Identifying Key Factors Influencing Advanced Hydrogen Evolution Reaction Catalysts

Data-driven approaches are increasingly vital in the field of catalyst design, significantly accelerating catalyst development. However, the mechanisms and rules underlying these approaches often lack transparency, potentially leading to unreliable outcomes due to an insufficient understanding of the specific processes involved. Here, we developed a method that combines analytical learning with constrained data mining techniques to not only identify high-performance materials but also elucidate the optimization pathways for enhancing their performance. Using this method, we screened over a thousand potential catalysts, identifying top-performing single-atom catalysts for the hydrogen evolution reaction and mapping out optimization pathways to progressively improve performance. Notably, our findings suggest that decisions aimed at enhancing material performance, when based on tuning key factors identified from entire data sets, can be misleading. Instead, a more effective strategy is to make decisions through a systematic, step-by-step analysis of subgroup data sets, specifically focusing on subsets of high-performance materials that exhibit common characteristics. This approach enhances both the development of knowledge from data and the trustworthiness of the results, offering new insights for advancing data-driven approaches in the rational design of material properties.