QSAR modeling of anti-HIV activity for DAPY-like derivatives using the mixture of ligand-receptor binding information and functional group features as a new class of descriptors

An accurate QSAR model was developed for the prediction of the anti-HIV activities of a set of DAPY-like derivatives as new non-nucleoside reverse transcriptase inhibitors (NNRTIs). The ligand-receptor (LR) interactions for all compounds were studied by the docking of compounds in the active site of appropriate receptors. The binding information of LR complexes at the best pose was called the molecular docking descriptors (MDDs). The mixture of 10 MDDs with about 154 simple, functional group counts was used as a new group of descriptors in the QSAR study of DAPY-like compounds. Among the 164 mixed descriptors, seven descriptors were selected as the most effective descriptors using the linear stepwise regression (SR) method and used as inputs of the artificial neural network (ANN) model. Levenberg-Marquardt (LM) method was used for the training of ANN through the backpropagation of errors algorithm. The Levenberg-Marquardt artificial neural network (LM-ANN) model with the architecture of 6-4-1 was selected as the optimal model. The predictability of the LM-ANN model was estimated by applying the external test set and the leave-one-out (LOO) method. The mean square errors (MSEs) and coefficient of determination (R-2) values for the test set were 0.16 and 0.89, respectively. The Q(2) value for the LOO method was calculated as 0.72. The results obtained demonstrated that a mixture of MDDs and functional group counts provided the required information for the construction of a QSAR model with acceptable prediction ability.