Wang Jiaqian, Hu Xiaojuan, Jiang Ying, Yuan Wentao, Yang Hangsheng, Han Zhong-Kang, Wang Yong
Center of Electron Microscopy, School of Materials Science and Engineering, Zhejiang University, Hangzhou 310027, China.
JACS Au. 2025 Jun 4;5(6):2762-2769. doi: 10.1021/jacsau.5c00339. eCollection 2025 Jun 23.
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.
数据驱动的方法在催化剂设计领域日益重要，极大地加速了催化剂的开发。然而，这些方法背后的机制和规则往往缺乏透明度，由于对所涉及的具体过程理解不足，可能导致不可靠的结果。在此，我们开发了一种将分析学习与约束数据挖掘技术相结合的方法，不仅可以识别高性能材料，还能阐明提高其性能的优化途径。使用这种方法，我们筛选了一千多种潜在催化剂，确定了用于析氢反应的顶级单原子催化剂，并绘制出逐步提高性能的优化途径。值得注意的是，我们的研究结果表明，基于调整从整个数据集中识别出的关键因素来提高材料性能的决策可能会产生误导。相反，一种更有效的策略是通过对子数据集进行系统的、逐步的分析来做出决策，特别关注具有共同特征的高性能材料子集。这种方法既增强了从数据中获取知识的能力，又提高了结果的可信度，为推进材料性能合理设计中的数据驱动方法提供了新的见解。
