Tox21_FeatureSelection_SMOTEENN_RF

This repository contains code accompanying the paper: Identifying Protein Features and Pathways Responsible for Toxicity using Machine learning, CANDO, and Tox21 datasets: Implications for Predictive Toxicology DOI:10.1101/2021.12.13.472455

In the directory there are three folders containing the data curation files, feature generation files, and the model training files.

• In the Tox21_DataPreproccessing folder Tox21_datacuration.ipynb: This is used for data curation using the 12 Tox21 assays (labels of activity).
• In the Feautregeneration_using CANDO folder Feature_generation_trainset.ipynb and Feature_generation_testset.ipynb: These are used for feature generation for train and test datasets using the SMILES from Tox21 data.
• In the ModelTraining folder Datapreprocessing_modeltraining.ipynb: This is used to preproccess the features and labels and perfrom model training. The train_utils.py and utilities.py files are helper functions for model training.

If you find this code, models or results useful, please cite the following paper:

@article{moukheiber2022identifying, title={Identifying Protein Features and Pathways Responsible for Toxicity Using Machine Learning and Tox21: Implications for Predictive Toxicology}, author={Moukheiber, Lama and Mangione, William and Moukheiber, Mira and Maleki, Saeed and Falls, Zackary and Gao, Mingchen and Samudrala, Ram}, journal={Molecules}, volume={27}, number={9}, pages={3021}, year={2022}, publisher={MDPI} }