK;B;Wr^;v0qH!kws`{%|l@)!PfYPawr#s2+f-e+^hC#P+1`^DWK z4;3E1+Hb(}v+V=cudem-&W#)W&K=7Cq~xyoDIb1NeTrlE{M)aW-+rg#)51H_e*VUR;A0Y-CB=cUHqNHLjuZm>GgUCYtbJcTLN&!+`|0H*Ng2h7Ie6$w%~y1 zfIjh^XXv{42Xy{)w!^TXs(#wM1|fdmSN~_-PqF!`cB?Zy{C28;pUSU|47%~-iPire z-uH0r_Xs7d^J*7vvx;wgRmwb1-*2{l>S)O5Ry3 z3bjgnCBA-<5Z_CO4lU9Z7~#|ZQSL@%yCg@rMr%rzM^Ee;T_Mc-8p{o!o6Jv z4Y^VO?Zlqnw5jrC{rIUJZ(Apnwd?Mc_W82-h}q{YZEPPeDc$Jg?IUqXeLo-A{qrqn zABMd0YpomYHt4P6*rtL-~2>Co6g%(Ce$rx3qR}Er*^*kO;dW-YW6rSY{2ya z=X}E=%66I(d8Xs;{pBWaSe$(E=gKw47yPl`yiPq&tttMIzG|7UGrKgV9VaSm-92IY znW!nJQ*RsnU(+?}bgS;#MzgQ&EYY>@F8i9&RloUZ|69MEKd@-k#;zZ(T=4SL0+&C2 zDfP9kc_&R=QDa4Hn=ebh;#l9X%~jvT>i$10TH0=NrJD~N7DwwX_Y+o+`{{bSmW~p? z{Z#bizUqfpS?}4Zwu(7WEp9`Z1Kmg6u3Gx<-d)0qcRS&?^mOO-O+IXO$H(V($XB&4 zRr_T{wKU`U0p`{VyRX`?ss2L4?~ZBR+b_S+`s%*dhc^fvys2f_ZzWf+xa{-#xGkMa z{iv!ajc)ZZ`i76&9PeMD`PKR zS#q+uvHv4;U;qAtnvWhctJ(1uwf|W=ZRr=^7klT6?LSPsxTDIdCG`?(4%z?RwClSr z-$GrzHFCJ>vVc@A+TWRBqsCrMviTsR?ffG&pl;@6t<`EKR>%+U~dF zU2fz_t=OQwy;|t*!%ceqQSQ*S)O=Of%vt&J#zm_uZ~K1P-eZe;tXe5gMOt3Esd*4{%df@4Bc`K;d+j{- zd&65FluS;leQWypXkM3nRm|azZM5S&`+rEf2YL81mn~; ze!ZJ*{lIbc+w02)4u0+7;#=FQ4;^TBG~WAf!t{;{Iw$xJEtb?&*XUuH$1P6S?|fM} z=({PtMJ7$tPK!#t7})wo-++`8`+UEtw*X?|GT<4CuABKJvv;ED@gS$j^ zs1iEAlJ?2{aR04A)pGBNuV(x+vEsak6HeBMUvq51#SMa1C67gKnB3;o@x%U^pE7Px z_r+l!KT0WhxAFea(UDg>9!|=CG`Vfy<=K6&>GHO}eI?JZjmNiEI#~9NLrd#*%{S*j zk;I3clINa&v~YFm_7)G$jJUa@{L%#}mNH`oJ*Yh9ljBjluK6a8T4!GMUa=AV-npJX zF85z+9$h?Xymq_6)b?ZgTv(YL)V-UoLe0N=_p|JnT<=n!*Ommny>(BiAx9(5ti81{ z?D2)GL)SkV{P*7#dL3@o@ZS={9Iy5rmdo;aO4m)fzA5#U)~Da>inCglZ29Wmm1QlB zA9iow;?U$yo$BbOmHcSN(q5M>rnmYu-gV$Jp!NRLpI20?X1&ombj!6K`n0m^+l5Xz z)?{JZWxj(x+m-lN!MtaW6bKw-Zn*us(YuGed*B{XLGq%iMU#$Ix83&?ERCrXtUtj)!8}A4kNIZ8r#kPUycx)|za}?Up^E-LW z*yCfCY7cySk{LUG@b8lElVrw*9{}VZ*zW^?`~#bP%yI)mk{Nq_%(9o6WVQk!8$j}d z*n4D_eW@f1{5P|l^Gvc}-w^=mk6C~c0P+umcm+W6PheqDu)t-hHj9DisZ4S>~mj_JwEMz95NjwrM6&htW(HLr-az)4}-P-@zyufNZmqdlew7 zcu(0Wr-#|Y9%ji-XPaj3Re)@xJ*8QW3iuyT0WR5TB}@6Nya3q)^pu@)RDh={Kxo|O zqyl6Y-cy?8sDS?g72r}ZdZsGCW|N~sPq`^a19++dT#5}@Cz%^3z$RUQg9zDElI19X z|9%A^m6qkg{vpCYf6W;AMe1M*;i~C;*q1imaOxw+cXF zu^a{P%uh3tQs~b|0puhCU${1AsRom6ZgQ&ta`J%Br;S;vlx3S_ZasjUG~kP;0G?MG z&?enq>_tGC69RY|0z9uQpv{((1bp!nz;jCiT62W+Qy{Twwx-!i*uf#)Ek*?ZgQ&ta*}}0#{-Z$ zShGzsw+bL93HZW=09HYBnVY=XIBFFPl$CKVIR~%4=MwmOHO?i(@oJolpY6>!DU8m- zx=0z5tc-K1;&_FpNYQ5Hds*ZcTNe4nmPLM%iqKh^Ul#dA>U#11duW^YX1hqn)>913 z+E!`|%o@vJVAgmB1GC077??G#!9ay=bUJ092Y@a*M5%aYdvi8k>t*1az1`VHbC=j( zE*8q$GS|Q!CH(pRBN-o0Sy8<`TZbEd@B%4$} zX>aG7cY4K*%Ts2}3Q4TDetO}a?{#09rfEJRZQ|~owG5Ba4o|6S9Z~ts<;8n0uF-b< zTi2F7IB+s?)7?$;wtiOkjVpSi&))uZyXf}p-BQ9>e@Na}Y6Op~*5cEZ;m02Ce>if} z)|AP{1=>rsABwHpw}#(r{Ogd?|N2@6nDf;7Jk@Xi>+cpytWk1&|CH7#U-U8L^$&lz z(XhJjt2?Y?5(?Ltp4ug<#>q*Mkqg6PR+iqq{%F-{J7TUz7MOfw`je>YpFTbI?=Z_% zU4_&?d-j@dob>5}Ja70c?iRlGQqcv&_Se>&iaqvl$lc?P>FXU+YubYfhpZ_0tACq_ z)VkPqT&%-)eZT&r!k0CVUVYH)=Xd%aXu5rTiLdij=s2WjyD@Fs&AzZQq1UFA&q8K~ z+#7N~?BgLDH#R;Kbn%mtPZs=aKVRacEq+4NNB0&t>)77VzTsPq*2lN&@%7O0zYgv? zcl0kkwjZhaTgk1pZwE~=KAKmz%RkpIR-IY*vud@96zksg`|@i?tQueXK)-r>FD>me zQ&-8q^&<1;?=RTbHF(mr+?%)ZUW_)6_~!GHjpmMwX!B_IgIQ}X4on%>xn$vb9|et_ zk#e|0%9>JvEnY5Dufh9|JD2)6)p}NcVc)!@pPP)^G~-Is&3R_^uy4}Z_2!5l+Ztx{ z4r|}cUOW8nTiY)mtDH8u-RR&p8#+Z@-#v5m(lcA~H~sFxo=r8jj=S;nqpNkMd^$KG zxz@Rxl?QKIUb3a**pJnJHQX8SBa8&R_)m;tlxL__jbD7Fm&!n zf4}id?r8cKh&9|>x8$|M`f?$;pVqB&!S~yI74!wdM=kE!I?kBCXUp-;N4=(d@X6){ zM@Bumd2f^EOr3seoC#J9@W0;03x$WBqKnSEkHdUGDV2UklpzE*;l#=IV7Pw_Uy*bl4iz zb;hh6m(QR7e&Dg7q59$jlkzWU+48l@4||*|aiDl&#M|pqx^KR_!ROm5fnDY4JUZs{T-B)vEiwQUTnmbqcUXv%;=jhzs*HR;6 zR~%Utxp)3*t8c}(`(1y_9Ik76tE%n6?TL|FR^@6~t@C+J`4h(*ypt<+UhAiy`}%eJ zt^2y|kE3tzDlqlf@7)eQICXx>`N{818_{^#pQm4XI%M;-8|Bghn%})1XqX&4yk+-# zx$YO=n)b%(f0pfAHmAzoW(7AsEB}G8>o{U%xv>|6@A*7lv-olDk=FPL`@%k} zb#!0TEzJU2R=(P{>D`bP+7)Y-*ZIAAvtHF7q*Pp=_ut*AYtG%?^vU?N79VvS)_lhC z4(I0hB+h!2f7QY1=iYhydLum0;lWEIZ*Inh>OIXS)NGpn>G6(#e)6Q@FUvOOnm?+> zj0-*+>~+jjoBWVx*@w+){WVxWDq7Q^SQ)>QHO5wqpJBgrXp&=4ss8T|T{viL=PgeP z)*9b&(cywuZQEi~{9m_^&!s>4t}$L6iO{`&V+iJ$0KD=;D(SHmCB|J6qSj`d6(oU1l%% zr&r%s625%ZKKta{9ev{}6}tO<+hVr$!)vrI@aC#>H#7ypI@LK?`-h!rd-gciwCz;x zar7boMacv6hOR&8U+Bw;+de%tVbtwYVfhApzHNBBuoFQ;50@`h@Y43T=F9|v6J$MCvEBN3Q_@&^3 zSKya|k2Fw1NA2cGGj-?Y{r}kujk&@9lBO&0%))(+cta+8p+Nzc=tA*nz262UEjQg* zi#tJca_KMDses@oCFq2L$~r$uBFgwFXItHK$)%{P+y5NGf#;VP zl`G4!B08ifXCs0OJdNA_FA3R^~zE22D)S+yc_5eZhALR z_>r=U>fOMswsku1&+?>_@7l|k#XGh@4i{za-5gz{*1LgO<(w>D#f3U~Ryjv1R%e9) zTyNu2G92DC)Ed+K5m>R)%Ro7M4Yj5Kxt}($O^FKb`c91Au(`UQ7B?kf~a-De@e zNLVPX9C8h~E~+ouS0W9ST4v-s{oB}YkOiuu)V8) zOO0H43|5Q9X42!1 z0JB!>V>Os?uaa48T*iW7=Ke9=Vw4%T2#$}4iN<%)VAColAv#{8)A)9@C*e=6a+^VC zu}JY6IgaiY5g%#S)M=3rA0Fd~tE)oUMVMW16Hu64hMdJ>@X;D{Mw?kH4sZ7mIW;z& z9ri?^DqXaOZ^+1XwTIdTVeM-CT(4tu!B3_p!_R24`55$8i`nzVGYO7##EI3D?r;N2 zZw99Slp|VsZH@AxxXwn@we7K_DfHfY7|YaT9kI<~s> zR(D{R&Ix3_fp?u&xSt~;Iw2+@P7@px85tN9;|Pq8NVIFB0;3b+f*ldD@tP1vU??66 zQ}K++#U8s%LrUei;TgTb>|@aytR|a5C>#%`!SGIBW}UJpLE-7(0nssoBJCkz_9$>X zS#KBn?}DhTgc4H8#A!qx;wdwS!p*MCk--}B7bTqKr4Nowh=Ujtgk2PJ z2Et0q6-U^t_rXmkdaFq%5<4DYY<6LHSp_NVE(?`1BblFFXXMK6GPz|sw2p`iPKXmT zQnQNqglr8wn2^qB0ITSYW{X8wU{5Awbj}AXp02gh=L2P!{!^ZhaujU(d@vI{?@AX< zoi>SqkqHn1ItTF{HU~)vp1F$5z)99B((W=vOA6NMkPsOk5fu{>7}-D*CH~$pJ}_=T z15IFb2n;3cBh;uS(jMrDW@78&DQ3_@@>t!?12ph~?y(vzW>1;R=A?yIxuQ_vEm?Oe zyd^JuCcGtGG~_zzZsB%KL{w~GaJ(iaR1;#i$2zB|iHOF$?I2rpm?k75&K?+Nmlh+$ zo@kGZi51{z`=In^+^7h*wmH*9F2o4g7Ng9___ zS-Uc*TF6>OMnp#h2S#EAf+K{iK~)c6WOi%Om1Yg5L$YAeGel0SoQ9*!NbU=h=?ZR^ zCm^_45;^7kEJbZN5cw&EODGN0Y;m$c<_xb=IX}ZmR+WHi>iFBE z?HYShtONK{?df_iQw9+YvLrA(b#O2Z5HOq0;Dr%ph{@Vhxns*omH_*}1UQ*uX@$Z} z40J>U!cT@wLA2Cfj_c(_8HlIfcIw8^6yB6Ip~9Oo?K0y{87G-H+eODa zVnPz62v!5Fs50ehnu^abuS<3#7)fwG67d2IinLPJ)wA8$Pxr$H!wUc zHLh-*3T2lwBZtw3z{FrQ>*0ihKRjW$3~QrYwU{AMGRHZWAHjyx&GSH5TrUK0(QUH3 z#gNb1N;7N~6C{7eiF`NVtP1llTv-`=brk4&8omp@GFLKu^#;8Uaz|F|ffC0Nc)8e& z*A+)6JximAh7${9bV5uwtyK#kV8Qy+QLN< z=?xhEFr*rzLmDEh7$2@d+QS}+1PHvWAjyZZN5_dANQ^@S95L``f+E$%#`UQ34C-X+ zm1dWla6-vnV(gX$>zN_j zier;OD8+op>?Zp`PP(;;v>@Sdh=>9KgdZcB{7H_8u<&?c0Ks=~oFhRP@sE?ikAnzx+**2N z!zH7RL1*w0Qsjki<1#g6CQ8oGNv?G2D6Wr_I;{ia12u8s@S-7C5pkOEz^GVcaYHmg zgAwq+xe0WH$lj(j?^s7fVjyz&!SLrI5agtLc51cZ*6u6KMQpwZ#C#7;G0#3n->uZ?~3R>EmFS8;urNcB{^S%n;`6-Fhpx9A~!P%3a4! z%sa$5V#6~clsAg^tvGoaSKvm++n zj>shrT1$w-U`Y;8|B8~YiGtrI`9pCtQ?pqcV(wf4AKi< zR4pjdNpBgBFGv*yVmZ>IGJAA*AgWQMT;_U2p$F5`S#>@ZtIcLKc%G8Vn1IZ|PRc0P zeLG3f78q-~wN(fq>=P zR6qnGqJEXGay^>QW%V=lRP<)#l5AEJ)UDULN?B~@JSg)}GcOgVrK*{*@}a;VwN^>m zAuA1KSrPOR`hd*Kt&Nu&tlfq4X0sM$AA?P2fhP7=gDFg(*}RlAsFOYcA_c?5+^v;X z4q;$o4pZl2gui97+B{8;Wq2iuTstv#k|y0sE5~~~F>d2tbOnDiH#bw1vz6gem^@RI zWv|zXvLXb8FWlSa%mvjfR@nkc6&*9)SvH%j=bb2N4}hCwK>Af*T|FCsmS zgQx>(Y$30S5%Ggn0_%EaWQM>>?L6)nTW>%C)@ZdD^k8m}C%`b4HH|VOX_4WE+cnWK ziAXu6mmK0E!qlZo*K;Z}a8))XGM3I-TqoNmWo=KPBTr^Or1y$HkhO^Stz?k2lLm5u7z8h7|KY!)9gLS>6S3wZ#gGM_;SId`E5@#&$jMt$|fiwQMF9t;xr#v!Pok3t_Ny5-o%()0t3dkiveK zS*U5?*3L?;hHeGd`{2crQ8!brWKPfA<1o*3)>2fJqE)3xSqcG^mJgXoqOVGnOk^8@ z@`T+H>p+w*w5)nsZf$X<)6!dzltqF`XSIkU7`&1Ylo5%9D7`CAUTT5FWSB4m$Q~i* zg6KXj7CixwZ7FT(K<*{ju8BjYOs$;KSBMFE`az&>s+tD){yH?Ep^`7o0P=7ePDWDj zPG4;(K{Cy3!gqP zIGirD7{zBG>oz99X5{TGVt<~=>T&ae5-B>@HPabTqtLZu#|(zaYyxf#u+m4%CXgL} z2=mT_;T6WAY2g;a+mwV9Eig&Yr@W+{E=8OInMI2RgVx80)z6AeLmU2ooPbu*Q%(Wy z5EvgEF60-528eK2m{Oe(9U|coAtCl?;qO6wMWYDSMi580+$Cedh+{_13&}>C71Q%r ziqn;-PC8p!WgT2~rpStB^v!9a>~J%J<20dC{Uj(5LK2KqI#rOR_<+eX2i%& zz0TRxnCXltamb9+l#t6`>e;xp-%2AFBM6$o1#1~iXuJ13W>exID))-VsnCbU60KrT zV34YKQj*9Xr|$So?|No3pDFU$sbx7kkuyl4PA@9mY-==y_XE4yU(&N%_)!_Q#0ifX6oIU{bNAnOMmZWIx{`Z(f5fRUUVK-piI}#YC)4C!Ow=>Y?N*`YBwuu)t6?@ocxsp)2M1Cc0HYvPv0TUMwH7Qj9Sgli<^k1iqcn%+)KmG;>&!;c zV4?a3x-_Yr>-FFvGB>^6s1@}&)mXjROl_WCZ?%a^wCXv74xu`QVb2o%hQZ8nHdxJU ztPzEMIu=zZs!y@i#YFiTA2164UWK0tn_ww^;66qp;f4{u(hHkH*TsazV&4!u=jeA0 zCJ{2IaWR;v9!JzemYis(sbajn=Itp0{ks@77NDE$zaUc^>v7#4=j7rn|v zUkMc}JVVu3z^E^U=dc@-K4$~>au|~Z8=Me27TTWiq>cL}dYi$<_!+E0X>QP(nXMHt zQDktPn}E@|8MIcEcIk6g6SaAG&Px3-%uGvdyqGQ3b%5!q-UZA+X(V6_7ZFtuzQ^h@ zUj;CW*i)dIFLaJ*rBuW8P;DG0s-omMBh@(ua1dJ;o!(6SD`a5tD*8TybPA*T8}Ogm zcg=e04;gf3J>`4A42%!ZZ9siO12)6ZcmXhk6cj($RSvU?-fuO|=(J+=ut}SF>2nMh zwA_eYsjBB-7{pFG0>ds$1~W6h((9y#S2EU0{Rjh^OsHN1%*Js+?>L<=Dv^wqdZUiU z0S3Jh9u~#fXykJDkr$pv=bFsEs!m(C*H@VU=#wP}9g} zrSTDJ8aA<`P4%1+hniCUgIhx5ZUej~w$5->7=A`Ai!U)&+SRQ7F4R3;7aShK^o9{` z4wYZH<4h-_(~Q{##A-|@0>*SAB%kAAGSHX~ENSBJ1DjGF2gPa34oo0~bnRjDsQxpW zxx4{JV=%nWOk*{`Snfu^m>d~FB8D@u9`h5ZmD>jUaA}Ur_dKg@(&2Gwiu^* z;8QveSV1PM$av6r5MvQ3lQl3KrTYm~vIQ88E6^FG<+zxoyE4>cae@e?C8A`S%Ro;g zoHL3GOaRqkXfvQW8H}|tnghoCFf-0GVedoUf~_COM)?_^A_YL>JOiR>=3`pi@UsQg5W3D5#HJJ%v}n*ABg_l}AA&PrjK^Vm7#~>R%CoW9 zwnk;pf@BqkLAdC5p(m+Lu^@3vc>^$xiv=MteIF97l*e%t4F9f$`@1+vh~`xc$b2zB z3^2yeh<_P9U}|Yz3g*|qbT%SH=0{kOlA-Gi?}g_1v7>)3?js#YlW{e zza2R@ri-n@2_b85g?^#(W`%xX^Dt@|J@6dI#VnnZs*){edz!OE8VMc(#RW=-!?@3YbP3X!OqZad#B>Q7dii{jUZebMwekD`U`&^wvcr53a0{b{ z4ZF9g-mrm9IgI;5m>T0OXJWPo?PJW}04bS11u2=o0T}Z)0OPp8l_$Cr87&r10|voV z;|v&19pf;{2Y8O@7XhR8SHS2RAPm3>BJ>;BTtn6rk`E?g&#~Hx&=7Q807FKRjYV~g z!r*2zT#&+JxIpbPT)0mS&j(IF{RZ4;5`W-1FcyD~+FE4WnZ5$dL~RdXZsu!~?w)kj ziALlTsm%in3r%qmTgy3&(G4br$`OQ@=@-DPG@p(1J%Tmz4IPROltwtxkm@>&#S*fy zEVdV?2vI!*7++INgX&^YjiPZfU{J>Ny9g|ZEkiY#)%OLA<~NMOtz%=+)X3HasaMLA zqKr-X3LXf}Js43iqdtRx(cFW8(cA+LiLo%*hgqOCeTX>GzUn5qT`hjofcW3jAgVHnMA>;%ZU%};P>kQ+~WCkuj_2nU0IG2%p1BQ^s5O@wz34e~+889@{ zYhYAnqXfgyH2^W0?}J<)^+k=MghAI7M53_)p0m*26eA|WY%*MB+Fyd_SUwRj927&> zz{qMlc#g%o0>%cFHe+FVKE(GOtt`HylcVsc1kyBCZ@+lxs37^%4N)F zMHtU`7m+FZ24HNhM7U3BiToO~`^c{`S|XHW`V^HrnjbP+dHw}3R0N0)1C03$fHA&@ zbTHeE4ah9UK?V* z-=l2Ba7NjHe?wPcs1C*1QAAxEz<3GR{{{cq%&x9NlhhYZ*nHy#cCr-`d zfUzD|z+h?EbC?{5QJgWB`PpJcDL$cN00E;ijjWWF z_5qmCp+tQtz|eR@ae5L%G$BKyL=f!qz_H9hjh zlpZ*qjM_$&=#dGb-+;-Z@+)#YRNg=~>PMhzpl328N~CNaIAW1~7v*2dSKw0q4dnkA zelSN2KN~!JIycaReFLhU(ih9ecmwT|wD%3h2wfQD8|W%v@&*|7JE53t)F*`3u4OVP z4sN1zGaz=R^FRj%lR-2ox%me3*YI65*V1{Qt$=+4$GWijn&D8gvG7(H3aBZ#ov~0~*8PInFPDF}npAwY6v@LDHPyhx`W9 zFGvP4{eoNu(=Q^I!QO|#Y`z#wWeb{w#tXPT03~#ap9Kjv2E)!124i(y6IudkZVoUO z&j1EVqVIz(qjN(FkF7IGl?-P{8{-4COwjx(jtXaeGJvuD6#~Zi1uz;LKoN7jhTUL{ z?@^7Sy*T3Ba<-->q4CMOpi7$RVtknCV!)_BiBt=x2U0RLAB*QuwxQoeBLL+gqy?DH zF~NpX*+))>=^?;qOo!qxcE*tJ!V_Tn6sDWT;uy>6hMWxJ7j$ehx*^lY{3gKIe9g$p zQ2BtSV|-=C_A)9T$jQ(gC@QH?_;kKtBD(fwNG4ksFcH;>X4In?en{s~`Gv<%_Zwhy z1)ODy3-0h?_@Vut;fFjN!w+YSQ2Z=9XkPLSctdQREl|(YF2DpZ8{npQOYBx-wi_^p zGhhs7WYTCZ4b6$H-w`l$O4E5jg6SGq_?}0MMG(fuLTDHa4W1N+v^-sVkets$oL0uh za$gx(?ihgY#!K4#{7G6_Zp=+JYeSE1IFfy7EUYLFXjM*M6Gt)1CG5rD@xMRVaMvGfENT(*cJu3CNp zGth4U#%dyfv7TALSkJ70(L5%EgZVF_i-+n6z*ua6YJhakg=^kIz-TTUFnom04aGY) zU)1CX@4`{G(z-ZctjEv{PnoVWU?@h@??T}d{=;*qfUxJxlwa_ih4LCslt8mC87mI> z2r9HeivyPx{m;@>}Wn`V^Jw)Fq{HGVQ>?PO+jXa_2vP_dQLleg!vz|F!jSS{bHmmfwtnJf2|5P@zD&$hrbrpFwL%G!))K@Hawc17HKnmB zN|iePT@efsdO%AunqvzQwaIYg*a6dFIIbFEZ*O_(<}7I|;VYluP_&SJkG&R75p9lUg%;fOI991bey=)__8p+T16htxmA z59Wy4X=JKcuO9%=~u|6m>PHAI00%0Swp@1>F zk1eP)_mB6n+5})Mj}92Ft!6m0S`nUO`^p83$pQiprcVK*y&5PSvpTkb(YOpSwl@th zmcJJ;w)O}tn6C#ITR#y*P}v8J=F;#D)>jG`%hv&h1TR@F z4r6=Mu#1Yv1~5`g$0F~@Y&?XO@fw7N)*QrEP-X`KWd0gjduYxQMv&$3FcJ!Z&ew{A z7bpzwJzE2~0(AYr%`_i@&LbqM$@>6f_Cti{^c#TjZ-7QrW^k@8(}^ej4rl z=F_0mMtidG9NG;iE;v7e;fH{`@^Li*0v3>rcD$e34WHy<8#s!76rxIhW zoNj1gW-^E}4xUYfgxim=jp6v`0N4Pt7(hl*uufZR2sCHj&0bFAkC2^rQ` z3YeR3paF}Dpu8G1K>0K8_2o zgW*bm>+$%C}te*=oG$XS2@q7+U1&cMHt(eXhd!X6- zknUt_itB>e`a#aw+>i%x8H-CXA|maMLWTVNeA`8b#%S