36. APPLICATION OF MACHINE LEARNING IN SCREENING FOR TYPE 2 ANTI-DIABETES NATURAL COMPOUNDS

Objective: Building and developing a machine learning model to predict natural compounds in the treatment of type 2 diabetes mellitus (T2DM), evaluating the multi-target impact ability of screened compounds. Materials and Methods: The study uses ImRMR and XGBoost algorithms to build a machine learning model that predicts natural compounds with anti T2DM activity. Combined with Molecular Docking on AutoDock Vina, the study evaluates the multi-target binding between screened compounds and potential impact targets of T2DM. Results: The DiMeNP model with an accuracy of 0.917 and an AUC of 0.962 was applied to screen for more than 13,000 natural compounds and predicted 11 flavonoid compounds and 3 lignan compounds. The compounds met the criteria of bioavailability and drug similarity, with a prediction probability of > 0.9 and all of them were well aligned on 5 potential targets in the treatment of diabetes including GLP1, DPP4, PPAR-γ, α-glucosidase, and α-amylase. Conclusion: Applying machine learning methods combined with molecular binding, the research and construction of the DiMeNP model is capable of screening compounds with multi-target impact activity on T2DM.