tmVar: a text mining approach for extracting sequence variants in biomedical literature

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Automated text-mining of mutation data is crucial for understanding sequence variations in complex diseases.
• Existing methods often focus on limited mutation types (e.g., protein point mutations) and require extensive manual rule development.
• There is a need for accurate, automatic approaches to extract a wider range of mutation information.

Purpose of the Study:

• To develop and evaluate tmVar, a text-mining approach for extracting diverse sequence variants from biomedical literature.
• To cover important mutation types not previously addressed by existing text-mining tools.
• To improve the accuracy and scope of mutation extraction compared to state-of-the-art methods.

Main Methods:

• Utilized a conditional random field (CRF) model for text mining.
• Developed a novel CRF label model and feature set tailored for mutation extraction.
• Applied the approach to protein, DNA, and RNA sequence variants adhering to Human Genome Variation Society nomenclature.

Main Results:

• tmVar achieved high performance, outperforming a state-of-the-art method on both a custom corpus and a gold-standard dataset.
• F-measure scores for tmVar were 91.4% (custom corpus) and 93.9% (gold standard), compared to 78.1% and 89.4% respectively for the comparison method.
• The approach successfully extracts a wide range of mutation types, including those not covered by previous studies.

Conclusions:

• tmVar is a high-performance tool for extracting diverse mutation information from biomedical literature.
• The developed CRF model and feature set significantly enhance mutation extraction accuracy.
• This tool facilitates the analysis of sequence variations and supports the creation of mutation databases.