Insight prediction of receptor binding activity of a set of benzamide derivatives using hybrid QSAR models: GA-MLR and GA-SVR

Abstract

In this study, we developed the hybrid QSAR models (HQSAR) for a set of benzamide derivatives by combining genetic algorithms with multivariate regression and support vector machine learning techniques. The genetic algorithm has assisted the selecting process of 2D and 3D molecular descriptors to get a globally optimal HQSARGA-MLR model with k = 7. The hybrid support vector regression model (HQSARGA-SVR) received from the selected descriptors of the multivariable regression model (HQSARGA-MLR) has been operated to predict the pIC50 activity of validation and prediction groups with MARE% of 0.8492 % and 2.8411 %. The hybrid support vector technique has improved the efficiency of the predictability of the multivariate regression model. The predicted activities pIC50 of benzamide derivatives resulting from the HQSARGA-SVR model are reliable enough and in good agreement with experimental data.