Deep-Learning Driven Identification of Novel Antimicrobial Peptides.

In this study, we report the identification of novel antimicrobial peptides (AMPs) via a machine learning-driven pipeline. Short Trp-rich peptide sequences were obtained using the HydrAMP deep learning (DL) algorithm, followed by the in silico screening for antimicrobial activity via the AMPlify DL model. Three candidates, namely AMP1, AMP2, and AMP3, were selected for synthesis and experimental validation. The antimicrobial activity was evaluated in vitro against a panel of Gram-positive and Gram-negative bacterial strains. Among them, AMP3 demonstrated the broader antibacterial spectrum. To investigate the mechanisms of action, we conducted detailed biophysical analyses of AMP3 interaction with liposomal models of bacterial membranes. The data revealed significant perturbation of membrane bilayer stability, supporting the proposed membrane-targeting activity of AMP3. Overall, our results underscore the potential of DL approaches for the accelerated discovery and mechanistic characterization of novel AMPs.