Modelling the quality of enantiomeric separations based on molecular descriptors

In this paper the selectivity and the resolution of enantiomeric separations are modelled. For each of the 50 molecules of the considered dataset, several molecular descriptors were calculated. The aim of this work is to determine whether it is possible to model the quality of the separations, based on the calculated descriptors. The chemometric methods used to explore and model the data were Principal Component Analysis, Projection Pursuit, Uninformative Variable Elimination by Partial Least Squares, Stepwise Multiple Linear Regression and Classification and Regression Trees. Stepwise Multiple Linear Regression gives the best models, both for selectivity and resolution, being the models able to predict the selectivity with an error lower than 4%, and the resolution with an error of 7%. The results seem to demonstrate that it is possible to predict quantitatively the quality of enantiomeric separations of related compounds on a given chromatographic system. (c) 2006 Elsevier B.V, All rights reserved.