CIViCmine

CIViCmine extracts clinically relevant cancer biomarkers and their clinical associations from PubMed and the PubMed Central Open Access subset using supervised text-mining to prioritize curatable evidence for precision oncology.

Key Features:

• Text-mining corpus: Processes PubMed abstracts and the PubMed Central Open Access subset to identify literature evidence.
• Supervised learning: Uses expert-annotated sentences as training data for automated extraction.
• Machine-learning model: Applies a trained model to identify sentences that describe biomarker–clinical associations.
• Extraction output: Extracted 121,589 pertinent sentences containing biomarker-clinical association evidence.
• Knowledgebase coverage: Contains over 87,412 biomarkers linked to 8,035 genes, 337 drugs, and 572 cancer types.
• Document provenance: Derives associations from 25,818 abstracts and 39,795 full-text publications.
• Annotation reliability: Training data were produced by cancer genomics experts with substantial inter-annotator agreement.
• CIViC integration: Provides prioritized, curatable biomarker evidence for integration with the Clinical Interpretation of Variants in Cancer (CIViC) knowledgebase.

Scientific Applications:

• Knowledgebase expansion: Prioritizes candidate biomarker evidence to expand CIViC and other variant interpretation resources.
• Clinical association discovery: Systematically identifies associations relevant to diagnosis, prognosis, predisposition, and drug response in cancer.
• Supporting precision oncology: Supplies curated evidence candidates for variant interpretation and biomarker-driven clinical decision support.

Methodology:

Expert annotation of sentences produced training data used in supervised learning to train a machine-learning text-mining model that extracts biomarker–clinical association sentences from PubMed and the PubMed Central Open Access subset.