Stepwise Leave‐One‐Isomer‐Out Free‐Wilson Approaches as Preprocessing Tools in QSAR Analysis of Racemates

QSAR analysis of racemates is complicated if specific substituent‐receptor interactions and, by that, specific spatial fits to the binding site result in individual but unknown activity differences of enantiomers, and even in structure‐dependent changes of which is the more active configuration. In a first approximation, additivity of substituent contributions should be assumed instead of major conformational effects. Then, Free‐Wilson analysis (FWA) can be used as preprocessing tool to reduce a starting set of all pairs of enantiomers into a final series of the probably (more) active configurations. A stepwise “leave‐one‐isomer‐out” approach is applied, where the model is successively improved by checking all remaining pairs and leaving out one enantiomer, determined by a special criterion of poorest prediction, in each step. The final model is given by the maximal F value. This approach was applied to histamine H 1 antagonistic activity (pK B , guinea pig ileum) of 19 racemic and six non‐chiral phenyl‐halogenated N‐(diphenylpropyl)‐N′‐(imidazolylpropyl)guanidines. Based on only eight variables because of additivity of meta and para contributions, the starting model with n = 44, r 2 = 0.29, s = 0.52, F = 1.8, r 2 ‐PRESS = −0.14 has been improved to a final one with n = 31 (only six remaining pairs), r 2 = 0.84, s = 0.24, F=14.0, r 2 ‐PRESS = 0.65. Additionally, each of the successive series was submitted to CoMFA. Statistical parameters of the parallel CoMFA and FWA models are closely related. QSAR results obtained with both methods correspond to well‐known structure‐activity relationships of diphenhydramine‐like H 1 antagonists. A direct application of the leave‐one‐isomer‐out strategy to CoMFA was less successful.