Harnessing machine learning for high-entropy alloy catalysis: a focus on adsorption energy prediction

npj | computational materials Review article Published in partnership with the Shanghai Institute of Ceramics of the Chinese Academy of Sciences https://doi.org/10.1038/s41524-025-01579-5 Harnessing machine learning for high-entropy alloy catalysis: a focus on adsorption energy prediction Check for updates 1234567890():,; 1234567890():,; Qi Wang 1 & Yonggang Yao 2 High-entropy alloys (HEAs) have emerged as promising candidates for catalyst applications due to their inherent compositional, structural, and site-level diversities, which enable highly tunable catalytic properties. However, these complexities pose grand challenges for traditional “trial-and-error” experimentation or computationally expensive “brute-force” ab initio calculations. Machine learning (ML) demonstrates great potential to address these challenges by establishing efficient, scalable mappings from composition, structure or site environment to HEA properties. Among these properties, adsorption energy, which quantifies the binding strength between catalytic intermediates and surface sites, is a crucial indicator of catalytic activity. This review provides a comprehensive overview of ML-driven strategies for adsorption energy prediction in the context of HEAs. Two primary strategies are introduced: “direct” prediction from unrelaxed structure and “iterative” prediction via ML potential-guided relaxation modeling. Both strategies can leverage handcrafted features or end-toend frameworks such as graph neural networks. We also discuss how pretrained models on largescale databases can extend to out-of-domain HEA systems. Beyond methodology, we address key challenges and future directions, including benchmarking ML strategies, developing HEA-specific datasets, pretraining and fine-tuning, integrating chained ML models, advancing multi-objective optimization, and bridging ML predictions with experimental validation. By critically evaluating existing strategies and highlighting emerging trends, this review underscores the critical role of ML in advancing adsorption energy predictions, offering a foundation for accelerating the discovery and optimization of HEA catalysts. Catalysis plays a fundamental role in a wide array of industrial and environmental processes, from energy conversion to chemical manufacturing1,2. The efficiency and selectivity of catalysts are critical in determining the sustainability and economic viability of these processes. Traditionally, catalysts have been composed of single or binary metals3, but the search for improved performance has led researchers to explore more complex materials, such as high-entropy alloys (HEAs)4,5. HEAs are initially defined as consisting of five or more elements in approximately equal proportions, with individual element concentrations ranging from 5% to 35%, as proposed by Yeh et al.6. In recent years, it has been found that maximized configurational entropy is not the only critical factor for superior properties7–9, and “high entropy” is increasingly recognized as an alloy design strategy rather than being strictly defined by compositional rules10. The focus shifts towards leveraging the intermixing of multiple elements to tailor material properties11,12. In addition to catalytic properties, HEAs exhibit a diverse range of remarkable properties, including mechanical strength and ductility13,14, thermal stability15, soft magnetic properties16, corrosion resistance17, and radiation tolerance18, making them promising for applications in energy storage, structural engineering, magnetic materials, or nuclear reactors. The largest advantage HEAs offer for catalysis lies in their diversity to potentially optimize catalytic performance for a given reaction19. By incorporating a diverse combination of elements, HEAs increase the likelihood of forming surface microstructures that exhibit optimal interaction characteristics for key catalytic intermediates20. Meanwhile, the enhanced stability and corrosion resistance of HEAs also makes them appealing for catalytic applications, especially under harsh conditions where conventional catalysts tend to degrade17. HEAs have already shown great promise across a 1 Institute of Materials, China Academy of Engineering Physics, Mianyang, Sichuan, China. 2State Key Laboratory of Materials Processing and Die & Mould e-mail: Technology, School of Materials Science and Engineering, Huazhong University of Science and Technology, Wuhan, China. npj Computational Materials | (2025)11:91 1 https://doi.org/10.1038/s41524-025-01579-5 Review article variety of reactions, including hydrogen evolution reaction (HER)21, oxygen evolution/reduction reactions (OER/ORR)22,23, nitrogen reduction reaction (NRR)24 and CO2 reduction reaction (CO2RR)25. Recent advances in fabrication techniques, such as carbothermal shock26,27, fast-moving bed pyrolysis28, microwave heating29, solvothermal synthesis30, and liquidmetal-based methods31, have notably expanded the range of possible element combinations and compositions for HEAs. These synthetic achievements underscore the feasibility of producing HEAs even beyond the quinary level or incorporating elements that are typically immiscible. Such advancements unlock new opportunities for catalyst design and enable the creation of HEAs with highly tunable performance profiles. Despite the remarkable potential of HEAs, their complexity presents significant challenges in deciphering the relationships between composition, structure, and catalytic property, hindering rational HEA catalyst design. Traditional experimental methods often lack the spatial and temporal resolution required to precisely probe adsorption behaviours or capture fleeting reaction intermediates, facing limitations in establishing clear relationships. Computational approaches—particularly first-principles calculations based on density functional theory (DFT)32–34—have emerged as vital tools for pinpointing the roles of specific active sites, probing adsorption interactions, and unraveling reaction pathways at the atomic level19. Techniques such as the computational hydrogen electrode method22 bolster our ability to assess electrocatalytic thermodynamics or kinetics. This detailed insight guides the identification of favorable reaction pathways on HEAs and fosters rational catalyst design. Nonetheless, the vast compositional variety of HEAs, coupled with their intricate structural and site-level diversity, creates a combinatorial explosion of possible configurations (Fig. 1, as detailed later). This complexity not only makes experimental synthesis and testing time-consuming and expensive, but also limits the ability of DFT calculations to explore the full range of HEA configuration space. This is where machine learning (ML) becomes critical. Renowned for its ability to identify patterns and make predictions from data, ML has emerged as a crucial tool in materials science35–39, aided by the emergence of large-scale databases such as the Mat (...truncated)