Evaluation of alignment dependence in 3D-QSAR model construction using 4D-QSAR analysis

Four different training sets of ligands, ranging from relatively rigid anal ogs to large flexible molecules, were studied using 4D-QSAR analysis with t he goal of building optimized 3D-QSAR models as a function of alignment. On ly three ordered-atom alignments were employed in each of the four applicat ion studies. The sensitivity of the quality of the best 3D-QSAR model, as m easured by the cross-validated correlation coefficient, is increasingly sen sitive to increasingly flexible molecules. However, even the training sets of the relatively rigid molecules exhibit a significant dependence on the c hoice of alignment in generating 3D-QSAR models. 4D-QSAR analysis permits t he alignment variable to be separated from the conformation and pharmacopho re variables in building 3D-QSAR models. Thus, the findings reported for al ignment dependence are not biased, or clouded, by the conformation and/or p harmacophore degrees of freedom.