PCSPred: Prediction of Short Chain Antimicrobial Peptides using Machine Learning Algorithms

Antimicrobial peptides (AMPs) have exhibited an effective and widespread impact in healthcare for evolving into a novel strategy against bacterial infections which have become immune to most of the conventional antibiotics. In-silico approach of evaluating antimicrobial activity for multiple combinations of peptide sequences proves to be optimizational in terms of time and effort when compared to wet lab analysis. In the current study, we have developed a predictive machine learning model which incorporates the physico-chemical and spatial properties of peptide sequences. We propose that the amino acid composition, α-helix and β-sheet propensities, charge-to-hydrophobicity ratio, isoelectric point, and their dipeptide composition are significant features that might be useful parameters for identifying novel AMPs. The PCSPred model utilizes the Random Forest algorithm of predicting, which is capable of providing the best accuracy and precision when compared to other techniques. Additionally, we have developed a means of predicting all possible peptide sequences having a maximum length of 10 amino acids with antimicrobial properties. Thus, we constructed an AMP library containing new sequences which can be implemented in-vivo to synthesize novel AMPs that are successful against resistant microbial infections and would overcome the drawbacks of conventional treatment.