Predicting Molecule-Protein Interactions using AI

This repository is meant to demonstrate the use of classical machine-learning and deep learning techniques for predicting molecule-protein binding classifications on the LEASH-BELKA Kaggle Dataset. It covers:

1. Classical machine-learning techniques in combination with FCHL featurization
   • Understanding the use of FCHL based descriptors
   • Using FCHLs for building the kernel matrix of all building block molecules
   • FCHL based feature matrix construction
   • FCHL feature matrix combined with logistic regression, random forest and gradient boosting classifiers
   • FCHL feature matrix combined with XGBoost and CatBoost (TBD)
   • Extending the feature matrix beyond molecular building blocks (including core information, Dy point of attachment information etc.) (TBD)
2. FCHLs combined with fully connected feed forward neural networks (TBD)
3. Transformer-based architectures (TBD)
4. Graph neural networks (TBD)