Machine learning and DFT coupling: A powerful approach to explore organic amine catalysts for ring-opening polymerization reaction

Abstract

Recently, using well-known data to drive the chemical feature of catalysts for the specified reaction has emerged as a prevalent approach in catalysis science. Amines as essential compounds play crucial roles in living organisms, pharmaceutical, agricultural applications, and various chemical reactions. However, developing a systematic strategy for the screening of amine as an organic catalyst in ring-opening polymerization (ROP) remains a significant challenge. In predicting the catalytic performance of ROP reactions, the proton affinity (PA) of amines serves as a significant indicator, which reflects the basicity of the catalyst and the activity of the ROP reaction. Herein, we conducted a systematical analysis on a dataset of 259 unique amine compounds, involving the application of seven classical machine learning models and 23 molecular descriptors. The predictive model was constructed with molecular descriptors and DFT-calculated PA values. Among the regression results, the Support Vector Machine (SV) model exhibited the optimal performance, with R2, average mean absolute error (MAE) and root mean square error (RMSE) achieving 0.84, 8.38 kJ/mol and 13.77 kJ/mol, respectively. We further employed SHAP (SHapley Additive exPlanations) to evaluate feature significance and provide insights into screening amine catalysts for ROP reaction. The top 10 amines with high PA values are enumerated and discussed. This workflow of coupling machine learning and the DFT approach proves an effective way for screening amine catalysts in ROP reaction.