OpenChemIE: An Information Extraction Toolkit For Chemistry Literature

TL;DR

OpenChemIE tackles the challenge of extracting comprehensive reaction data from chemistry literature by integrating modality-specific extractions from text, figures, and tables. It introduces a modular pipeline with figure (MolDetect, MolCoref, RxnScribe, MolScribe) and text (ChemNER, ChemRxnExtractor) models, coupled with multimodal integration components for reaction-condition alignment and R-group resolution. On a manually curated substrate-scope dataset (1007 reactions across 78 figures), OpenChemIE achieves an F1 of 69.5% and outperforms several baselines, with end-to-end Reaxys alignment reaching 64.3% accuracy; individual modules show strong performance, especially MolCoref and RxnScribe. The toolkit, available as open-source software and a web portal, enables broader access to robust multimodal chemistry data extraction and sets the stage for future improvements in PDF parsing, molecule recognition, and integration with large language models.

Abstract

Information extraction from chemistry literature is vital for constructing up-to-date reaction databases for data-driven chemistry. Complete extraction requires combining information across text, tables, and figures, whereas prior work has mainly investigated extracting reactions from single modalities. In this paper, we present OpenChemIE to address this complex challenge and enable the extraction of reaction data at the document level. OpenChemIE approaches the problem in two steps: extracting relevant information from individual modalities and then integrating the results to obtain a final list of reactions. For the first step, we employ specialized neural models that each address a specific task for chemistry information extraction, such as parsing molecules or reactions from text or figures. We then integrate the information from these modules using chemistry-informed algorithms, allowing for the extraction of fine-grained reaction data from reaction condition and substrate scope investigations. Our machine learning models attain state-of-the-art performance when evaluated individually, and we meticulously annotate a challenging dataset of reaction schemes with R-groups to evaluate our pipeline as a whole, achieving an F1 score of 69.5%. Additionally, the reaction extraction results of \ours attain an accuracy score of 64.3% when directly compared against the Reaxys chemical database. We provide OpenChemIE freely to the public as an open-source package, as well as through a web interface.

Figures (13)

• Figure 1: Example of a multimodal reaction description drawn from doi:10.1021/jo202590wdoi:10.1021/jo202590w. The reaction template is displayed in a figure, but information regarding R-groups is only contained in the highlighted sections of the table. Moreover, detailed reaction conditions are described in the table and accompanying footnotes.
• Figure 2: OpenChemIE addresses the problem of extracting a list of reactions, containing chemical structures for reactants and products, as well as reaction conditions from a PDF document segmented into figures, text, and tables.
• Figure 3: Overview of OpenChemIE, which receives segmented figures, texts,and tables for processing. The results from individual neural models for each modality are combined through reaction condition alignment and R-group resolution in OpenChemIE to yield a final list of reactions.
• Figure 4: OpenChemIE provides four models for analyzing figures in chemistry literature, including molecule detection, text-figure coreference, reaction diagram parsing, and molecule recognition.
• Figure 5: Illustration of extracting molecules and identifier coreferences from a figure. First, MolCoref determines entity bounding boxes and their correspondence. Then, the molecules are recognized by MolScribe.