OntoCodex GPT built with this framework as a showcase: https://chatgpt.com/g/g-6734e1cd43008190b7245d756c8e75ef-ontocodex

OntoCodex is a multi-agent, LLM-enabled framework for semi-automated ontology enrichment. It integrates large language models, biomedical ontologies, curated knowledge bases, and standard terminologies to support scalable, reproducible, and human-in-the-loop ontology development for real-world biomedical and clinical research.

This repository provides the core framework, agent design, and example workflows for enriching OWL/RDF ontologies with new concepts, relationships, provenance, and standardized codes (e.g., ICD, SNOMED CT, RxNorm, LOINC).

• Multi-agent architecture for ontology enrichment
  • Decision Agent
  • Ontology Reading Agent
  • Knowledge Base Agent
  • Terminology Agent
  • Script Generation Agent
• Ontology-aware parsing of user-provided OWL/TTL ontologies
• Automated concept extraction from trusted biomedical sources
• Terminology normalization to standard vocabularies (ICD, SNOMED CT, RxNorm, ATC, LOINC)
• Provenance annotation and confidence scoring
• Executable Python code generation for OWL-compliant ontology updates
• Human-in-the-loop design for expert review and validation
• Extensible and modular, suitable for new domains and data sources

Ontology enrichment is essential for semantic interoperability but remains labor-intensive and error-prone. OntoCodex addresses this challenge by:

• Reducing manual curation burden
• Improving consistency of concept definitions and mappings
• Supporting downstream analytics using EHR, claims, and clinical trial data
• Enabling rapid iteration and reuse across biomedical domains

OntoCodex follows a coordinated, agent-based workflow:

1. Decision Agent
   Interprets enrichment goals and orchestrates agent interactions.

2. Ontology Reader Agent
   Parses OWL/TTL ontologies to identify existing classes, properties, and gaps.

3. Knowledge Base Agent
   Retrieves candidate concepts and relationships from curated sources such as:

   • MedlinePlus (API / cached TTL)
   • PubMed
   • NIH / CDC / ADA / ACC / AHA resources
   • Disease Ontology (DOID)

4. Terminology Agent
   Normalizes extracted concepts to standard vocabularies using CSV, OWL, or TTL datasets.

5. Script Generation Agent
   Generates executable Python scripts to update ontologies with new classes, axioms, annotations, and provenance.

• ICD-9 / ICD-10
• SNOMED CT
• RxNorm
• ATC
• LOINC
• DOID
• Custom CSV / OWL / TTL vocabularies

OntoCodex supports:

• Source attribution for extracted concepts
• Timestamped annotations
• Confidence scores (LLM-estimated or heuristic-based)
• Reproducible enrichment scripts

This design supports auditability and regulatory-aligned research workflows.

• Multiple Chronic Conditions (MCC)
• Alzheimer’s Disease / ADRD
• Opioid and substance use research
• Clinical trial Common Data Elements (CDEs)
• EHR and claims-based phenotyping

• Outputs require expert review before production use
• LLM-generated content may reflect source bias or incompleteness
• Not intended for direct clinical decision-making

This project is released under the MIT License. See LICENSE for details.

Contributions are welcome!

1. Fork the repository
2. Create a feature branch
3. Add tests and documentation
4. Submit a pull request

If you use OntoCodex in your research, please cite:

Feng J, et al. OntoCodex: A Multi-Agent Biomedical Ontology Enrichment Framework. (Manuscript under review).

Tao AI Group
Mayo Clinic, AI&Informatics

For questions or collaboration inquiries, please open an issue or contact the maintainers via GitHub.

OntoCodex is a research framework designed to augment—not replace—domain expertise.

MIT © 2025 Jingna Feng