Abstract
Dipeptidyl peptidase-4 (DPP-4) inhibitors play a critical role in the management of type 2 diabetes; however, some synthetic drugs may cause adverse effects. Natural peptides derived from rice offer a promising alternative due to their favorable biocompatibility and development potential. In this study, an AI-assisted virtual screening pipeline integrating machine learning, molecular docking, and molecular dynamics (MD) simulations was established to identify and evaluate rice-derived DPP-4 inhibitory peptides. A random forest classification model achieved 85.37% accuracy in predicting inhibitory activity. Peptides generated by simulated enzymatic hydrolysis were screened based on machine learning and docking scores, and four proline-rich peptides (PPPPPPPPA, PPPSPPPV, PPPPPY, and CPPPPAAY) were selected for MD analysis. The simulation results showed that PPPSPPPV formed a stable complex with the DPP-4 catalytic triad (Ser592-Asp670-His702) through electrostatic and hydrophobic interactions, with low structural fluctuation (RMSF < 1.75 Å). In vitro assays revealed that PPPPPY exhibited the strongest DPP-4 inhibitory activity (IC(50) = 153.2 ± 5.7 μM), followed by PPPPPPPPA (177.0 ± 6.0 μM) and PPPSPPPV (216.3 ± 4.5 μM). This study presents an efficient approach combining virtual screening and experimental validation, offering a structural and mechanistic foundation for the development of natural DPP-4 inhibitory peptides as candidates for functional foods or adjunct diabetes therapies.