Accelerating reaction rate predictions: a machine learning approach using a novel quantum chemical property for nucleophilic aromatic substitutions

Abstract Accurately predicting reaction rate coefficients is crucial for various chemical engineering tasks, such as kinetic modeling and drug synthesis planning. Traditional methods like group additivity and rate rules have limitations, prompting the exploration of machine learning methods to predict these coefficients. These machine learning models are often combined with quantum chemical calculations to improve their performance. While often accurate, these approaches slow down predictions due to the need for quantum chemical calculations, particularly for transition states. This study addresses the issue by introducing a quantum chemical property that does not require transition state calculations but still correlates well with the rate coefficient. We further enhance the prediction speed by training a machine learning model to predict this property. Finally, we propose two approaches using this machine learning model to predict the rate coefficients of nucleophilic aromatic substitution reactions in fractions of a second.