Additive chr19 cholesterol companion

Open chr19-cholesterol.vitessce.json in Vitessce. Original chr19.json and all its inputs remain unchanged.
New runtime inputs: chr19.map.embedding.csv, chr19.links.embedding.csv, chr19.sets.obsSets.json.
All original CSV files are reused at their existing URLs.

The landscape is UMAP (cosine distance on L2-normalized rE2G profiles, seed 0), not expression or evidence of causal regulation. Map x is midpoint in Mb; map y is log2(1+ClinVar overlaps) times the factor recorded in chr19.extras.provenance.json. Coordinates have a common display scale.

Gene selections use full matrix feature IDs. LDLR, APOE and TM6SF2 are present. CYP2A6 is absent; the supplied CYP2A6–2B6 locus contains zero elements. Its 2 Mb neighborhood contains 97 elements. The dot plot therefore uses three available genes. No synthetic gene columns or elements were added. Initial locus selection covers 35 elements. Correspondence with Gosling is by supplied windows/colors, not cross-application synchronization.

Rebuild extras in a NEW output directory:
  mkdir -p /tmp/vitessce-numba-cache /tmp/chr19-extras-rebuild
  NUMBA_CACHE_DIR=/tmp/vitessce-numba-cache python make_vitessce_extras.py --matrix chr19.matrix.csv --clinvar chr19.clinvar.metadata.csv --out /tmp/chr19-extras-rebuild
Requires numpy; umap-learn is optional, with explicitly recorded PCA fallback if unavailable. Output replacement requires --overwrite. The separate config builder reads adjacent original chr19.json and extras files and refuses to overwrite an existing companion config.

Validation: real-data row IDs, finite embedding coordinates, map formula, independent reconstruction of every set, config references/scopes and source preservation checked. Browser preview uses Vitessce 4.0.10 with real local CSV bytes intercepted at the configured URLs because outbound HTTPS to users.abdenlab.org times out in this environment. This does not validate the public HTTP delivery layer.