QSAR investigations and structure-based virtual screening on a series of nitrobenzoxadiazole derivatives targeting human glutathione-S-transferases

Quantitative structure-activity relationship (QSAR) models are useful tools for understanding the relation between biological activity and chemical structure, and for the design of new drugs. In this work, the performances of two QSAR approaches for modelling and predicting glutathione-S-transferases (GSTP1-1) inhibition are compared, namely Multiple Linear Regression (MLR) and Artificial Neural Network (ANN). These models are applied to 38 thiol substituted nitrobenzoxadiazole inhibitors. For these compounds, we found that the nonlinear model performs better than the linear one in terms of predictive ability, indicating a non-linear relation between the selected molecular descriptors and GSTP1-1 inhibition. The validity of the proposed models was established using the following techniques: separation of data into independent training and test sets, leave-one-out cross-validation, and Y-randomization. Otherwise, the domain of applicability indicating the area of reliable predictions was defined. Through the adopted QSAR model and in silico virtual screening, we identified 23 hits with good predictability. Our work should motivate future in vitro investigations on these compounds.