Predicting Rate Constants of Reactive Chlorine Species toward Organic Compounds by Combining Machine Learning and Quantum Chemical Calculation

Reactive chlorine species (RCS), such as chlorine (HOCl/OCl–), chlorine dioxide (ClO2), chlorine atom (Cl•), and dichlorine radical (Cl2•–), play a crucial role in oxidation and disinfection worldwide. In this study, we developed machine learning (ML)-based quantitative structure–activity relationship (QSAR) models to predict the rate constants of RCS toward organic compounds by using quantum chemical descriptors (QDs) and Morgan fingerprints (MFs) as input features along with three tree-based ML algorithms. The ML-based models (RMSEtest = 0.528–1.131) outperform multiple linear regression-based models (RMSEtest = 0.772–4.837). Moreover, the QSAR models developed by combining QDs and MFs as input features (RMSEtest = 0.528–0.948) show better prediction performance than that by QDs (RMSEtest = 0.616–1.875) or MFs alone (RMSEtest = 0.636–1.439) for all four RCS. The SHapely Additive exPlanation (SHAP) analysis reveals that the energy of the highest occupied molecular orbital (EHOMO), charge, and −O––NH2 and −CO are the most important descriptors affecting the rate constants of RCS. This study demonstrates that the combination of QDs and MFs as input features achieves much better model prediction performance for RCS, which can be extrapolated to other oxidants in water treatment.