How it's built¶

Building the vocabulary graph is a three-step process: ingest, build, merge.

1. Ingest¶

A curator downloads a release (a ZIP file) and hands it to sssom-rosetta. sssom-rosetta checks the file is the one expected, then stores it under a version-specific folder. Nothing is guessed: every release keeps its own copy.

2. Build¶

Each vocabulary is turned into its own graph file: one for OMOP, one for the Diagnosethesaurus, one for the Verrichtingenthesaurus. These are built separately, so a problem in one release never affects the others.

Every relationship in OMOP is kept in the graph, not just the handful used for mapping, each with a readable label. For example, "has ingredient" stays "has ingredient", instead of being collapsed into a generic term. This means nothing gets silently dropped.

3. Merge¶

The three graphs are combined into one. Codes that already match across vocabularies (for example, a shared SNOMED CT code) get connected automatically.

The merge step uses maplib, a fast graph-processing tool. This matters because OMOP's release is huge, with tens of millions of connections, and older tools would take much longer to process it.

Building everything at once¶

By default, building the vocabulary graph includes OMOP and both DHD thesauri: they are the standard set, not optional extras. A single command runs all three steps and produces the combined graph.