image: The schematic diagram illustrates the data flow and core components of Catal-GPT: from the data collection to the AI-driven catalyst optimization.
Credit: ©Science China Press
In the intersection of artificial intelligence (AI) and catalytic science, researchers at Shenyang University of Chemical Technology have leveraged the open-source large language model qwen2 to create Catal-GPT: an intelligent platform established specifically for accelerating catalyst discovery and optimization. The Catal-GPT streamlines the generation of catalyst formulations, marking a significant leap toward efficient, data-driven industrial catalysis.
Unlike traditional trial-and-error approaches, the Catal-GPT establishes a collaborative human-AI workflow. Its modular architecture integrates the datasets encompassing catalyst synthesis, characterization, and performance. After the data cleaning and encoding, these inputs train a customized GPT model capable of responding to researcher queries much like an expert consultant. With a remarkable 92% accuracy in knowledge extraction from scientific literature, the Catal-GPT not only retrieves relevant information but also generates experimental protocols for catalyst preparation.
Looking forward, the team is tackling cross-system adaptability: by decoupling general knowledge from reaction-specific parameters from task-specific parameters, the Catal-GPT aims to enable seamless switching between diverse catalytic systems, which will transform catalyst discovery from serendipity-based “searching blind” to AI-guided “precision targeting”. As higher-quality data enriches its training corpus, this platform holds immense potential for advancing the development of industrial catalysts.
Journal
National Science Review