4D-QSAR studies of CB 2 cannabinoid receptor inverse agonists: a comparison to 3D-QSAR

Over the years QSAR methods have developed from 2D-QSAR to more complex 4D-QSAR which features freedom of alignment and conformational flexibility of individual ligands. This approach takes advantage of conformational ensemble profile (CEP) generated for individual compounds by molecular dynamics simulations. In present study, the 4D-QSAR methods called LQTAgrid-QSAR has been performed on a series of potent CB 2 cannabinoid receptor inverse agonists. Step-wise method was used to select the most informative variables. Partial least squares (PLS) and multiple linear regression (MLR) methods were used for constructing the regression models. Y-randomization and leave-N-out cross-validation (LNO) were carried out to verify the robustness of the model and to analysis of the independent test set. Best 4D-QSAR model provided the following statistics: R 2 = 0.862, q 2 LOO = 0.737, q 2 LNO = 0.719, R 2 Pred = 0.884 (PLS) and R 2 = 0.863, q 2 LOO = 0.771, q 2 LNO = 0.761, R 2 Pred = 0.877 (MLR). The comparison of the 4D-QSAR to 3D-QSAR was performed.