OntoCodex: a multi-agent biomedical ontology enrichment framework

Area of Science:

• Biomedical Informatics
• Knowledge Representation
• Artificial Intelligence

Background:

• Ontology enrichment is crucial for semantic knowledge representation but is often labor-intensive.
• Existing methods may lack efficiency and scalability in handling complex biomedical data.

Purpose of the Study:

• To introduce OntoCodex, a novel multi-agent framework for semi-automated ontology enrichment.
• To evaluate OntoCodex's effectiveness in improving concept extraction and terminology mapping for chronic diseases.

Main Methods:

• OntoCodex integrates large language models (LLMs), ontologies, knowledge sources, and vocabularies.
• A five-agent system (parsing, decision-making, retrieval, normalization, script generation) supports OWL-based integration and a feedback loop.
• A ChatGPT-4o implementation was used to enrich concepts across five chronic diseases.

Main Results:

• OntoCodex outperformed baseline ChatGPT-4o in concept extraction precision, recall, and F1 scores for most domains.
• Perfect performance was achieved in laboratory test extraction; improved accuracy in medication terminology mapping was observed.
• Automatically generated Python scripts successfully enriched the MCC-CDO ontology without errors.

Conclusions:

• OntoCodex significantly enhances the efficiency and accuracy of ontology enrichment.
• The framework shows strong potential to accelerate clinical and translational research by improving semantic knowledge representation.