Exploring Impact of Protein Sequence Local Information to Predict Enzyme-Ligand Binding Residues Using Machine Learning

Abstract

Enzymes are important for various biochemical reactions in living cells. In drug discovery and drug design, identifying small molecule binding residues in enzymes is a crucial step. Although, identifying ligand-binding residues using computational techniques are improving, accurate prediction remains difficult. Therefore, to address this problem, we used the sequence local information, i.e., sequence neighbors around the target residue, and transformed it into two ways. First, into a Chaos Game Representation (CGR) to form a feature vector and train with an Extreme Gradient Boosting classifier (XGB) and second into a non-numeric feature space and apply the conditional probabilistic based approach. Our results suggest that local protein sequence information along with global information can help to develop precise predictors for small molecule binding residues. We hope our observations could facilitate the researchers to investigate more into enzyme-ligand interaction and drug discovery.