Predicting antibacterial activity, efficacy, and hemotoxicity of peptides using an explainable machine learning framework

Abstract

Antimicrobial peptides have emerged as a potential alternative to combat the growing threat towards antimicrobial resistance. Owing to a large number of possible combinations of twenty naturally occurring amino acids, it is extremely resource intensive to experimentally identify whether a given peptide has desired therapeutic properties. To expedite the screening of therapeutic peptides, we propose a classification framework that can simultaneously predict the antibacterial activity, hemotoxicity, and efficacy against three most common pathogens i.e., Staphylococcus aureus, Escherichia coli, and Pseudomonas aeruginosa for any given peptide. The proposed framework uses support vector machine algorithm with amino acid compositions, sequence analysis, and physicochemical properties as features to develop three binary classifiers. Our models resulted in accuracies of 97.3 %, 86.2 %, and 84.1 % for antibacterial activity, combined efficacy against all three pathogens, and hemotoxicity, respectively. Explainable machine learning algorithm was implemented for each model to elucidate meaningful insights. It was evident that physicochemical properties along with the occurrence of certain amino acids play the most important role in determining antibacterial activity, efficacy, and hemolytic activity of peptides. The entire framework is made accessible freely in form of a web tool, which will further aid in rapid screening of antibacterial peptides with high therapeutic potential.