UNMIRI Publishes Open FHIR Genomics Schema and Variant Retrieval Research
UNMIRI、オープンFHIRゲノムスキーマとバリアント検索研究を発表
Summary
UNMIRI has published an open-source FHIR Genomics R4 schema for cross-vendor Next-Generation Sequencing (NGS) data. Additionally, the company released research on a failure mode in how clinical AI tools retrieve genomic evidence, highlighting issues with standard similarity comparisons.
Details
UNMIRI has released two key resources: an open-source FHIR Genomics R4 schema and a research preprint detailing flaws in genomic evidence retrieval by AI. The new FHIR schema normalizes reports from major labs (e.g., Foundation Medicine, Tempus) into a structured format, including HGVS variant conventions and biomarker fields. The accompanying research highlights that relying on cosine similarity for comparing text embeddings can dangerously conflate clinically distinct cancer variants, treating them as nearly identical. UNMIRI argues that identity must not be treated as a similarity problem at all. Instead, the platform resolves identity using a typed knowledge graph, matching variants based on standardized identifiers before any retrieval step. This approach ensures accuracy, which is critical because different variants in the same gene can require vastly different drugs. These tools provide foundational infrastructure for precision oncology applications, including genomics-aware clinical decision support and variant-grounded trial matching.
Original content copyright by respective publishers