Unraveling the ranitidine removal mechanism in a novel riboflavin-mediated ultraviolet/peracetic acid system: An integrated density functional theory-machine learning framework.

In this study, a novel riboflavin-mediated ultraviolet/peracetic acid system (RF/UV/PAA) was developed to enhance the degradation of ranitidine. Under optimal conditions, ranitidine was completely removed within 15 min, leading to a 68.9% mineralization rate due to the combined action of radical (hydroxyl radicals (·OH) and carbon-centered organic radicals (R-C·) and nonradical (singlet oxygen (1O2)) reactive oxidative species (ROS). Electron paramagnetic resonance and quenching experiments indicated that RF significantly promoted ROS generation, with the respective contributions of ·OH, R-C· and 1O2 to ranitidine removal being 54.5%, 15.8%, and 24.3%, respectively. Demethylation, monooxygenation, dioxygenation, and C-S bond cleavage were identified as the primary degradation pathways for ranitidine. These pathways were initially inferred from structural analysis of transformation products (TPs) and subsequently validated through density functional theory (DFT) calculations and the TP-Transformer model. Furthermore, an integrated DFT-machine learning (ML) framework was employed to elucidate the degradation mechanism, with a specific focus on the interactions of principal molecular features with ROS. The results revealed that FractionCSP3 (SHAP value = 0.30) and MolWt (SHAP value = 0.44) were the dominant predictors of ·OH and 1O2 reactivity, respectively, whereas FormalCharge (SHAP value = 0.40) governed the R-C·-mediated processes. In addition, SHAP-based interaction analysis revealed that electron-rich sites facilitate ·OH attack, conjugated π-systems promote 1O2 reactivity, and molecular polarization drives R-C· reactions. Our study provides not only an effective technology for ranitidine removal but also a quantitative, interpretable framework for predicting reactivity and guiding the design of photo-assisted advanced oxidation processes.