Data-Driven Electrocatalyst Discovery: Recent Trends in Machine Learning Approaches and Descriptor-Based Design Principles

Conspectus The urgent transition to sustainable energy systems has intensified the search for advanced electrocatalysts that efficiently promote key reactions, including hydrogen evolution reaction (HER), oxygen evolution reaction (OER), oxygen reduction reaction (ORR), carbon dioxide reduction reaction (CO 2 RR), and nitrogen reduction reaction (NRR). However, the enormous chemical and structural diversity among candidate materials makes traditional trial-and-error screening highly inefficient. Recent advances in data-driven discovery, especially descriptor-based strategies and machine learning (ML), are transforming this landscape. By combining high-throughput first-principles calculations, curated materials databases, and interpretable ML algorithms, researchers can systematically reveal quantitative relationships between structure and activity, identify promising catalyst candidates, and accelerate the design and screening of efficient catalytic systems. This Account highlights how computational modeling, ML algorithms, data mining techniques, and descriptor engineering (e.g., d-band center, e g orbital filling, and ionization energies), along with experimental verification, together provide a robust framework for rational catalyst development. In this Account, we survey recent progress in data-driven electrocatalyst discovery across the major energy conversion reactions. The integration of interpretable ML models, such as least absolute shrinkage and selection operator (LASSO), sure independence screening and sparsifying operator (SISSO), and subgroup discovery (SGD), with high-quality data sets enables the discovery of both global and local relationships between structure and activity, overcoming limitations in conventional models like volcano plots and linear scaling relations. Through a series of case studies, we demonstrate that the end-to-end, data-driven workflows enable rapid, reliable screening of large catalyst libraries, including single-atom and dual-atom motifs on two-dimensional (2D) materials, basal planes of 2D substrates, and metal–organic frameworks (MOFs). Together, descriptor-based design and interpretable ML not only yield mechanistic insights that guide experimental synthesis and optimization, but also establishes a new paradigm for catalyst discovery, paving the way for breakthroughs in sustainable energy technologies.