NCBI disease corpus: a resource for disease name recognition and concept normalization

Area of Science:

• Biomedical Natural Language Processing
• Computational Linguistics
• Bioinformatics

Background:

• Biomedical literature contains crucial information, but accessing it requires advanced text mining tools.
• Developing effective tools for automatic disease concept detection relies on high-quality annotated data.
• Existing resources may lack the comprehensive annotation needed for robust natural language processing (NLP) model training.

Purpose of the Study:

• To present the NCBI disease corpus, a novel annotated dataset for biomedical NLP research.
• To provide a standardized resource for training and evaluating disease name recognition and concept normalization systems.
• To facilitate the development of machine learning-based approaches for biomedical information extraction.

Main Methods:

• Manual annotation of 793 PubMed abstracts by two independent annotators.
• Utilizing PubTator for pre-annotation and consensus-building through discussion phases.
• Mapping disease mentions to standardized concepts using Medical Subject Headings (MeSH) and Online Mendelian Inheritance in Man (OMIM).

Main Results:

• The NCBI disease corpus includes 6892 disease mentions mapped to 790 unique concepts.
• 88% of concepts link to MeSH identifiers, and 12% link to OMIM identifiers.
• High inter-annotator agreement was achieved, with 91% of mentions linked to a single concept.

Conclusions:

• The NCBI disease corpus serves as a high-quality gold standard for disease recognition and normalization research.
• This resource enables the development and benchmarking of advanced NLP tools for biomedical text mining.
• The corpus has the potential to significantly advance the state-of-the-art in automated disease information extraction.