The extraction of structured chemical information from literature is essential for constructing reaction databases that drive data-driven and AI-powered chemical research. A major challenge lies in the multimodal and complex nature of chemical data. In order to convert the raw data into structured, machine readable datasets, a modeling framework is needed with the ability to comprehend text, tables, and graphical representations of molecular structures. Existing approaches predominantly focus on single tasks [1-3], limiting their ability to fully capture reaction details and leading to incomplete datasets.
To overcome this, we present ChemEagle, a multimodal large language model (MLLM)-based multi-agent system that integrates diverse chemical information extraction tools. By integrating seven expert-designed tools and six chemical information extraction agents, ChemEagle not only processes individual modalities but also utilizes MLLMs' reasoning capabilities to unify extracted data, ensuring more accurate and comprehensive reaction representations. We demonstrated the model’s capability in parsing chemical reaction diagrams, converting molecular structures to SMILES, and substituting functional groups or molecular fragments from tabulated lists. In all these tasks our model performed better than existing methods by a significant margin. Our approach presents a significant step towards automated chemical knowledge extraction, facilitating more robust AI-driven chemical research.
Reference
[1] Guo, J.; Ibanez-Lopez, A. S.; Gao, H.; Quach, V.; Coley, C. W.; Jensen, K. F.; Barzilay, R. Automated chemical reaction extraction from scientific literature. J. Chem. Inf. Model. 62, 2035-2045 (2021).
[2] Vaucher, A. C.; Zipoli, F.; Geluykens, J.; Nair, V. H.; Schwaller, P.; Laino, T. Automated extraction of chemical synthesis actions from experimental procedures. Nat. Commun. 11, 3601 (2020).
[3] Qian, Y.; Guo, J.; Tu, Z.; Coley, C. W.; Barzilay, R. RxnScribe: A sequence generation model for reaction diagram parsing. J. Chem. Inf. Model. 63, 4030-4041 (2023).

Yong Huang

Appearances