Unraveling the catalytic potential of two-dimensional conjugated metal–organic frameworks based on hexaazanaphthalene: machine learning-driven insights into the origin of oxygen evolution-reduction activity

Inspired by hexaazanaphthalene-based conjugated copper metal–organic framework (HATNA-Cu-MOF), we designed 161 HATNA-TM-MOF-based SACs (TM@N_xO_4−x-HATNA) with varying TM or ligands creating distinct coordination environments (x = 0–4) with superior thermodynamic and electrochemical stabilities. Volcano plots can be constructed using (ΔG_OOH* − ΔG_O*)/ΔG_O* as descriptors for oxygen evolution/reduction reaction (OER/ORR) activity, also serving as target parameters for machine learning (ML) models to identify high-performance OER/ORR catalysts. The efficient monofunctional and bifunctional electrocatalysts were successfully predicted, where the ML prediction results well matched the DFT calculation results. We employed Shapley additive explanations (SHAP) for feature analysis and utilized sure independence screening and sparsification operator (SISSO) for generalization. ML analyses reveal that TM-based OER/ORR activities predominantly correlate with three key descriptors: metallic atomic radius, d-orbital electron population, and the heat of formation of the oxide, demonstrating the pivotal role of TM’s inherent electronic configuration and physicochemical characteristics in governing electrocatalytic efficacy. The constant-potential approach emphasizes the key role of electric double-layer capacitance in adjusting the kinetic barrier, where changes in the Fermi level influence the occupation of d-orbitals. Variations in electrochemical potential significantly alter the electronic structure of representative Rh@N₁O₃-HATNA, affecting both the Fermi level and adsorption properties, with the unique 4d⁸5s¹ configuration leading to inverted O₂ adsorption energies as the potential decreases. This study contributes insights into the origin of oxygen evolution-reduction activity for the HATNA-TM-MOF-based SACs and reveals the fundamentals of structure–activity relationships for future applications.

Graphical abstract