Support Vector Machine‐Based Prediction of Enantioselectivity in Fluorination of Allylic Alcohols

Abstract

AbstractQuantitative structure‐selectivity relationship (QSSR) models in asymmetric catalytic reactions can accelerate catalyst design by predicting enantioselectivity ΔΔG (=−RTln(er), where er is the enantiomeric ratio). Support vector machine (SVM) as the most popular machine learning algorithm, can fit nonlinear relationships between structural feature and enantioselectivities. Three Dragon‐based descriptors were used to develop SVM Model A for enantioselectivities in fluorination of allylic alcohols by using boronic acids and chiral phosphoric acids as catalysts. The training and test sets of SVM Model A, respectively, have determination coefficients R2 of 0.951 and 0.945, which are higher than that (0.870 and 0.884, respectively) from SVM Model B based on four descriptors (non‐covalent interaction, Sterimol parameters, and infrared intensity). Both of the SVM models are proved to be accurate in predicting enantioinduction of fluorination of allylic alcohols with boronic acids and chiral phosphoric acids as catalysts.