ViMRT

ViMRT automates recognition and extraction of virus mutation mentions from biomedical literature for downstream analyses including vaccine design and drug usage.

Key Features:

• Natural language processing: Applies advanced NLP techniques for text mining of biomedical publications.
• Rule-based extraction: Implements 8 optimized rules and 12 regular expressions for mutation recognition.
• Training corpus: Rules and patterns were derived from an initial dataset of 830 papers covering five severe disease-associated viruses.
• High accuracy: Achieved an F1-score of 99.17% on a separate test dataset of 1662 papers.
• Cross-virus validation: Maintained an F1-score of 96.99% across 212 papers for influenza virus and SARS-CoV-2.
• Literature retrieval: Includes a search capability for retrieving mutation mentions, associated viral genes, and disease associations from texts.

Scientific Applications:

• Vaccine design: Provides extracted mutation data that can inform vaccine antigen selection and design considerations.
• Drug usage and resistance studies: Supplies mutation annotations relevant to drug efficacy and usage analyses.
• Disease progression research: Enables study of mutation roles in virus-driven disease progression by extracting mutation mentions from literature.
• Literature curation: Facilitates creation of curated mutation datasets and retrieval of mutation–gene–disease associations from publications.

Methodology:

Uses natural language processing combined with a rule-based extraction system employing 8 optimized rules and 12 regular expressions; rules were developed from 830 papers across five viruses and evaluated on datasets of 1662 and 212 papers yielding F1-scores of 99.17% and 96.99%, respectively.