Elsevier and LG AI Research have deployed chemistry-specific AI vision technology that extracts chemical substance information from images in patents and scientific literature at scale. The system is now live inside Reaxys, Elsevier's discovery chemistry database, converting drawings, figures, and reaction schemes that were previously invisible to search engines into structured, queryable data.
The technology addresses a long-standing bottleneck in chemical research. Much of the published evidence chemists need - molecular structures, reaction diagrams, stereochemistry details - appears only as images rather than searchable text. When that information cannot be indexed, researchers must manually page through documents to verify whether a compound or reaction has already been described. In fields like inorganic and organometallic chemistry, where complex structures are harder to extract, the problem is acute.
How the AI vision model works
The model from LG AI Research combines three functions in a single architecture: molecule detection, reaction-diagram parsing, and optical chemical structure recognition (OCSR). Unlike general-purpose OCR, it reads chemical meaning - bonds, atoms, stereochemistry, and spatial relationships - not just pixels. A misread bond can identify the wrong compound. A missed structure gives chemists an incomplete picture.
In published benchmarking, the model outperformed alternatives at extracting chemistry from full document pages. Each extraction pipeline is validated against existing Reaxys benchmarks before going live. Elsevier said accuracy is not traded for scale.
Mirit Eldor, Managing Director of Life Sciences at Elsevier, said: "Every hour a chemist spends deciphering figures or images to see what has already been made is an hour that could instead be spent on chemistry discovery. 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. A structure buried in a figure should be evidence rather than a dead end."
Hwayoung Edward Lee, lead of the AI Biz Transformation Unit at LG AI Research, said: "Understanding scientific images requires AI engineered specifically for chemistry - where every bond and spatial layout holds critical meaning. We designed our AI vision model to decode these complex visual representations with human-expert precision."
What the partnership covers next
Reaction extraction is the next phase of the collaboration. That work will extend image-based extraction beyond individual substances to capture full reaction evidence within Reaxys. The two organizations are also exploring additional customer challenges to address together, combining LG AI Research's AI capabilities with Elsevier's chemistry content and curation expertise.
The initiative operates under Elsevier's Responsible AI Principles and Privacy Principles. More details are available on the Reaxys product page.
Why this matters for research scientists
For chemists conducting novelty searches, competitive intelligence, or synthesis planning, this changes the completeness of the evidence base. A structure that previously required manual document review is now findable through a database query. The gain is not just speed - it reduces the risk of missing prior art or duplicating known work. For research scientists looking to apply AI to their own workflows, understanding how domain-specific vision models outperform generic tools is increasingly relevant to AI Learning Path for Research Scientists and broader AI for Science & Research Training.
Your membership also unlocks:
