• Novel model predicts sulfate radical reactivity using simple structural descriptors. • It is mechanistically interpretable without requiring advanced computations. • It provides high accuracy (Q²=0.83) across diverse organic contaminants. • Electron-rich groups accelerate reactivity, whereas fluorine atoms hinder it. • pH and temperature strongly influence predicted reactivity. Sulfate radical-based advanced oxidation processes (SR-AOPs) are critical for removing persistent organic pollutants from water, yet predicting degradation kinetics remains a significant engineering challenge. While machine learning offers predictive capabilities, models often lack mechanistic transparency. In this study, a mechanistically interpretable model was developed using a diverse dataset of 557 organic contaminants to predict second-order rate constants with sulfate radicals. The framework utilizes accessible structural descriptors, including functional group counts and the pH / T ratio, achieving high predictive accuracy with an external Q 2 of 0.83. Mechanistic analysis reveals that electron-rich moieties, such as ethers and conjugated double bonds, accelerate degradation through enhanced electron density. Conversely, electronegative fluorine atoms significantly hinder reactivity. This finding has critical implications for the treatment of per- and polyfluoroalkyl substances (PFAS), highlighting the inherent resistance of these "forever chemicals" to conventional SR-AOPs. By incorporating specific corrections for complex molecular geometries, this model provides a practical, low-cost tool for environmental scientists. The approach enables the prioritization of emerging contaminants and the optimization of water treatment strategies without the need for computationally intensive "black box" simulations.
Keshavarz et al. (2026) studied this question.