This repository contains data, training and testing scripts, models, and figure generation scripts for the preprint Spectral neural approximations for models of transcriptional dynamics by Gennady Gorin, Maria T. Carilli, Tara Chari and Lior Pachter.
Contains data used to train, validate, and test models for kernel weight regression that contain conditional examples and conditional distributions for 256 unique rate vectors {${b,\beta,\gamma}$} (corresponding to thousands ofs conditional examples and conditional distributions per file). It also contains testing files with 256 unique rate vectors {${b,\beta,\gamma}$}, the same as those in the correspondingly numbered conditional testing files, and their full joint distributions:
- 6 train files:
256_train_*.npy
(conditional) - 2 validation files:
256_valid_*.npy
(conditional) - 3 testing files:
256_test_*.npy
(conditional) - 3 full joint distribution testing files:
256_test_full_*.npy
(joint)
Contains final architecture for kernel weight regression (best_model_MODEL
), as well as models used to test varying the number of kernel functions in kernel_models
, nodes in hidden layer in node_models
, and number of training parameters in param_models
. Contains models used for direct regression in direct_models
.
Figures used in preprint (generated by notebook Figures.ipynb
).
Also contains scripts used to generate results stored in results/
for figures: figure*_script.py
.
Results from figure scripts in figs/
. Used to generate figures in Figures.ipynb
notebook.
Figures.ipynb
: Generates preprint figures.Training_Environment.ipynb
: Environment to train model to predict conditional probabilities using kernel weight regression. Allows user to change various model and training configurations and visualize the results on testing data.
train_conditional.py
: Contains all functions necessary to train conditional model (kernel weight regression).tools_conditional.py
: Contains tools to visualize training and testing results and assess performance for kernel weight regression.ypred_module.py
: Contains predefined function for kernel generating and functions for calculating probabilities for kernel weight regression.direct_module.py
: Contains model definition, training function, and functions for predicting probabilities for direct regression.exact_cme.py
: Contains functions for calculating exact solution over grid of user specified size and generating parameters within user specified bounds.