Skip to content

Known Issues

This page documents the current release status, data-coverage limitations, and transformation caveats of the ClinVar-GKM pipeline.

Release status

The current release is 1.1.0. It regroups the proposition sections by meaning and carries a set of data corrections, all delivered by one full rebuild of every published release. It is built on the finalized GKM (Genomic Knowledge Model) standards — VRS 2.1.1, Cat-VRS 1.1.1, VA-Spec 1.1.0, and GKM-Core 1.3.0 — and published under v1/1.1/; the 1.0.1 releases are kept, frozen, under v1/1.0/. See the Changelog for what 1.1.0 changed.

Data coverage

Each release includes 100% of the Germline and Somatic submission (SCV) records from the corresponding ClinVar XML release, and their classification content is carried through without modification to what ClinVar provides. However, the pipeline does not yet extract 100% of every attribute and tertiary content:

  • Case-level and functional data — aggregate case values, grouped/individual case-level evidence, and functional-data (incl. MaveDB) submissions are not currently extracted.
  • Rarely-provided attributes — some attributes that ClinVar provides only occasionally and sparsely may also not yet be represented.

If data that exists in ClinVar is missing here and would be useful to you, it can be added to the Roadmap and prioritized by community demand — open or upvote a discussion in the Ideas category. Case-level / functional data are already tracked as roadmap items (case data, functional data).

Delta reconstruction — VRS allele records (resolved in 1.1.0)

In 1.0.1, reconstructing a release's allele section from the prior release plus its weekly delta was inexact for about a dozen VRS allele records, most at the 2026-08-29 release, where the VRS line advanced, and the rest from non-deterministic ordering in allele assembly. Resolved in 1.1.0: alleles are built once per allele id with a total order, and the delta-reconstruction oracle finds no difference (0,0,0) on every section for all 16 delta pairs, including the cross-version pair 2026-08-22 → 2026-08-29. The frozen 1.0.1 deltas under v1/1.0/ keep the old behavior.

VRS normalization errors

VRS normalization does not resolve every variation. In the 2026-10-18 release, 6 of the 43,868 variations sent for normalization returned an error. The error text is in the vrsProcessingException extension on the variation's variation record. See Categorical Variant Extensions.

Malformed HGVS expressions in ClinVar

16 HGVS expressions on 14 variations in ClinVar's VCV XML are malformed at the source, most without an HGVS type letter (g., c.). From 1.1.0 each is kept verbatim as an extension of a listed expression, either merged onto ClinVar's valid expression for the same accession or on a repaired one; see HgvsListItem. They were reported to ClinVar on 2026-10-18 (NCBI case CAS-1713568-P1W1Q1), tracked in #163.

DOIs that VA-Spec's pattern rejects

VA-Spec's Document.doi pattern rejects some valid DOIs, such as SICI-style DOIs with parentheses (ga4gh/va-spec#461). Such a DOI is kept in the clinvarDoi extension of its Document, with its doi.org URL in urls, instead of in doi. See Citations.

RCV membership lists SCVs missing from the VCV data

ClinVar's RCV-format XML runs out of sync with its VCV XML: an RCV's SCV list sometimes names SCVs that are absent from the VCV data, 2–3 a week. This is ClinVar's data, faithfully parsed. RCV statements are built only from SCVs present in the VCV data, so an RCV's evidence can list fewer SCVs than ClinVar's RCV record.

Name-derived gene context genes are not in the gene section

An SCV proposition's gene context can be a gene named in the variant's title that ClinVar does not associate with the variant (for example NR_164162.1(GLRA4):n.1091C>T → GLRA4). The bundle's gene section lists only the genes ClinVar associates with a variant, so such a gene has no gene record, and its Parquet gene_context_id does not join to gene.parquet. In the 2026-10-11 release this is 25 variants and 31 SCV propositions. Genes with no HGNC id (LOC*) and NCBI-discontinued gene ids get no gene context, by design of the upstream method.

Past Parquet cannot yet be rebuilt

Only the latest Parquet set is published. Past Parquet is meant to be rebuilt from the monthly bundles with clinvar-gkm-parquet, which has no release yet.

Additional edge cases, transformation caveats, and specific limitations will be documented here as they are identified.