A Graph Convolutional Network for Elution Order Prediction in Chromatographic Enantioseparation

Chromatography using chiral columns is the most effective method for determining the optical purity of enantiomer pairs, although identifying the absolute configuration of each peak is difficult and remains a challenging issue. We demonstrate a solution to this challenge based on machine learning in this study. We used a dataset that consists of retention times of absolute configurations of enantiomer pairs on CROWNPAK CR-I(+) and CR(+) columns to train a variety of neural network architectures. Our findings show that graph-based neural networks achieve superior prediction accuracy despite the relatively small dataset. Specifically, the model shows enantiomer elution order prediction accuracy of 88.9% for CR-I(+) and 88.4 % for CR(+) columns. We then predicted elution order for molecule types not contained in the original dataset, in order to test this model’s ability to generalize to new molecules. The overall accuracy across both column types was 81.0 %, indicating that our prediction model can be used for virtual experiments on enantioseparation before conducting actual experiments and analysis of column fractions.