Machine Learning‐Assisted Prediction and Generation of Antimicrobial Peptides
Sukhvir Kaur Bhangu, Nicholas Welch, Morgan Lewis, Fanyi Li, Brint Gardner, Helmut Thissen, Wioleta Kowalczyk
Abstract
Antimicrobial peptides (AMPs) offer a highly potent alternative solution due to their broad-spectrum activity and minimum resistance development against the rapidly evolving antibiotic-resistant pathogens. Herein, to accelerate the discovery process of new AMPs, a predictive and generative algorithm is build, which constructs new peptide sequences, scores their antimicrobial activity using a machine learning (ML) model, identifies amino acid motifs, and assembles high-ranking motifs into new peptide sequences. The eXtreme Gradient Boosting model achieves an accuracy of ≈87% in distinguishing between AMPs and non-AMPs. The generated peptide sequences are experimentally validated against the bacterial pathogens, and an accuracy of ≈60% is achieved. To refine the algorithm, the physicochemical features are analyzed, particularly charge and hydrophobicity of experimentally validated peptides. The peptides with specific range of charge and hydrophobicity are then removed, which lead to a substantial increase in an experimental accuracy, from ≈60% to ≈80%. Furthermore, generated peptides are active against different fungal strains with minimal off-target toxicity. In summary, in silico predictive and generative models for functional motif and AMP discovery are powerful tools for engineering highly effective AMPs to combat multidrug resistant pathogens.