ClinVar

ClinVar is a freely accessible NCBI archive that collects and reports relationships between human genetic variants and phenotypes, with supporting evidence. It is the de facto reference for pathogenicity classifications in clinical genomic medicine: clinical laboratories, expert panels, and individual submitters deposit variant classifications, and downstream consumers — diagnostic labs, variant-interpretation pipelines, computational predictors — anchor their evidence chains to ClinVar entries.

Classification Tiers

ClinVar submissions are categorized along the ACMG five-tier system (Pathogenic, Likely Pathogenic, VUS, Likely Benign, Benign), with conflicting classifications surfaced explicitly when submitters disagree. A clinical variant interpretation pipeline reading ClinVar will encounter four practical states:

1. Reviewed by an expert panel — strongest evidence; rare across the variant landscape
2. Multiple submitters, no conflicts — strong evidence
3. Multiple submitters, conflicting interpretations — flag for re-interpretation
4. Single submitter / no assertion criteria — weak evidence; supplementary only

Live Status and Drift

ClinVar classifications change over time as new evidence accumulates. A variant classified pathogenic in 2018 may be reclassified VUS in 2023 if functional evidence emerges that the 2018 classification rested on weak segregation data. Any analysis or marketing artifact that cites a ClinVar classification carries a freshness obligation — the cited classification reflects ClinVar at the date of access, not necessarily ClinVar at the date of consumption. Best practice is to record the access date and to re-check before downstream clinical use.

ClinVar in Variant Interpretation Pipelines

ClinVar provides one of the strongest single inputs to ACMG criterion PS3 (functional evidence) and PP1 (segregation) when classifications carry expert-panel review. For computational predictors (REVEL, AlphaMissense, ESM-2), ClinVar pathogenicity labels are also the training data — ClinVar entries supervise the model, then the model’s predictions feed back as PP3/BP4 computational evidence. The recursive dependency is one reason unsupervised methods (ESM-2 specifically) are an interesting cross-reference: they do not consume ClinVar labels during training.

API Access and Programmatic Use

ClinVar exposes a public RESTful API (E-utilities) and bulk FTP downloads. Programmatic live-status checks against the API are standard for variant-interpretation pipelines; bulk-download snapshots are standard for benchmarking analyses where reproducibility against a fixed reference is more important than freshness.