Detection of antimicrobial peptides from fecal samples of FMT donors using deep learning.

Introduction: Antimicrobial peptides (AMPs) represent a class of short peptides that are widely distributed in organisms and are regarded as an effective means to tackle bacterial resistance, potentially functioning as substitutes for onventional antibiotics. Methods: We employed metagenomics in combination with deep learning to mine AMPs from the 120 fecal microbiota transplantation (FMT) donor metagenome. Subsequently, a comprehensive analysis of the candidate AMPs was conducted through metaproteomic cross-validation, solubility analysis, cross-validation with other prediction tools, correlation analysis, and molecular dynamics simulations. Finally, four candidate AMPs were selected for chemical synthesis, and experimental validation identified two with broad-spectrum antimicrobial activity. Furthermore, molecular docking was utilized to further analyze the antimicrobial mechanisms of the candidate AMPs. Results: Our approach successfully predicted 2,820,488 potential AMPs. After a comprehensive analysis, four candidate AMPs were selected for synthesis, two of which exhibited broad-spectrum antimicrobial activity. Molecular docking provided further insight into the binding mechanisms of these peptides. Discussion: This study demonstrates the feasibility of discovering functional AMPs from the human fecal microbiome using computational and experimental approaches, highlights the potential of mining novel AMPs from the fecal microbiome, and provides new insights into the therapeutic mechanisms of FMT.