Chemists have long faced a frustrating problem: crucial chemical information buried in images, drawings, and reaction schemes across millions of scientific papers and patents remains invisible to search engines. Now, a partnership between Elsevier and LG AI Research is changing that by deploying specialized artificial intelligence (AI) vision technology that can automatically identify and extract chemical structures from visual content, converting them into searchable, curated data.

Why Are Chemical Drawings So Hard for AI to Read?

Chemical structures are not simple pictures. A bond drawn at the wrong angle, a missed atom, or a misinterpreted spatial relationship can completely change what compound a chemist is looking at. Traditional computer vision models trained on everyday images struggle with this level of precision because they lack chemistry-specific understanding. When a model misreads a single bond, researchers may incorrectly conclude that a compound has never been synthesized before, wasting time on redundant research.

This problem is especially acute in areas like novelty searching, where chemists need to confirm whether a specific compound has already been described in the literature, and in competitive intelligence, where companies track what competitors have already patented. Inorganic and organometallic chemistry, which involves complex three-dimensional structures, has been particularly difficult to index at scale.

How Does This New AI Technology Work?

LG AI Research developed a chemistry-specific AI vision model that combines three specialized capabilities into a single system:

• Molecule Detection: Identifying where chemical structures appear within a document page
• Reaction-Diagram Parsing: Understanding the flow and relationships between reactants, products, and conditions in chemical reaction schemes
• Optical Chemical Structure Recognition (OCSR): Converting visual chemical drawings into machine-readable molecular representations

In published benchmarking tests, this integrated approach outperforms existing alternatives at extracting chemistry from full document pages, placing it at the forefront of innovation for reading chemistry across different formats. The technology is now integrated into Reaxys, Elsevier's discovery chemistry solution, where it processes substance information from patents and journal content far more quickly and accurately than was previously possible.

Importantly, every extraction pipeline is validated against existing Reaxys benchmarks before going live, ensuring that speed does not come at the cost of accuracy. The system has undergone rigorous testing across Elsevier's data and workflow tools to maintain quality standards.

What Does This Mean for Researchers?

The practical impact is significant. Chemists currently spend considerable time manually checking documents to confirm whether a compound or reaction has already been described in the scientific literature. By making that chemistry discoverable and searchable at scale, researchers can redirect that time toward actual chemistry discovery and innovation.

"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," said Mirit Eldor, Managing Director of Life Sciences at Elsevier.

Mirit Eldor, Managing Director, Life Sciences, Elsevier

The technology also addresses a longstanding gap in chemical knowledge management. Much of the substance and reaction information chemists rely on is communicated through figures, drawings, and reaction schemes rather than searchable text. When that chemistry is not indexed, researchers are left with incomplete pictures of what has already been discovered.

What's Next for This Partnership?

Elsevier and LG AI Research are already planning the next phase of their collaboration. Reaction extraction, which extends image-based extraction beyond individual substances to capture the full reaction evidence available in scientific literature, is the immediate priority. The companies are also exploring additional customer challenges to tackle together, combining LG AI Research's specialized AI capabilities with Elsevier's chemistry content expertise and scientific curation processes.

"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," explained Hwayoung Edward Lee, lead of the AI Biz Transformation Unit at LG AI Research.

Hwayoung Edward Lee, Lead, AI Biz Transformation Unit, LG AI Research

This work represents a shift in how AI is being applied to materials science and chemistry research. Rather than building general-purpose AI models and hoping they work for specialized domains, companies are now investing in domain-specific AI systems engineered from the ground up to understand the unique visual and conceptual language of chemistry. The partnership also follows Elsevier's Responsible AI Principles and Privacy Principles, reflecting growing attention to ethical AI deployment in scientific research.

For the broader chemistry community, the implications are clear: the knowledge trapped in images across millions of scientific documents is becoming accessible, searchable, and actionable. That shift could accelerate the pace of chemical discovery and reduce the time researchers spend on redundant work.