The traditional development of catalysts has long depended on trial-and-error methods—an approach that is time—consuming and often yields inconsistent data. To advance the precision and efficiency of catalyst design, transitioning to a data-driven, automated paradigm has become imperative.
In a perspective article published in Matter, a research group led by Prof. DENG Dehui from the Dalian Institute of Chemical Physics (DICP) of the Chinese Academy of Sciences (CAS), in collaboration with Dr. LI Haobo's group from Nanyang Technological University, systematically reviewed the transformative role of artificial intelligence (AI) in the design and synthesis of heterogeneous catalysts. The perspective also outlines future directions for AI-driven innovations in this field.
Schematic illustration of the AI-driven catalyst design and synthesis paradigm, and the development of automated chemical synthesis techniques alongside the advancements in machine learning methods (Image by ZHANG Longhai and BING Qiming)
The perspective highlights machine learning (ML) as a powerful tool for predicting catalyst structure-property relationships, optimizing synthesis conditions, and enabling high-throughput automated calculations and experiments. By identifying key performance descriptors, ML reduces reliance on resource-intensive theoretical calculations such as density functional theory (DFT), thereby accelerating the catalyst discovery process. Advanced techniques such as active learning and generative models further enhance design efficiency by prioritizing critical experiments and proposing novel catalyst candidates.
A central focus of this perspective is the development of AI-powered closed-loop systems that integrate automated synthesis, characterization, and optimization. These systems improve data quality, minimize human error, and ensure reproducibility across the entire catalyst development cycle.
The authors also pointed out current challenges, including the limited generalizability of AI models across diverse catalytic systems, the difficulty of integrating multidisciplinary datasets, and the need for better anomaly detection in automated workflows. To address these, the authors proposed technological roadmaps emphasizing cross-institutional data sharing and adaptive AI frameworks.
"This Perspective provides a blueprint for transitioning catalysis research toward fully automated and intelligent paradigms, unlocking unprecedented efficiency in catalyst development," said Prof. DENG.