The research team led by Assistant Professor Yulian He of the University of Michigan-Shanghai Jiao Tong University Joint Institute (UM-SJTU JI, JI hereafter) has made a significant contribution in catalyst design optimization. Their findings, published in the prestigious scientific journal, Proceedings of the National Academy of Sciences of the United States of America (PNAS), introduce a novel method for determining the pivotal physical quantity influencing chemisorption energy on catalyst surfaces.

The study, titled “Interpreting Chemisorption Strength with AutoML-based Feature Deletion Experiments,” utilizes automatic machine learning (AutoML) and the extraction of knowledge from the high-throughput density functional theory (DFT) database. This methodology addresses the intrinsic complexity of catalyst surfaces and chemisorption reactions, providing valuable insights into the optimal catalyst.

For binary alloy catalyst surfaces, the research reveals that the local geometric information of adsorption sites is the crucial physical quantity determining chemisorption energy, surpassing the influence of intrinsic electronic or physicochemical properties. Combining feature deletion experiments with explainable artificial intelligence (XAI) tools, the study identifies an optimal feature set containing 21 inherent, non-DFT-calculated physical quantities. This approach achieved an impressive average absolute error (MAE) of 0.23 eV in over 8,400 chemical adsorption reactions involving more than 1,600 alloy surfaces.

The research not only advances catalyst design optimization but also introduces a stable and consistent methodology for feature deletion experiments based on AutoML. Unlike traditional models, this approach avoids the complexity-interpretability trade-off, emphasizing the evaluation of feature set performance. The strategy aligns with traditional experimental science methods, providing a promising avenue for knowledge extraction in various research directions.

JI Assistant Professor Yulian He and Associate Professor Cheng Hua of the SJTU Antai College of Economics and Management are the corresponding authors of the paper, with postdoctoral fellow Zhuo Li as the first author, and undergraduates Changquan Zhao and Haikun Wang as co-second authors. Yanqing Ding and Yechao Chen are co-authors.

The project received support from the Young Scientists Fund of the National Natural Science Foundation of China, the Major Program of the National Natural Science Foundation of China, Shanghai Committee of Science and Technology and Chenguang Program of Shanghai Education Development Foundation and Shanghai Municipal Education Commission.

Weblink of the paper:https://www.pnas.org/doi/10.1073/pnas.2320232121

Author Introduction

Zhuo LiPostdoctoral Researcher at UM-SJTU Joint Institute, graduated from Peking University with a bachelor’s degree, and graduated from the University of Rochester with master’s and doctoral degrees. Honors include the 2022 Shanghai Jiao Tong University (SJTU) Morning Star Postdoctoral Fellowship and the 2024 Ministry of Education Postdoctoral Overseas Talent Program.

Changquan Zhao

Senior student at the SJTU School of Mathematical Sciences, soon to pursue a Ph.D. at the UM-SJTU Joint Institute. Research interests primarily focus on catalyst design optimization.

Haikun Wang

Senior student at the SJTU Antai College of Economics and Management, soon to pursue a Master’s degree at SJTU. Research interests mainly involve machine learning.

Yanqing Ding

Graduated in 22nd batch from the SJTU School of Chemistry and Chemical Engineering, obtained a Master’s degree from Columbia University.

Yechao Chen

Senior student at the SJTU Antai College of Economics and Management, research interests mainly include machine learning and deep learning.

Yulian He

Assistant Professor jointly appointed at the UM-SJTU Joint Institute and SJTU School of Chemistry and Chemical Engineering. Ph.D. graduate from the School of Chemistry and Environmental Engineering, Yale University. Published over thirty SCI papers and authored four book chapters. Research interests include catalyst synthesis and design, C1 catalysis, heterogeneous catalysis, structure-activity relationships, catalytic oxidation, hydrogenation reactions, and data-driven rational design of catalysts. Funded by various organizations including the National Natural Science Foundation of China, the National Development and Reform Commission, the Shanghai Municipal Science and Technology Commission, and multiple enterprises.

Cheng Hua

Associate Professor at SJTU Antai College of Economics and Management. Graduated from SJTU with a Bachelor’s degree and obtained a Ph.D. degree from the Yale School of Management. Published numerous academic papers in renowned international journals such as Manufacturing & Service Operations Management, Naval Research Logistics, and Transportation Research Part E. Received 7 international paper awards in management science and operations research. Also nominated for the Kaiyuan Top Ten Teachers and recognized as one of the most popular teachers among undergraduate students at SJTU. Research interests include data-driven decision-making, stochastic processes and approximation algorithms, healthcare optimization management, service systems, machine learning, and artificial intelligence.