3D-QSAR illusions

Abstract

3D-QSAR is typically used to construct models (1) to predict activities, (2) to illustrate significant regions, and (3) to provide insight into possible interactions. To the contrary, examples are described herein which make it clear that the predictivity of such models remains elusive, that so-called significant regions are subject to the vagaries of alignment, and that the nature of possible interactions heavily depends on the eye of the beholder. Although great strides have been made in the imaginative use of 3D-descriptors, 3D-QSAR remains largely a retrospective analytical tool. The arbitrary nature of both the alignment paradigm and atom description lends itself to capricious models, which in turn can lead to distorted conclusions. Despite these illusionary pitfalls, predictions can be enhanced when the test set is bounded by the descriptor space represented in the training set. Interpretation of significant interaction regions becomes more meaningful when alignment is constrained by a binding site. Correlations obtained with a variety of atom descriptors suggest choosing useful ones, in particular, in guiding synthetic effort.