Construction of QSAR model between the ligand and γ-Aminobutyric acid type A receptor using support vector regression algorithm

Abstract

Quantitative structure-activity relationship (QSAR) plays an important role in the prediction of biological activity based on machine learning. According to the characteristics of the binding interface between ligands and the γ-Aminobutyric acid type A (GABAA) receptor, we used random forest feature selection and support vector regression (SVR) to establish three QSAR models. The best QSAR model features include docking ligand molecular descriptors and ligand-receptor interactions. We also used Leave-One-Out-Cross-Validation (LOOCV) to select the appropriate value C = 2, g = 0.0221. The result of cross validation (QLOO2) is 0.8225, R2 of test set is 0.8326, and MSE is 0.0910. In addition, we found that BELm2, BELe2, Mor08v, Mor29m, refRMS and intermol _ energy are key features, which helps to build QSAR model more accurately.