PubMind: literature-based genetic variant extraction and functional annotation using large language models.

PubMind extracts and annotates over 1.3 million genetic variants from biomedical literature with high accuracy, revealing many novel variants beyond existing databases.

• Why it matters: Current resources like ClinVar and HGMD are limited by incomplete coverage, bias, and sparse annotations, hindering comprehensive variant interpretation essential for precision medicine.
• What they did: Using large language models, PubMind triages and extracts variant-function-disease associations from over 41 million PubMed abstracts and 5 million full-text articles, normalizing data to genomic coordinates.
• The result: The database enables improved variant interpretation, with >80% of variants showing concordant pathogenicity labels, and offers a scalable resource to advance genomic research and clinical decision-making.