Prediction of the enantiomeric excess value for asymmetric transfer hydrogenation based on machine learning

Abstract

Asymmetric transfer hydrogenation has a wide range of applications in organic synthesis. In this work, we predict the enantiomeric excess value of asymmetric transfer hydrogenation reactions by building a machine learning black-box model. Based on DFT calculations, we extracted some molecular descriptors (such as sterimol parameters, buried volume parameters, NBO charges, etc.) as features, which can be inputted into the machine learning model, and then calculated the enantiomeric excess value. We found that the random forest model performed the best on this dataset, with the test-set root-mean-square error being 8.6 and the coefficient of determination R² being 0.86 in the prediction of the enantiomeric excess value compared to the experimental value. The results demonstrate that our model can be used for the prediction of the enantiomeric excess value for asymmetric transfer hydrogenation.