Effective Screening Descriptors of Metal–Organic\nFramework-Supported Single-Atom Catalysts for Electrochemical CO<sub>2</sub> Reduction Reactions: A Computational Study

Metal–organic\nframework-supported single-atom catalysts\n(SACs@MOF) show considerable promise in CO<sub>2</sub> reduction reactions\n(CO<sub>2</sub>RR). However, efficiently screening and designing optimal\ncatalysts is hindered by the lack of effective descriptors for encoding\nthe complex chemical microenvironments in SAC@MOF systems. Herein,\nthrough combining an intuition-guided dimensionality reduction strategy\nwith machine learning (ML), we identified critical descriptors based\non atomic features and the SAC’s constrained coordination geometry,\nwhich capture the effects of complex chemical microenvironments on\nelectrochemical CO<sub>2</sub>RR activity and selectivity for UiO-66-supported\nSACs. With these descriptors, accurate ML models were developed to\npredict the limiting potentials for producing HCOOH, CO, and CH<sub>4</sub>/CH<sub>3</sub>OH on 48 SACs@UiO-66-X (X = H, NH<sub>2</sub>, and Br). Moreover, the transferability of the developed descriptors\nand ML models was demonstrated on 48 additional systems with X = CH<sub>3</sub>, OH, and NO<sub>2</sub>. The accuracy of the predicted activity\ntrends for specific SACs combined with different linker groups and\nthe selectivity of the top-performing catalysts were validated through\nadditional DFT calculations. This study provides an effective framework\nfor understanding and modulating chemical microenvironments, enhancing\nthe design and development of MOF-supported SACs for the CO<sub>2</sub>RR.