Predictive modeling and mechanistic interpretation of single electron transfer reactivity between antibiotics and...

In water treatment systems and natural aquatic environments enriched with halogen ions, reactive halogen species (RHS) can degrade antibiotics via addition, hydrogen abstraction, or single electron transfer (SET). SET is often the dominant pathway for electron-rich antibiotics and initiates subsequent transformation reactions. This study employs quantum chemical methods to calculate the SET reaction activation energy (Δ^‡G⁰_SET) for reactions between five key RHS and 40 antibiotic-like contaminants. Building on these calculations, we constructed both a quantitative structure-activity relationship (QSAR) model and nine distinct machine learning (ML) models to predict the Δ^‡G⁰_SET between antibiotic contaminants and halogen radicals. The constructed models demonstrated satisfactory predictive performance and robustness, including Artificial Neural Network (Cl^•), Decision Tree (Br^•), QSAR (Cl₂^•-), catboost (Br₂^•-), and Decision Tree (BrCl^•-). Their scope of application encompasses a wide range of antibiotics. The SHapley Additive exPlanations (SHAP) interpretability analysis confirmed significant variations in the dominant molecular descriptors across different RHS systems, indicating that the SET reaction activity between RHS and antibiotics is governed by system-specific structural and electronic characteristics. In Br^•/Cl^• systems, electronic and spatial descriptors (RDF020m, VE3sign_B(m), Dm) drive the prediction process, while molecular topological descriptors (TDB10r, SpMax_AEA(dm)) play a central role in Br₂^•-/BrCl^•- systems. For the Cl₂^•- system, long-range spatial interactions characterized by RDF110m reveal strong correlations between antibiotic molecular structures and Δ^‡G⁰_SET. Collectively, this work provides a predictive and interpretable modeling framework for assessing the transformation potential of antibiotic-like pollutants in halogen-rich aquatic environments.