Advancing CO<sub>2</sub>RR with O‑Coordinated\nSingle-Atom Nanozymes: A DFT and Machine Learning Exploration

Electrochemical CO<sub>2</sub> reduction\nreaction (CO<sub>2</sub>RR) offers a promising route toward zero-carbon\nemissions. Recently\nemerged single-atom nanozymes (SANs), which combine the advantages\nof single-atom catalysts (SACs) and nanozymes, show potential in CO<sub>2</sub>RR electrocatalysis applications. Herein, we designed 260\ncarbon-supported TMO<sub>4</sub> SANs, introducing heteroatoms outside\nthe coordination sphere to modulate the interaction between the first-coordination\nsphere and the support structure. Using a collaborative workflow of\nthe density functional theory (DFT) and machine learning (ML), we\nassessed these SANs for the CO<sub>2</sub>RR performance. Among them,\n185 SANs demonstrated high stability. Our model, based on extreme\ngradient boosting regression (XGBR) and trained on 100 labeled SANs,\nsuccessfully predicted the limiting potentials (<i>U</i><sub>L</sub>) for 85 SANs. From these, we distinguished 33 SANs with\nsuperior activity, good specific product selectivity, and greater\nhydrogen evolution reaction (HER) suppression compared with benchmark\nTMN<sub>4</sub>-C catalysts. Particularly, MoO<sub>4</sub> supported\non external H-doped graphene (Mo-I) and FeO<sub>4</sub> supported\non external N,O-doped graphene (Fe-VI) exhibited remarkable enzyme-mimicking\ncharacteristics, with Mo-I showing a formate dehydrogenase-like behavior\nand Fe-VI mimicking CO dehydrogenase with <i>U</i><sub>L</sub> values of −0.05 and −0.06 V, respectively. Through\nML feature engineering and electronic structure property analysis,\nwe determined the first-coordination sphere–support interaction\n(FCSSI) as a key descriptor in regulating the catalyst performance.\nWe hope that these insights may provide valuable guidance for exploring\npotential SANs for CO<sub>2</sub>RR.