An innovative method for predicting oxidation reaction rate constants by extracting vital information of organic contaminants (OCs) based on diverse molecular representations

The reaction rate constant (k) of oxidants with organic contaminants (OCs) is an important parameter to assess the efficiency of oxidants in removing contaminants. In this study, the degradation of OCs in three oxidation systems was evaluated. The modeling process applied three molecule representations (molecular descriptors (MD), quantum chemical descriptors (QCD) and MACCS fingerprints) and their variable integrations. Models based on integration molecule representations show significant performance improvements. Eventually, the optimal models for ozone, chlorine dioxide and hypochlorite were found to be (MD+QCD)-XGBoost (R2tra = 0.982, Q2tra = 0.715), (MD+QCD+MACCS)-XGBoost (R2tra = 0.982, Q2tra = 0.778), and (MD+QCD+MACCS)-CatBoost (R2tra = 0.856, Q2tra = 0.709) model, respectively. Here, we introduced a new perspective that differed from focusing on machine learning (ML) algorithm optimization. This perspective centered on the input variables (i.e., molecular representations) of models to improve model performance by capturing the key properties of OCs comprehensively. Furthermore, the key effects of pH, ionization potential, orbital energy, polarizability and electronegativity on the oxidation reaction in different oxidation systems were clarified. We hope that the mechanism explanation in this study can provide valuable insights for understanding the mechanism of various oxidation reactions of complex OCs. [Display omitted] •ML-assisted model development for predicting k1, k2 and k3 at different pH.•The model performance was improved by expanding the descriptor type and range.•Factors such as pH, ionization potential and electronegativity have effects on k.