BioChatter: Open Source Framework Simplifies Application of LLMs in Biomedical Research

In recent years, large language models (LLMs) have been increasingly widely used in various fields, from content creation to programming assistance to search engine optimization, all of which have demonstrated their powerful capabilities. However, in biomedical research, the application of these models still faces challenges in terms of transparency, repeatability and customization.

In response to this issue, the University of Heidelberg and the European Institute of Bioinformatics (EMBL-EBI) jointly proposed an open source Python framework, BioChatter, aiming to help biomedical researchers use LLMs more easily.

BioChatter’s design philosophy is to simplify technical complexity, allowing researchers to focus on their research without having to worry about the expertise in programming or machine learning. Through this framework, researchers can extract relevant data from biomedical databases and literature and enable real-time information access with external bioinformatics tools. This is all thanks to the seamless integration of BioChatter and BioCypher knowledge graphs, which are able to link important data such as gene mutations and drug-disease associations, greatly supporting the analysis of complex datasets.

The core functions of BioChatter include: basic Q&A interaction with various large language models, reproducible prompt engineering, knowledge graph query, search enhancement generation, model chain call, etc. More humane, BioChatter provides an intuitive API interface that researchers can easily integrate into web applications, command-line interfaces, or Jupyter notebooks.

During the experimental evaluation, the research team created customized benchmarks designed to evaluate BioChatter's performance more accurately. The results show that the model using BioChatter is significantly better than the model without the prompt engine in generating correct queries, and this discovery provides strong support for the practical application of BioChatter.

Going forward, the BioChatter team will continue to work with life science databases such as Open Targets, aiming to help users more efficiently identify and prioritize drug targets by integrating human genetics and genomics data. In addition, they are developing a complementary system called BioGather, which aims to extract information from other clinical data types such as genomics, medical notes and images to solve complex problems in personalized medicine and drug development.

Through BioChatter, scientists in the field of biomedical research will be able to utilize LLMs more efficiently, thereby promoting progress and innovation in scientific research.