
Selective photocatalytic CO2 reduction to value-added chemicals offers an important route for solar-to-chemical energy conversion. However, the multi-electron and multi-pathway nature of CO2RR makes product selectivity difficult to regulate and limits both the rational design and rapid screening of photocatalysts toward targeted products. In this work, density functional theory (DFT) calculations and machine learning (ML) were combined to investigate CO2RR selectivity on M@N-WS2 catalysts. Rh@N-WS2 and Pd@N-WS2 preferentially produce CH3OH, whereas Ir@N-WS2 and Pt@N-WS2 favor CH4 formation. Integrated crystal orbital Hamilton population (ICOHP) analysis identifies *CH2OH as the pivotal branching intermediate and shows that its C–metal bonding strength governs product preference. A descriptor derived from the sure independence screening and sparsifying operator (SISSO) was established to predict ΔG values associated with the subsequent conversion pathways of *CH2OH and identify the favored C1 product, with ICOHP identified as the dominant factor governing pathway preference. This work provides atomic-level insight into product selectivity regulation in WS2-based single-atom photocatalysts and offers a descriptor-guided strategy for rapid selectivity prediction.
single-atom catalysts; product selectivity; density functional theory; machine learning; ICOHP