Synergistic interactions among QSAR descriptors

Quantitative structure–activity relationships (QSARs) and quantitative structure–property relationships (QSPRs) rely on regression equations containing numerical descriptors of molecular structure. In constructing these models, highly correlated descriptors are sometimes excluded from the regression equations. Although this exclusion seems reasonable, in fact it can lead investigators to overlook significant descriptor combinations, because the small differences between highly correlated descriptors sometimes encode important structural information. Furthermore, the multicollinearity that results from employing correlated descriptors is not as serious a problem as is often assumed. Described are several examples of cases in which pairs of highly correlated, poorly performing single‐parameter descriptors yield highly significant structure–property regression equations. In effect, the descriptors act synergistically and yield regression equations that model the systems examined better than the sum of the individual components. A discussion of practical approaches to this problem is given. © 2004 Wiley Periodicals, Inc. Int J Quantum Chem, 2004