Chemically Interpretable Machine Learning for Predicting HER Activity in Au-Based Alloys

Hydrogen evolution reaction (HER) catalysts are essential for renewable hydrogen production, but conventional screening through density functional theory (DFT) remains computationally demanding. Here, we present a proof-of-concept machine learning (ML) framework that predicts hydrogen adsorption energies on Au-based alloys using a curated dataset and nine chemically interpretable descriptors. Tree-based ensemble models achieve quantitative accuracy (mean absolute error ≈ 0.13 eV), sufficient to resolve weak, optimal, and strong adsorption regimes. Feature attribution highlights generalized coordination number and site-specific electronic descriptors as dominant factors, consistent with established chemisorption principles. Guided by these insights, the trained model was applied to a library of Au–M alloys, with targeted DFT validation identifying Au–Mo, Au–W, Au–Ta, Au–Hf, and Au–Tc as promising candidates that couple favorable adsorption with alloy stability. Beyond specific predictions, this work introduces an accessible, reproducible workflow that connects ML predictions to catalytic intuition. The framework serves as a reference for experimental chemists, lowering the barrier for integrating data-driven approaches into catalyst discovery.