Machine learning combined with molecular simulations to screen α-amylase inhibitors as compounds that regulate blood sugar

Abstract

Diabetes, a metabolic disease characterized by hyperglycemia, seriously endangers the health and the lives of people. α-Amylase inhibitors have become effective substances to control blood glucose, and attracted extensive attention. In this study, a database of α-amylase inhibitors derived from naturally active small molecules in food was created and a quantitative structure-activity relationship model was developed by combining three machine learning methods (SVM, RF, and LDA) with four descriptors (MOE, ChemoPy, Mordred, and Rdkit). Hydrogen bond and hydrophobic interaction in the inhibition of α-amylase activity was confirmed by molecular docking. Enzyme inhibition experiments showed that the predicted compound had α-amylase inhibitory activity. Nevadensin was identified as a promising candidate of α-amylase inhibitors. The stability of α-amylase binding reaction was verified by molecular dynamics simulation. Optimal process conditions for the extraction of nevadensin from L. pauciflorus maxim were derived from single-factor experiments and response surface modeling. A promising method for digging natural α-amylase inhibitors was developed and the mode between inhibitors and α-amylase was explained in this research.