Elsevier is partnering with LG AI Research to integrate specialized chemistry artificial intelligence into its Reaxys database and search engine, automating the extraction of chemical structures, reaction schemes, and images from scientific publications and patents. According to Elsevier, the technology captures substance information far more quickly and at a larger scale than was previously possible.
How LG AI Research Built MolMole for Chemical Extraction
The underlying technology, developed by LG AI Research and called MolMole, combines molecule detection, reaction-diagram parsing, and optical chemical structure recognition (OCSR) into a single model. This architecture extracts machine-readable structures straight from images found in scientific literature. Mirit Eldor, managing director, life sciences, at Elsevier, notes that the system lifts chemistry out of static images and into a searchable format. “Our partnership with LG AI Research gives that time back, lifting more chemistry out of the image and into Reaxys – curated, searchable and ready to act on,” Eldor says. “A structure buried in a figure should be evidence rather than a dead end.”
Did you know? Traditional manual data curation requires chemists to spend hours deciphering figures and images to identify previously synthesized compounds, an operational bottleneck this AI integration aims to eliminate.
Validation, Benchmarking, and Industry Response
LG AI Research states that its tool outperforms its competitors when extracting chemistry from a full document page. To maintain data integrity, Elsevier validates each extraction against existing Reaxys benchmarks before deploying it live. The pipeline underwent rigorous testing across Elsevier data and workflow tools prior to wider use.
Lutz Weber, co-founder of the German software company MolGenie, points out that while the tool improves curation speed, it cannot yet extract complex structures like metal-organic complexes or metal-organic frameworks. Weber also highlights that the source code remains unavailable for independent testing, meaning reported performance advantages over open-access tools like Decimer or MolScribe cannot be validated, especially since the paper appeared merely as an unreviewed arXiv publication.
Echoing these concerns, Christoph Steinbeck, an analytical chemist at Friedrich-Schiller-University Jena whose team develops Decimer, questions the reproducibility of the findings. Steinbeck points out that the model has not been published and carries a licence barring commercial use and derivative works, which complicates independent benchmarking and peer validation.
Frequently Asked Questions
What is MolMole?
MolMole is an AI model developed by LG AI Research that combines molecule detection, reaction-diagram parsing, and optical chemical structure recognition to extract machine-readable data from document images.
How does Elsevier use this AI technology?
Elsevier integrates the technology into its Reaxys database to rapidly extract images, drawings, and reaction schemes from patents and scientific journals.
Can independent researchers test the MolMole model?
Stay Updated on Chemical Informatics
Explore more articles on our site to keep pace with how artificial intelligence is transforming research data management and scientific discovery.