Elsevier, a global leader in advanced information and decision support in science and healthcare, has introduced Reaxys AI Search, an innovative addition to the Reaxys platform that leverages AI-driven natural language processing to transform chemistry research. Reaxys is the first chemistry database to introduce natural language document search to enhance the discoverability of relevant documents as researchers navigate vast amounts of complex chemical research information. Mirit Eldor, Managing Director, Life Sciences, Elsevier, discusses
How does Reaxys AI Search transform chemistry research compared to traditional keyword searches?
Chemists in both pharmaceutical and chemical R&D are expected to accelerate innovation at a faster pace than ever before, reduce development costs and set higher standards for sustainability and competitiveness; and to do this while also driving greater levels of productivity. They need trusted data and technology to help them deliver on these heightened expectations.
Elsevier’s Reaxys is a comprehensive chemistry database that delivers critical insights from over a billion data points combined with innovative technology. Chemists can easily explore chemical, bioactivity and toxicity data from literature and patents and accelerate synthesis planning, competitive analysis and the Design-Make-Test-Analyse (DMTA) cycle.
We recognised that keyword search will always be less than optimal when trying to surface every single relevant data point from that volume of information. Finding the relevant insight can be like finding a needle in a haystack – and with the extensive data in Reaxys, it is a pretty large haystack. Our goal was to help users find answers and insights even when there is no clear search filter or pre-defined field.
Reaxys AI search fills this gap. Built on advanced Machine Learning models trained specifically on chemistry texts, it understands scientific terminology, abbreviations and synonyms to deliver accurate, comprehensive and context-relevant results. This reduces the time researchers spend refining searches and increases the likelihood of discovering relevant insights. In testing, Reaxys AI Search achieved high relevancy and accuracy scores, delivering a substantial improvement over traditional keyword searching.
By eliminating the need to construct complex keyword searches, Reaxys AI Search is especially impactful for those working in interdisciplinary fields such as materials science, drug discovery, chemical engineering and polymer science.
What makes Reaxys the first chemistry database of its kind?
Reaxys is the first chemistry database to introduce AI-driven natural language document search specifically tailored for chemistry. Reaxys holds a billion chemistry data points from chemical patents and literature, and Reaxys AI Search understands the context of this data and knows exactly where to look for answers.
Reaxys’ goal is to accelerate R&D in the chemistry discipline and make chemical information accessible to users with varying levels of expertise, improving quality and speed of getting to insights, increasing confidence in decision-making, improving productivity and positively impacting a business’ ROI.
How does Reaxys AI Search interpret user queries differently from traditional search methods?
Reaxys AI Search uses natural language processing that allows it to interpret user intent and handle spelling variations, abbreviations and synonyms. It then applies the natural language search over an immense vectorised database to find the best matches. Needless to say, that’s a huge step up from traditional lexical search techniques that typically only return results that exactly match keywords.
How does Reaxys AI Search save researchers time?
Aside from reducing the amount of time needed to build complex search strings, Reaxys AI can be especially powerful in hit-to-lead and lead optimisation, where fast access to prior knowledge and reaction data can reduce time spent searching and planning and improve decision-making. Instead of manually searching through articles and patents, researchers can ask a question like ‘What small molecules inhibit XYZ pathway’ and instantly retrieve results, including bioactivity data and synthetic pathways, all in one place.
What previous AI-powered feature did Elsevier introduce before this?
For Reaxys, we have previously launched Reaxys Predictive Retrosynthesis, a tool that accelerates chemical synthesis projects by instantly revealing published and AI-predicted synthesis routes using Reaxys data and commercial availability insights.
Across Elsevier, we see AI as a way to improve how we support customers, and we have launched a number of AI tools in the last few months, including Embase AI and PharmaPendium AI.
What additional AI capabilities are expected in future releases of Reaxys?
We are moving towards Reaxys becoming a fully conversational, chat-based interface. That includes developing advanced summarisation capabilities, plus discovery tools for exploring answers in more detail and asking follow-up questions. This will enable even more intuitive exploration of chemical data.
We’re also continuing to solicit feedback from chemists using Reaxys AI search to ensure that we deliver what users need most. Suggestions for better integration with filters and summaries are already being prioritised for future releases.
Which guiding principles influenced the development of Reaxys AI Search?
Reaxys AI Search was developed through testing with hundreds of chemists, and in accordance with Elsevier’s Responsible AI Principles. As with all product development at Elsevier, our Privacy Principles were also central.
In practice, these principles mean that Reaxys AI aims to enhance human decision-making, not replace it. User interactions are private, and no data is used to train external models. Results are generated solely from trusted, curated content within Reaxys’ extensive database, ensuring confidentiality and security.
How does this innovation aim to lower barriers for researchers?
Natural language search is vital if we want to lower barriers for researchers. The traditional way of searching Reaxys requires users to build complex keyword strings, which in turn required advanced knowledge of query syntax or controlled vocabularies. Now, users at all expertise levels can ask questions in a conversational way and receive accurate answers.
