Leveraging Generative AI to Accelerate Biocuration of Medical Actions for Rare Disease

Area of Science:

• Computational biology
• Medical informatics
• Rare disease research

Background:

• Structured clinical data and ontologies are crucial for translational research and computational analysis.
• The Medical Action Ontology (MAxO) represents patient treatments and clinical management actions.
• Manual biocuration for MAxO is time-consuming and difficult to scale for over 10,000 rare diseases.

Purpose of the Study:

• To develop and evaluate AutoMAxO, a semi-automated workflow using Large Language Models (LLMs) to streamline MAxO biocuration for rare diseases.
• To improve the efficiency and scalability of creating structured representations of medical actions for rare disease clinical management.

Main Methods:

• AutoMAxO employs LLMs to extract candidate medical action curations from publication abstracts.
• Candidate curations are matched to ontology terms (MAxO, HPO, MONDO) using LLMs and post-processing.
• Matched terms are reviewed and approved by human curators.

Main Results:

• Processed 4,918 unique medical abstracts, identifying annotations for 21 rare genetic diseases.
• Extracted 18,631 candidate disease-treatment curations, with 538 confirmed and added to the MAxO dataset.
• Demonstrated the potential of generative AI to accelerate the curation of biomedical literature.

Conclusions:

• AutoMAxO significantly enhances the efficiency of MAxO biocuration for rare diseases.
• Generative AI offers a powerful approach to accelerate precision medicine through structured data representation.
• The AutoMAxO methodology can be adapted for other biomedical curation tasks.