Using Evolutionary Algorithms and Machine Learning to Explore Sequence Space for the Discovery of Antimicrobial Peptides

Summary We present a proof-of-concept methodology for efficiently optimizing a chemical trait by using an artificial evolutionary workflow. We demonstrate this by optimizing the efficacy of antimicrobial peptides (AMPs). In particular, we used a closed-loop approach that combines a genetic algorithm, machine learning, and in vitro evaluation to improve the antimicrobial activity of peptides against Escherichia coli . Starting with a 13-mer natural AMP, we identified 44 highly potent peptides, 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 changed from a random coil to an α-helical form. This strategy not only establishes the potential of in vitro molecule evolution using an algorithmic genetic system but also accelerates the discovery of antimicrobial peptides and other functional molecules within a relatively small number of experiments, allowing the exploration of broad sequence and structural space.