BioRAGent: natural language biomedical querying with retrieval-augmented multiagent systems.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Artificial Intelligence in Medicine

Background:

• Biomedical research requires understanding complex gene, phenotype, and disease relationships.
• Efficient retrieval of this interconnected data is a significant challenge.
• Existing methods struggle with the complexity and accessibility of biomedical knowledge.

Purpose of the Study:

• To introduce BioRAGent, an intelligent assistant for natural language querying of biomedical knowledge.
• To leverage retrieval-augmented generation (RAG) and a multiagent system for improved data access.
• To facilitate accurate and efficient retrieval of information on genes, phenotypes, diseases, and their interrelationships.

Main Methods:

• Developed BioRAGent, integrating Tool-augmented RAG with a multiagent system.
• Employed three specialized agents: Guide (query optimization), Retriever (data retrieval), and Reviewer (answer validation).
• Utilized authoritative biomedical databases for data retrieval and response generation.

Main Results:

• BioRAGent demonstrated superior performance on benchmark tasks compared to state-of-the-art models.
• Achieved high accuracy on eleven single-hop and three multi-hop query tasks.
• Ablation experiments confirmed the contribution of each agent to retrieval accuracy.

Conclusions:

• BioRAGent effectively addresses the challenge of biomedical knowledge retrieval through an intelligent multiagent system.
• The system provides a practical and robust user experience, especially for complex queries.
• BioRAGent enhances the accessibility and accuracy of biomedical information for researchers.