Using AI to improve how we extract information from biomedical texts

Research papers and other biomedical documents contain knowledge about how diseases work. This knowledge could help researchers discover new ideas for treatments and research questions.

However, because of the way papers are written, it is difficult for computers to automatically find, organise and analyse this information.

To analyse documents, researchers have traditionally had to:

• Read papers
• Extract important information
• Use text-analysis tools to help with the process

This was slow and time-consuming.

Large language models (LLMs), such as modern AI systems, have changed this. LLMs can now:

• Extract claims from text – for example: gene X increases the risk of disease Y
• Map those claims to ontologies by translating concepts into standard terms and relationships so we can compare them easily
• Normalise the information by converting information into a standard, consistent format
• Use it to build databases and knowledge graphs – knowledge graphs show relationships between diseases, genes, drugs and biological processes

Project aims

We will develop new approaches to extracting important information from biomedical texts such as research papers and other documents.

What we hope to achieve

We hope that by developing new methods of data extraction we can help us identify research questions and inform how new drugs are developed.

Our work also has the potential to convert unstructured medical notes into structured data that we can then add to standard ontologies and dictionaries.

This PhD project is being led by Freddie Yu with supervision provided by Yi Liu, Tom Gaunt, Zahraa Abdallah, Michael Tillich and Martin Baron.