Machine learning for predicting enantioselectivity in chiral phosphoric acid-catalyzed naphthyl-indole synthesis

Abstract

The design of enantioselective axially chiral compounds is of great importance in modern synthetic chemistry, biochemistry, and material science due to their potential applications in the pharmaceutical and chemical industries. Traditional approaches to predicting enantioselectivity involve repetitive trial-and-error routines driven by chemical intuition. However, the fast-paced advancements in machine learning offer an alternate way to predict selectivity by leveraging data from laboratory experiments and computational analyses. In our study, we explore various machine learning (ML) techniques to predict the enantioselectivity of reactions using metal-free chiral phosphoric acid (CPA) catalysts in the synthesis of the naphthyl-indole scaffolds. We developed regression-based ML models using molecular descriptors of the reactants, catalysts and key intermediate complexes involved. Despite the limited dataset size, the random forest regression model performed remarkably well, achieving an R² score of 0.88 and RMSE of 0.32 on the test set. This demonstrates the effectiveness of integrating computational and machine learning methodologies in predicting enantioselectivity, marking a significant step forward in the pursuit of efficient, selective, and sustainable asymmetric catalysis.

Graphical abstract