Application of PC-ANN and PC-LS-SVM in QSAR of CCR1 antagonist compounds: a comparative study.

Principal component regression (PCR), principal component-artificial neural network (PC-ANN), and principal component-least squares-support vector machine (PC-LS-SVM) as regression methods were investigated for building quantitative structure-activity relationships for the prediction of inhibitory activity of some CCR1 antagonists. Nonlinear methods (PC-ANN and PC-LS-SVM) were better than the PCR method considerably in the goodness of fit and predictivity parameters and other criteria for evaluation of the proposed model. These results reflect a nonlinear relationship between the principal components obtained from molecular descriptors and the inhibitory activity of this set of molecules. The maximum variance in activity of the molecules, in PCR method was 45.5%, whereas nonlinear methods, PC-ANN and PC-LS-SVM, could explain more than 93.7% and 95.6% variance in activity data respectively.