STRUCTURE ACTIVITY CORRELATIONS FOR PSYCHOTOMIMETICS .2. PHENYLALKYLAMINES - A TREATMENT OF NONLINEARITY USING THE ALTERNATING CONDITIONAL EXPECTATIONS TECHNIQUE

A refinement of the quantitative structure-activity relationship presented in Part 1 of this series using the alternating conditional expectations (ACE) technique to account for nonlinearity is described. It is shown that the hydrophobicity interaction effect found in Part 1 is nonlinear in nature, and is a major term in accounting for the activity of phenylalkylamine psychotomimetics. The volume interaction can be accounted for equally by nonlinear behaviour in the remaining terms. Para-group hydrophobicity dependence is nonlinear, confirming the results of Part 1. A receptor site model is postulated. The plausibility of the model is confirmed by a satisfactory prediction of some data which were not used in its construction. The ACE method is tested on synthetic data and a shortcoming of the cross-validation technique is demonstrated. The stepwise use of ACE in selection of variables is shown, at least with this data set, to be unsatisfactory. An alternative method of validating stepwise regression results is presented.