Interpretable QSAR Modelling and QSAR-Based Virtual Screening of Novel 3H-Thiazolo[4,5-b]pyridin-2-one Derivatives as Potential Antioxidant Drug Candidates

Abstract

Quantitative structure-activity relationship (QSAR) study has been carried out for 32 N3 substituted 3H-thiazolo[4,5-b]pyridin-2-one derivatives as potential antioxidant drug candidates. The genetic algorithm (GA) and multiple linear regression analysis (MLRA) were used as appropriate techniques for descriptors selection and correlation models generation. The four best regressions for the prediction of the ability to scavenge the DPPH radical were generated as three-parameter QSAR models with the highest statistical characteristics and predictive ability. Based on the validation parameters of the generated models, it may be stated that they all satisfy the statistical requirements for their goodness-of-fitting with no current overfitting. The predictive ability of the constructed models was assessed with both internal and external validation approach and estimated with the leave-one-out and leave-group-out cross-validation coefficients (Q2LOO and Q2LGO). The values of Q2LOO (0.7060  0.7480) and Q2LGO (0.6647  0.7711) are reasonable, showing that the models are significant and robust to predict the free radical scavenging activity of the compounds from both training and validation sets. Applicability domain defining technique was employed to the obtained models and it was indicated that most structures were adequately represented by the chemical space of the models.