PREDICTION OF CHIRAL CHROMATOGRAPHIC SEPARATIONS USING COMBINED MULTIVARIATE REGRESSION AND NEURAL NETWORKS

A new method for the prediction and description of enantioselective se parations on HPLC chiral stationary phases (CSPs) is described. Based on the combination of multivariate regression and neural networks, the method was successfully applied to the separation of a series of 29 a romatic acids and amides, chromatographed on three amylosic CSPs. Comb inations of charge transfer, electrostatic, lipophilic, and dipole int eractions, identified by multivariate regression, were found to descri be retention and enantioselectivity, with highly predictive models bei ng generated by the training of back-propagation neural networks.