The oxygen evolution reaction (OER) is a key bottleneck in clean energy conversion due to sluggish kinetics and high overpotentials. Transition metal single-atom catalysts offer great promise for OER optimization thanks to their high atomic efficiency and tunable electronic structures. However, intrinsic scaling relationships between adsorbed intermediates limit catalytic performance and complicate discovery through conventional machine learning (ML). To overcome this, we combined density functional theory (DFT) with a progressive learning strategy within an active learning framework. By first predicting adsorption energies as auxiliary features, our ML model achieved improved sensitivity to rare, high-activity candidates. High-throughput screening of 261 transition metal single-atom-doped metal oxides (MSA-MOx) identified nine top-performing catalysts (theoretical overpotential < 0.5 V), including MnSA-RuO2 and FeSA-TiO2 (theoretical overpotential < 0.3 V). Data mining revealed key theoretical descriptors governing OER activity, while electronic structure analysis pinpointed intermediate binding strength as the key performance driver. Further constant-potential DFT calculations and experimental evaluation of MnSA-RuO2 confirmed its low overpotential and excellent durability under acidic conditions. This integrated framework, which connects theoretical modeling, machine learning prediction, and experimental validation, accelerates the discovery of efficient OER catalysts and provides mechanistic insights for the rational design of materials in sustainable energy technologies.
Keywords: Active Learning / Machine Learning; Oxygen evolution reaction; Transition Metal-Doped Oxides; density functional calculations; single-atom catalysts.
© 2025 Wiley‐VCH GmbH.