Deep learning-driven discovery of novel anti-Salmonella AMPs from the chicken gut microbiome

Salmonella is one of the leading causes of foodborne diseases, with poultry meat and eggs serving as major transmission routes, posing a serious threat to food safety. Antimicrobial peptides (AMPs) naturally occur in immune systems and exhibit antimicrobial activity along with improved biocompatibility. Thus, they represent promising alternatives to antibiotics. In this study, we developed a deep learning model for identifying anti- Salmonella AMPs based on features extracted from PLMs. Using this model, we identified novel AMPs from metaproteomic data. This data was translated from a gene catalog obtained via metagenomic sequencing of fecal samples from healthy chickens. Alignment revealed no homologous proteins or related gene annotations, suggesting these peptides are previously uncharacterized. Additionally, we attempted sequence optimization and successfully shortened the peptides while reducing their minimum inhibitory concentration. Our findings demonstrate the feasibility of discovering and improving AMPs using AI-driven approaches. These potentially naturally derived AMPs hold promise as feed additives or environmental sprays for controlling Salmonella transmission in poultry and livestock. • We developed a deep learning model based on protein language models to identify anti- Salmonella AMPs. • Novel AMPs were discovered from metaproteomic data derived from the metagenomic sequencing of healthy chicken fecal samples. • The identified AMPs showed broad-spectrum antibacterial and bactericidal activities against multiple Salmonella subspecies. • Sequence optimization reduced AMP length while maintaining or improving antimicrobial potency.