Prediction of Enantiomeric Excess in a Catalytic Process: A Chemoinformatics Approach Using Chirality Codes
Qingyou Zhang, Dandan Zhang, Jing-Ya Li, Hailin Long, Lu Xu · 2012
ABSTRACT: The enantiomeric excesses obtained with 296 chiral catalysts in the asymmetric hydrogen transfer to acetophenone are extracted from a combinatorial database constructed by Riant et al., and used to investigate their relationships with the molecular structures of the catalysts. These catalysts incorporate three components, namely a β-amino alcohol, an aldehyde or ketone, and a metal complex precursor. The structures of the catalysts are featured by molecular descriptors, including chirality codes, calculated for their three components. Selection of variables, from the initial pool of molecular descriptors, is performed by a genetic algorithm. These are fed to a random forest to predict the enantiomeric excess. A square correlation coefficient of R2=0.82 and RMSE=9.96 are obtained for the test set, and the results for cross-validation of the whole dataset (in the out-of-bag procedure) are R2=0.79 and RMSE=10.96. The method can be helpful for computer-aided design of enantioselective catalysts. Usually, a pair of enantiomers exhibit different physical, chemical and biological activity. In order to reduce known or unknown side effects of the non-functional enantiomer, the needs