Machine Learning-Guided Design of Cu-Based Single-Atom Alloy Nanoclusters for Selective CO2 Electroreduction

Metal nanoclusters are highly promising for the electrochemical CO2 reduction reaction (eCO2RR) due to their abundant low-coordination sites and size-dependent electronic properties. Alloying these clusters with isolated transition metal atoms to form Cu-based single-atom alloys (Cu-SAAs) provides a powerful strategy to finely tune their local structures and electronic properties. Because the dopant element identifies and its specific surface site strongly dictate *CO adsorption and subsequent eCO2RR pathways, systematically mapping this combinatorial space is critical. To clarify these effects, we synergistically couple high-throughput density functional theory (DFT) calculations and machine learning (ML) to evaluate *CO adsorption across 1,944 small Cu-SAAs nanocluster candidates. Using reliable DFT-derived adsorption data, we developed an iterative Extreme Gradient Boosting (XGBoost) model capable of rapidly predicting binding energies by identifying the governing structural and electronic descriptors. Analysis of these large-scale predictions highlights how specific combinations of dopant elements and geometrical sites modulate *CO adsorption strength, directly influencing catalytic trajectories. This work establishes an interpretable, data-driven framework for the rational design of Cu-SAAs nanoclusters tailored for targeted eCO2RR products.