Mapping membrane activity in undiscovered peptide sequence space using machine learning

Source abstract

Significance We use machine learning on membrane-permeating ⍺-helical host defense peptides to study the nature of their functional commonality and sequence homology. Machine learning is combined with calibrating experiments to show that the metric in our support vector machine model correlates not with antimicrobial activity but with a peptide’s ability to generate the negative Gaussian membrane curvature necessary for membrane permeation. Moreover, we use the classifier reflexively to map the undiscovered sequence space of antimicrobial peptides and identify taxonomies of peptides with similar topological membrane remodeling activity, including endogenous neuropeptides, viral fusion proteins, topogenic peptides, and amyloids.