Machine Learning-Enhanced Design of 2D TM3(HXBHYB)@MOF-Based Single-Atom Catalysts for Efficient Oxygen Electrocatalysis

This study integrates machine learning (ML) and density functional theory (DFT) to systematically investigate the oxygen electrocatalytic activity of two-dimensional (2D) TM₃(HXBHYB) (HX/YB = HIB (hexaaminobenzene), HHB (hexahydroxybenzene), HTB (hexathiolbenzene), and HSB (hexaselenolbenzene)) metal-organic frameworks (MOFs). By coupling transition metals (TM) with the above ligands, stable 2D TM₃(HXBHYB)@MOF systems were constructed. The Random Forest Regression (RFR) model outperformed the others, revealing the intrinsic relationship between the physicochemical properties of 2D TM₃(HXBHYB)@MOF and their ORR/OER overpotentials. Model predictions identified promising systems, including Co₃(HXBHYB) and Ir₃(HXBHYB), with Co₃(HHBHSB) and Co(HIB)₂ exhibiting exceptional ORR (η^ORR = 0.276 V) and OER (η^OER = 0.294 V) activities. SHAP analysis highlighted the valence electron count and atomic radius of the TM as critical descriptors, with the interaction between coordinating atoms and TM valence electrons governing catalytic activity. This work provides universal design principles for evaluating ORR/OER activities, offering a high-precision, low-cost method for catalyst screening.