Fusarium head blight caused by Fusarium graminearum threatens global wheat production, causing substantial yield reduction and mycotoxin accumulation. This study harnessed machine learning to accelerate the discovery of antifungal peptides targeting this phytopathogen. By developing a de novo antimicrobial peptide database and extracting six critical physicochemical features, we established four predictive models with XGBoost demonstrating superior performance (R2 = 0.77, RMSE = 1.8). The machine-identified peptide TP achieved near-complete suppression of F. graminearum at 13.33 μM concentration. Molecular dynamics simulations elucidated its action mechanism, involving electrostatic interaction followed by hydrophobic insertion and binding to myosin disrupting cellular functions. This work highlights the paradigm shift of machine learning framework in agricultural antimicrobial development through data-driven biotechnology.
Keywords: computational drug design; machine learning; membrane penetration mechanism; molecular dynamics; myosin.