Elsevier Adopted MolMole AI for Chemical Extraction
The new tool automates structure data mining from patents and journals to accelerate Reaxys database curation.
Updated on Sept. 25, 2026 in Chemistry
Live Poll
Should AI models used for scientific research be required to be open for independent peer review?
Elsevier has partnered with LG AI Research to deploy the MolMole AI model for the automated extraction of chemical structure data. The tool, which is now operational, parses images from scientific literature and patents for inclusion in the Reaxys database.
Why it matters
This implementation aims to scale the curation speed of chemical substance information by replacing manual extraction processes with automated recognition. It reflects a broader shift toward integrating multimodal AI models to digitize legacy scientific data.
MolMole utilizes molecule detection, reaction-diagram parsing, and optical chemical structure recognition to process image data. The model is currently unable to extract metal-organic complexes or metal-organic frameworks.
The players
Elsevier
An academic publishing and information analytics company that maintains the Reaxys chemical database.
LG AI Research
The artificial intelligence research arm of LG focused on developing multimodal models for industrial and scientific applications.
The details
MolMole functions by identifying structural diagrams within images found in technical documents and translating them into machine-readable data. Each automated extraction is validated against existing Reaxys benchmarks before being committed to the database. The source code for the model is not publicly available, keeping the architecture proprietary to the partnership.
Timeline
September 25, 2026: Article publication date.
The Tech Race
This development follows an industry-wide trend of digitizing vast, image-heavy scientific literature archives using proprietary computer vision models. It positions Elsevier to improve the update frequency of the Reaxys database relative to competing chemical indexing services.
Researchers and database users will see accelerated updates to chemical structure entries within the Reaxys platform as manual bottlenecks are removed. There is no public interface or open-source version of MolMole, meaning the tool remains restricted to Elsevier's internal production pipeline.
The takeaway
The transition to AI-automated extraction highlights the increasing reliance on proprietary vision models to manage the volume of chemical data trapped in static image formats. Users should monitor updates to the Reaxys database for changes in the scope of supported chemical classes, particularly as the model evolves to handle complex metal-organic structures.
Further reading
For broader context on current methods in chemical data management, visit Chemistry.
Live Poll
Should AI models used for scientific research be required to be open for independent peer review?
