Synonym Augmentation for Rare Disease Identification in Unstructured Data.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Rare Disease Research

Background:

• Information scarcity and unstructured formats pose challenges in rare disease research.
• Existing methods need improvement for identifying relevant rare disease information.

Purpose of the Study:

• To develop effective methods for identifying rare disease information from text.
• To advance rare disease research by systematically providing relevant data.

Main Methods:

• Utilized a fine-tuned BioMedBERT encoder to assess text relevance using confidence scores and semantic similarity.
• Fine-tuned the encoder with data from Online Mendelian Inheritance in Man (OMIM), Orphanet, and STS benchmark datasets.
• Employed a Neo4j knowledge graph to host data on rare diseases and NIH-funded projects.

Main Results:

• Successfully identified mentions of rare diseases in relevant texts.
• Demonstrated the approach's effectiveness through case studies retrieving NIH-funded projects.
• Created a knowledge graph containing data on 2,067 GARD diseases and over 320,000 NIH-funded projects.

Conclusions:

• The developed approach systematically provides rare disease-related data.
• This method enhances understanding of rare diseases for future investigations.
• Effective information retrieval is crucial for advancing rare disease research.