The discovery of wheat-derived anticaries peptides by a ML-based computational strategy.

This study proposes a machine learning-based peptide screening strategy for identifying wheat-derived peptides with anti-caries potential. By integrating multiple feature descriptors and algorithms (Random Forest, XGBoost), we constructed a Screening Funnel Model and identified a wheat-derived peptide AP-2 (FPVTWRWWKWW) as the prioritized candidate. Experimental validation demonstrated that AP-2 exhibits potent antibacterial activity against Streptococcus mutans (MIC = 4 μM), achieving rapid bactericidal effects through bacterial membrane disruption, and significantly inhibits biofilm formation at 1/2 × MIC concentration. AP-2 exhibited an extremely low hemolysis rate and demonstrated favorable stability in saliva within 1 h. In vivo, studies confirmed that AP-2 effectively prevents early caries formation in rats at low concentrations without inducing organ toxicity or oral microbiota dysbiosis. These results demonstrate the utility of machine learning in discovering wheat-derived anti-caries peptides and indicate that AP-2 represents a safe and effective candidate for clinical translation.