Machine learning-guided design principles for single-atom catalysts in persulfate-based advanced oxidation processes

Single-atom catalysts (SACs) have recently garnered significant attention in persulfate-based advanced oxidation processes (PS-AOPs) owing to their distinctive coordination environments, maximum atom utilization efficiency, and tunable reaction pathways. However, exhaustive experimental screening of SAC configurations is prohibitively time-consuming, necessitating machine learning (ML) to rapidly establish predictive structure–activity relationships. This review highlights ML-guided design principles for SACs in PS-AOPs, shifting the paradigm from empirical trial-and-error to a rational inverse design. The fundamental principles of SAC-mediated PS activation establish a mechanistic foundation for catalytic oxidation. Furthermore, the pivotal microenvironmental effects governing spatial organizations and electronic configurations are underscored. Within this framework, common ML descriptors are analyzed with special emphasis on the multidimensional features governing catalyst–oxidant–pollutant interactions. A closed-loop “data–model training–validation” workflow is established, integrating ML models with experimental evidence to ensure mechanistic accuracy for SAC-driven PS-AOPs and decode algorithmic “black boxes”. Finally, future challenges, from optimizing ML application frameworks to enhancing comprehensive datasets, are discussed to drive next-generation experiment–machine coordination. This review aims to chart pathways for rational design of high-quality SACs by ML methodologies to establish efficient PS activation platforms.