Predicting enantioseparations on polysaccharide-based stationary phases using achiral and chiral molecular descriptors

• Forty-eight chiral molecules were analyzed on six polysaccharide chiral phases. • Models were built to predict the retention, separation and elution sequence. • Chiral descriptors were developed in-house to distinguish between enantiomers. • Predicted chromatograms were generated to visualize the expected separation. Achiral and chiral molecular descriptors were used to build models that predict the global retention of chiral molecules on the one hand, and their enantioselectivity, separation and elution sequence on the other. Chromatographic separations of 48 structurally diverse molecules were performed on six polysaccharide-based chiral stationary phases (CSPs) with two mobile phases (i.e. twelve chromatographic systems). Subsequently, “achiral” and “chiral” models were constructed. For the achiral models, either the mean retention time of the enantiomers ( t R ‾ ) or the logarithm of the mean retention factor (log k ¯ ) was modelled as a function of achiral descriptors, whereas for chiral models, the enantioselectivity was modelled using “Ensemble Steric and Electrostatic Chirality” (ESEC) descriptors. Initially, various types of ESEC descriptors were used to create models. A variable reduction technique, known as the Variable - Wootton, Sergent, Phan-Tan-Luu’s (V-WSP) algorithm, was applied to discard highly correlated descriptors. Stepwise multiple linear regression was applied to build the models. The achiral models for log k ¯ predicted the global retention rather well. For the enantioselectivity, separation and elution sequence, chiral models with good performance parameters and predictive abilities were obtained. Finally, the results from the achiral and chiral models were combined to predict the retention times of individual enantiomers. Predicted chromatograms were generated, to visualize both the expected and experimental separation. In most cases, the combined models yielded a moderate error (median relative error ranged from 5.8% to 22% across all systems) for the individual enantiomers.