Exploring Peptide Sequence Space Using Artificial Intelligence for Antimicrobial Peptides

Abstract

The number of effective antibiotics is rapidly decreasing due to the emergence of resistant bacteria, which necessitates the development of not only new, but novel strategies for developing antibiotics. Antimicrobial peptides (AMPs) are considered to be a promising new class of antibiotics, and effective strategies to predict potential candidates are still actively sought after. Here, we present a closed-loop artificial intelligence-based approach that combines an in silico genetic algorithm, machine learning prediction, and in vitro evaluation to improve the antimicrobial activity of peptides. Starting with a 13-mer natural AMP, 44 highly potent peptides were identified achieving up to a ca. 160-fold increase in antimicrobial activity within just three rounds of experiments. During these experiments, the conformation of the peptides selected was observed to alter from a random coil to α-helical form through the optimization process, and this is thought to significantly contribute to the improvement of antimicrobial activity. This strategy therefore shows it is possible to accelerate the discovery of antimicrobial peptides within a relatively small number of experiments, and to explore broad sequence and structural space.