Simple and accurate approaches to predict the activity of benzothiadiazine derivatives as HCV inhibitors

Abstract

Quantitative structure–activity relationship (QSAR) models of Hepatitis C virus (HCV) inhibitors were developed for 118 benzothiadiazine derivatives based on linear and nonlinear approaches. The best multi-linear regression (BMLR) was utilized to select appropriate molecular descriptors and develop linear model. Using the same descriptors, radial basis function neural networks (RBFNN) and support vector machine (SVM) were employed to generate the non-linear models. When comparing with the simpler BMLR model, RBFNN and SVM models had more accurate results with the square of correlation coefficients (R 2) of 0.850 and 0.875 for the training set, and R 2 of 0.893 and 0.854 for the test set. It proved that RBFNN and SVM were useful tools to predict the biological activity of benzothiadiazine derivatives as HCV inhibitors. Meanwhile, those factors governing the biological activity of benzothiadiazine derivatives were discussed. The proposed methods will be of significance in the HCV inhibitors research, and can be expected to apply to other similar research fields.