Prediction of the anti-HIV activities of PETT analogs as non-nucleoside HIV-1 reverse transcriptase inhibitors by linear and non-linear QSAR models

Abstract

Quantitative structure-activity relationship (QSAR) models were constructed in order to predict the anti-HIV activity of a set of phenethyl thiazole thiourea (PETT) analogs by calculated descriptors. Molecular descriptors calculated by Dragon software were subjected to variable reduction using the stepwise regression. The variables were then used as inputs for QSAR model generation using multiple linear regression (MLR) and artificial neural network (ANN). Validation study of the MLR and ANN models was performed using the test set and leave-one-out techniques. The results obtained for prediction of the test set by the MLR and ANN models showed squared correlation coefficients of 0.766 and 0.913, respectively.

Keywords

• bayesian regularized algorithm
• anti-hiv
• thiourea
• inhibitors
• artificial neural network (ann)