Prediction of IC50 Values of 2−benzyloxybenzamide Derivatives using Multiple Linear Regression and Artificial Neural Network Methods

Abstract

In this study, six molecular descriptors were selected from a pool of variables using stepwise regression to built a QSAR model for a series of 2-benzyloxy benzamide derivatives as an SMS2 inhibitor to reduce atherosclerosis. Simple multiple linear regression (MLR) and a nonlinear method, artificial neural network (ANN), were used to modeling the bioactivities of the compounds. Modeling was carried out in total with 34 compounds of 2-benzyl oxybenzamide derivatives. PCA was used to divide the compounds into two groups of two training series and tests. The model was constructed with 27 combinations as training set, then the validity and predictive ability of the model were evaluated with the remaining 7 combinations. While the MLR provides an acceptable model for predictions, the ANN-based model significantly improves the predictive ability. In ANN model the average relative error (RE%) of prediction set is lower than 1% and square correlation coefficient (R2) is 0.9912.