Linear and nonlinear QSAR modeling of 1,3,8-substituted-9-deazaxanthines as potential selective A2BAR antagonists

Abstract

A QSAR study of 74 derivatives of 1,3,8-substituted-9-deazaxanthines as potent and selective A2B adenosine receptor (A2BAR) antagonists is described. pKi of all the studied compounds were acquired by three linear and nonlinear methods namely stepwise multiple linear regression, partial least squares (PLS), and general regression neural networks (GRNN). The performances of developed models were tested by several external and internal validation methods and also the criteria recommended by Tropsha and Roy. Predictability and possible overfitting in the resulting models were examined by cross-validation. Results revealed the significant role of topological and geometrical descriptors in binding of the studied compounds to A2BAR. PLS and GRNN models had good statistical qualities with PLS showing better performance (R 2 = 0.863 and Q 2 = 0.817). Applicability domain of the models was also defined. The prediction results were in good agreement with the experimental data.