The hidden engine of AI in electrocatalysis: Databases and knowledge graphs at work
Di Zhang, Hao Li
Abstract
The development of efficient electrocatalysts is crucial for advancing renewable energy technologies, such as hydrogen production, carbon dioxide reduction, and fuel cells. However, the complexity of electrocatalytic systems and the vast chemical space of potential materials present significant challenges for traditional discovery methods. Databases and knowledge graphs have emerged as indispensable tools in this domain, enabling the fine-tuning of large language models (LLMs) and the development of robust machine learning (ML) models. By organizing and integrating vast amounts of experimental and computational data, these tools facilitate the extraction of relevant knowledge, improve model performance, and accelerate the discovery of high-performance electrocatalysts. This perspective explores the synergistic role of databases and knowledge graphs in electrocatalyst discovery, focusing on their contributions to LLM fine-tuning, ML model development, and the creation of predictive tools. We also discuss the challenges associated with data quality, standardization, and integration, and propose strategies to maximize the potential of these tools in AI-driven electrocatalysis research. • A systematic workflow is proposed to construct physically meaningful knowledge graphs from electrocatalysis databases. • Overview and comparison of databases: DigCat, ElectroCat, Catalysis-Hub, and others. • DFT-based descriptors are integrated into knowledge graphs for mechanistic insights. • A unified framework links experiments, theory, and AI via knowledge graphs. • Key challenges in data quality, AI errors, extraction, and sustainability are discussed.