Machine Learning-Guided Discovery of Alloy Nanoclusters: Steering Morphology-Based Activity and Selectivity Relationships in Bifunctional Electrocatalysts

Nanoclusters designed with atomic precision are poised to transform next-generation electrode materials for energy devices due to their exceptional performance. However, traditional computational studies often focus solely on individual nanoclusters, neglecting the impact of structurally diverse, low-energy isomers that coexist in a sample. Herein, we present a data-driven approach to screen late-transition metal-based core-shell nanoclusters for bifunctional electrocatalysis. Utilizing geometric and electronic properties, we establish morphology-based relationships for activity and selectivity, emphasizing the critical role of structural diversity in fuel cell applications. We identify the unique single-cluster catalyst identity of M₅₅ nanoclusters, where intermediate adsorption is primarily governed by the constituent metals' electronic and elemental characteristics. Our findings identified the Au₄₈W₇ nanocluster as the most efficient electrocatalyst, exhibiting the lowest bifunctional overpotential of 0.76 V, with η_OER = 0.33 V and η_ORR = 0.43 V, highlighting its outstanding catalytic performance at the nano regime. Guided by the Sabatier principle, we highlight the limitations of conventional numerical methods and reshape the activity volcano, transitioning from RuO₂ and Pt to Au/Ag-based nanoclusters. Furthermore, the trained ML model enables the screening of electrocatalysts for two- and four-electron pathways, steering selectivity between H₂O₂ and H₂O formation. This study provides intuitive guidelines for designing efficient bifunctional electrocatalysts, redefining activity volcanoes, and modulates selectivity in nanocluster alloys.