An improved design of fluorophilic molecules: prediction of the ln P fluorous partition coefficient, fluorophilicity, using 3D QSAR descriptors and neural networks

A combination of 3D QSAR molecular descriptors and artificial neural networ ks have been used to predict fluorophilicities, the natural logarithm of th e perfluoro(methylcyclohexane)/toluene partition coefficients, for a wide r ange of partially fluorinated organic compounds. The average error of the p redictions was less than twice the 0.2 experimental error. Multiple Linear regression proved to be much less efficient. To better characterise the flu orous partition phenomenon, specific fluorophilicity was defined as the pro duct of fluorophilicity and of the ratio of the van der Waals volumes of th e expelled fluorous solvent and the entering solute molecules. This dimensi onless term correlates well in a compound family with the calculated Hildeb rand parameters of the fluorous molecules. The trifluoromethyl group was fo und highly effective for increasing the fluorous phase affinities of model compounds when used in combination with longer perfluoroalkyl groups. (C) 2 001 Elsevier science B.V. All rights reserved.