Abstract

COX-2 inhibitory activities of some thiazinan, benzdiazinan and benzthiazinan derivatives were modeled by quantitative structure–activity relationship (QSAR) with stepwise-multiple linear regression (SW-MLR) method. The built model was robust and predictive with correlation coefficient (R²) of 0.840 for training and 0.522 for test groups. The quality of the model was evaluated by leave-one out (LOO) cross validation and LOO correlation coefficient (Q²) was 0.639. We also investigated a leverage approach for defining of applicability domain of model. According to QSAR model results, COX-2 inhibitory activity of selected data set had correlation with VE3_Dzm (Logarithmic coefficient sum of the last eigenvector from Barysz matrix weighted by mass), GATS6c (geometrical structure of the considered molecules in a complex way) and GATS5i (Geary autocorrelation of lag 5 weighted by ionization potential) descriptors which derived from their structures.