mims-harvard/ClinVec

Generates 153,166 clinical code embeddings by training a graph transformer (HGT) on ClinGraph, a unified knowledge graph integrating 8 EHR vocabularies (ICD-9/10, CPT, RxNorm, LOINC, ATC, PheCode, SNOMED-CT, UMLS). Embeddings are available in multiple formats (DGL, PyTorch Geometric, NetworkX) across vocabulary-specific CSV files, enabling semantic retrieval and downstream tasks like risk scoring and medical QA without patient-level data dependencies.