A Quantitative Structure-Activity Relationship Model

Abstract

Structure-activity relationships were studied for a series of 46 2.6-dimethyl-3.5-dicabomethoxy-4-phenyl-1.4-dihydropyridine derivatives by means of multiple linear regression (MLR) and artificial neural network (ANN) techniques. The values of log (1/EC50), which represents the 50% effective concentration for blocking the Ca2+ channel of the studied compounds were correlated with the descriptors encoding the chemical structures. Using the pertinent descriptors revealed by the regression analysis, a correlation coefficient of 0.99 (s = 0.23) for the training set (n = 46) was obtained for the ANN using the Levenberg-Marquardt algorithm with a 3-10-1 configuration. The results obtained from this study indicate that the activity of 2.6-dimethyl-3.5-dicabomethoxy-4-phenyl-1.4-dihydropyridine derivatives is strongly dependent on molar refractivity (MR), electronic factors (especially on the connectivity indices (IC0)) and hydrogen-bond donor's (HBD) of the molecule. Comparison of the descriptor's contribution obtained with MLR and ANN models indicates the presence of non-linearity in the data and the interaction effect between them since the efficiency of these descriptors was increased by the ANN model.On the other hand, we have used a new, robust structure-activity mapping technique, a Bayesian-regularized neural network, to develop a quantitative structure-activity relationship (QSAR) model, and the ability of the model was tested by using the cross validation technique. The results show that the method is robust and reliable and gives good results. Comparisons of Bayesian neural net models with those derived by classical MLR model analysis showed its superiority in generalization.