Application of machine learning in the discovery of antimicrobial peptides: exploring their potential for ulcerative colitis therapy.

Background: Ulcerative colitis (UC) is a chronic, relapsing inflammatory bowel disease with complex aetiology and limited treatment options. Antimicrobial peptides (AMPs), as endogenous immune effectors, have recently emerged as promising therapeutic agents in UC. However, systematic identification and functional evaluation of AMPs remain underexplored. We aimed to discover novel AMPs with potential therapeutic efficacy in UC by leveraging machine learning-based prediction and validating their impact in an experimental colitis model. Methods: We established a machine learning-driven pipeline to predict candidate AMPs based on their structural and functional features. Top-ranked peptides were synthesised and subjected to in vitro antibacterial assays and proteolytic stability tests. Their therapeutic potential was evaluated using a dextran sulfate sodium (DSS)-induced colitis mouse model, assessing clinical indicators, histopathology, inflammatory markers and gut microbiota alterations. Metagenomic and metabolomic analyses provided insights into microbial community dynamics and metabolic pathways. To probe the role of gut microbes in AMP-mediated gut homeostasis, we conducted Results: Several AMPs identified by machine learning exhibited potent antimicrobial activity and resistance to proteolytic degradation. In vivo, AMP administration ameliorated DSS-induced colitis symptoms, including body weight loss, Disease Activity Index and histological damage. Treatment also modulated the gut microbiome, increasing the abundance of Conclusions: Our findings demonstrate that machine learning-guided discovery of AMPs is a viable approach to identify promising therapeutic agents for UC. By integrating multi-omics analyses, we reveal potential microbiota-mediated mechanisms underlying AMP efficacy. These insights provide a strong foundation for advancing AMP-based strategies in UC management.