Evidence-Grounded Pathogen Knowledge Extraction
A reusable literature-curation service that extracts pathogen-related claims and maps them to BRC entities and schemas with passage-level provenance.
Goal
Develop a reusable literature-curation service that extracts pathogen-related claims and maps them to BRC entities and schemas with passage-level provenance.
Three-Day MVP
Process 200–500 open-access papers and extract three relation types, such as pathogen–host phenotype, pathogen–mechanism–disease, and biomarker–disease. Produce structured JSON-LD or RDF, BV-BRC/NCBI identifiers, confidence scores, and a “cite or refuse” validation step. COMIC/InfectioVision and glycan-biomarker extraction can serve as the first two plug-in use cases.
Model and Evaluation
Calibrate or fine-tune an extraction model on a small manually reviewed training set. Evaluate entity linking, relation precision/recall, citation correctness, and unsupported-claim rate. Incorporate GDB-Lit-style benchmark tasks.
Team
Team assignments are being finalized ahead of the codeathon. Participants can review project teams, and request reassignment, in the shared participant spreadsheet circulated by the organizing team.