SNOMED, or SNOMEDCT_US in the US edition, is a multilingual clinical terminology repository with more than 450,000 concepts and rich relationships such as Causative agent and Finding site. It supplies standardized biomedical identifiers, names, aliases, classifications, and graph connections for clinical information systems. section 7.2 section 8.2
SNOMED becomes especially useful when unstructured biomedical text must connect to structured knowledge. UMLS provides a cross-vocabulary entry point, while SNOMED contributes clinical concepts and relationships that can be integrated into a knowledge graph. section 7.5.2 section 7.2
A practical SNOMED ingestion and integration pipeline is:
- Load concept relationships from the SNOMED relationship file into Neo4j, enforcing unique identifiers and indexing names, relationship IDs, types, and UMLS identifiers. section 7.6.4
- Load descriptions to assign preferred names and aliases to concepts and relationship edges. section 7.6.4
- Represent hierarchical connections with SNOMED_IS_A and propagate first-level semantic categories through deeper nodes, allowing entities such as diseases and pharmaceutical products to inherit usable types. section 7.6.4
- Connect SNOMED entities to UMLS-linked MedicalEntity nodes, creating a unified graph that can also connect to HPO disease and phenotype nodes. section 7.6.4
Biomedical entity recognition
Retrieve propagated SNOMED categories and place them in an LLM prompt that restricts named-entity recognition to predefined medical types. For text about Zika, the resulting JSON can label Zika as an Organism and microcephaly and congenital malformations as Clinical finding (finding), with character offsets for each mention. section 8.4.1
