Aj. Hopfinger et al., Evaluation of alignment dependence in 3D-QSAR model construction using 4D-QSAR analysis, INTERNET J, 3(9-11), 2000, pp. NIL_23-NIL_42
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.