Machine Learning-Guided Coordination Engineering of M–N–C Single-Atom Electrocatalysts for Superior Oxygen Reduction

Precisely tailoring metal–nitrogen-carbon (M–N–C) single-atom catalysts (SACs) with high catalytic activity and selectivity for specific chemical reactions remains challenging due to the lack of a qualitative descriptor between their catalytic properties and coordination geometries. Herein, we bridge this gap by integrating density functional theory (DFT) calculations with machine learning (ML) algorithms to deconvolute the electrocatalytic oxygen reduction reaction (ORR) activity of M–N–C SACs across various possible coordination configurations. By correlating the theoretical overpotentials with structural features, an interpretable descriptor simultaneously reflecting the coordination number and the metal–support interaction is identified. This descriptor not only reliably describes the ORR performance trends across diverse metal centers in SACs but also provides a general guideline for engineering coordination geometry to optimize catalytic performance. Guided by these insights, the predicted Cu-SAC featuring low-coordinated Cu–N 3 moieties is synthesized, delivering remarkable ORR activity compared with the conventional Cu–N 4 sites while maintaining robust structural stability under prolonged electrochemical operation. This study highlights the exceptional potential of interpretable ML combined with theoretical and experimental strategies in elucidating complex structure–property relationships in SACs and accelerating the development of next-generation electrocatalysts for sustainable and efficient energy conversion.