Deep Learning-Driven Discovery of Novel Antimicrobial Peptides from Large-Scale Protist Genomes and Experimental...

The escalating issue of antibiotic resistance has created an urgent global demand within the biomedical field for the discovery of novel antimicrobial molecules as alternatives to traditional antibiotics. Previous studies have reported the identification of potential candidate antimicrobial peptides (AMPs) from extensive bacterial genomes by using deep learning. However, protists, as unique microorganisms capable of thriving in complex ecological environments, represent a largely untapped reservoir of unknown AMP resources that are awaiting systematic exploration. In this study, we harnessed deep learning techniques to identify novel antimicrobial peptides from large-scale protist genomes. Our results indicate that from 2120 protist genome data sets, through a multistage screening process involving sequence redundancy, the application of existing models (C_AMPs_Ptrdict, AMPEP, and AMPidentifier), and the integration of our optimized BERT and CNN models, we ultimately identified 3133 novel candidate AMPs from approximately 6.6 billion sequences. Experimental validation of 18 synthesized candidate peptides demonstrated inhibitory activity against at least one tested bacterial species. This high validation rate, where all 18 tested peptides exhibited inhibitory activity against at least one bacterial species, underscores the exceptional accuracy of the multimodel comprehensive identification approach for obtaining AMPs with antibacterial capability. Among the synthesized peptides, AMP_N2 and AMP_N3 exhibited strong antimicrobial activity in liquid culture while maintaining hemolysis rates below 3%. Our findings indicate that protists are a significant source of novel antimicrobial peptides and that deep learning techniques can be efficiently employed for AMP discovery. To our knowledge, this study represents the first large-scale exploration of novel antimicrobial peptides from over 2000 protist genomes.