peptide-array-ml

Neural network modeling of peptide sequences and binding data

Installation

1. Download and install the Anaconda distribution of Python 3 (https://www.anaconda.com/download/)
2. Install PyTorch with one of the following commands in terminal/command prompt (https://pytorch.org):
   • Mac: conda install pytorch torchvision -c pytorch
   • Linux/Windows: conda install pytorch torchvision cpuonly -c pytorch
3. Download this GitHub repository

Running on Command Line

 Call __init__.py with Python and provide a csv-formatted arguments file as in the following example:

python peptide_array_ml/__init__.py peptide_array_ml/arguments.csv

Running with Scripts

 See publications directory for examples

Publications

1. Chowdhury, R., Taguchi, A. T., Kelbauskas, L., Stafford, P., Diehnelt, C. W., Zhao, Z-G., C. Williamson, P. C., Green, V., Woodbury, N. W. (2022) Submitted to PLoS Computational Biology

   Dataset: https://doi.org/10.5281/zenodo.7377844

   Matlab scripts (publications/2022_PLoS_Computational_Biology):

   • example_code_chowdhury_et_al.m: Run examples of neural network training and prediction from publication.

2. Taguchi, A. T., Boyd, J., Diehnelt, C. W., Legutki, J. B., Zhao, Z-G., and Woodbury, N. W. (2020) Combinatorial Science, 22 (10), 500-508 (DOI: 10.1021/acscombsci.0c00003)

   Scripts to reproduce figures (publications/2020_Combinatorial_Science):

   • fit_targets.py: Model all nine protein targets with a neural network. The parameter settings (Parameters.txt), trained models (Model.pth), predicted vs real correlation plots (Correlation.png), and amino acid similarity matrix plots (Similarity.png) are all saved along with other metadata in fits/.
   • multifit_target.py: Fit a neural network multiple times to the same target. This is useful for estimating the variability in model performance and predictions.
   • extrapolate_binding.py: Train a neural network on weaker binding values, and then use it to predict stronger binding values not previously seen by the model.
   • plot_specificity.py: Plot measured vs predicted binding specificity between targets.
   • plot_specificity_control.py: Plot measured vs predicted binding specificity for replicates of the same target.
   • plot_self_correlation.py: Plot correlation between replicates of the same target.
   • plot_loss.py: Plot loss during model training.
   • plot_distributions.py: Plot length and amino acid distributions of the peptide arrays.

License

The use of these algorithms is protected under intellectual property filed by Arizona State University