Deep Learning Methods for Binding Site Prediction in Protein Structures

Abstract

This work is an overview of deep machine learning methods aimed at predicting binding sites in protein structures. Several classes of methods are selected: prediction of binding sites for small molecules, proteins, and nucleic acids. For each class, various approaches to prediction are considered (prediction of binding atoms, residues, surfaces, pockets). Specifics of feature selection and neural network architectures inherent to each class and approach are highlighted, and an attempt is made to explain these specifics and foresee the further direction of their development.