A simple and robust model to predict the inhibitory activity of α-glucosidase inhibitors through combined QSAR modeling and molecular docking techniques

Abstract

Quantitative structure-activity relationships (QSAR) and molecular docking studies have been performed on a series of 35 α-glucosidase inhibitory derivatives. The QSAR models have been developed by genetic algorithm-multiple linear regression (GA-MLR) and least squares-support vector machine (LS-SVM) methods to correlate the conformational descriptors to the inhibitory activity. The obtained models with 5 descriptors were validated and illustrated to be statistically significant. They had desirable prediction based on squared correlation coefficient (R²), cross-validated correlation coefficient (Q²), root-mean-squares error (RMSE) and Fisher (F) parameters (R² = 0.951, Q² = 0.931, RMSE = 0.121, and F = 114.629 for GA-MLR model, and R² = 0.989, Q² = 0.987, RMSE = 0.056 and F = 543.754 for LS-SVM model). The crucial descriptor named DELS was explored to have the highest correlation with the inhibitory activity and thus has been chosen to build a simple model. The QSAR model developed with this mono-descriptor showed appropriate results of the predicted model using LS-SVM method (R² = 0.888, Q² = 0.872, RMSE = 0.185 and F = 221.459). Also, molecular docking which focuses on the interaction between ligands and α-glucosidase in the protein active site considered different binding positions to find the best binding mode. It helped the QSAR study to propose more comprehensive details of the compounds structures and was used to design more active compounds. The most active designed compound had a high inhibitory activity of 9.22 that can be proposed for the treatment of diabetes type 2.

Keywords: Diabetes; Docking; Inhibitory activity; Medicinal chemistry; QSAR; α-Glucosidase.