Receptor and ligand-based 3D-QSAR study on a series of pyrazines/pyrrolidylquinazolines as inhibitors of PDE10A enzyme

The 3D-QSAR analysis was performed on the set of 175 potent inhibitors of the PDE10A enzyme. Four separate models were built based on different conformations and superimposition methods. They were generated following next criteria: (I) database alignment based on the conformations most similar to the co-crystalized ligand conformation, (II) database alignment based on the minimum energy conformations, (III) docking alignment based on the docked conformations, and (IV) database alignment based on the docking conformations. The best CoMFA and CoMSIA models, derived from superimposition III, show leave-one-out cross-validated correlation coefficient (q2) values of 0.673 and 0.707 as well as the non-cross-validated correlation coefficient (r2) values of 0.936 and 0.924, respectively. In addition, the satisfactory results, based on the bootstrapping analysis and 10- and 50-fold cross-validation, further indicate the highly statistical significance of the models. The external predictive abilities of these models were evaluated using a prediction set of 35 compounds, producing the predicted correlation coefficients. Results were graphically interpreted in terms of field contribution maps. A DISCOtech pharmacophore model was also constructed to light important structural features that could be responsible for the low- or high-inhibition activity.