This repository contains datasets, model code and notebooks used for all experiments in the PROSTATA: Protein Stability Assessment using Transformers paper.

Directories

DATA contains the datasets used for PROSTATA training and testing in the format used by the datasets authors. Also the dataset introduced in this article is available here.
DATASETS contains the same datasets converted to a format used for model training.
PDB contains the PDB files downloaded during conversion.
ACDC_FOLDS - converted acdc-nn train folds from here and PROSTATA test results on Ssym and Ssym_r folds from here.

Notebooks

00.generate_datasets.ipynb - Process the DATA directory and generate the DATASETS directory.
01.add_megadataset_and_split_on_train_test_sets.ipynb Expand dataset with megadatasets data. Split on train and test sets. Generate PROSTATA_EXPERIMENTS directory.
02.test_models_by_folds.ipynb - Test each individual model in the ensemble using 5-fold cross validation
03.test_models_on_other_datasets_ensemble.ipynb - test the PROSTATA ensemble on various combinations of train and test datasets. test_*_with_predictions.csv containts tests set with prediction (pred_ddg) column.
04.train_final_ensemble.ipynb - train the ensemble on all data for the online tool.
PROSTATA_tool.ipynb - Colab notebook for PROSTATA. Predict DDG Values for single mutation on a user sequence.

environment.yml - conda environment.
PROSTATA_experiments_pearson.log - Logs of experiments run by 03.test_models_on_other_datasets_ensemble.ipynb notebook.
LICENSE - Apache License 2.0
Readme.MD - This file