Sources

Clinical significance: the aggregate classification on each axis ClinVar classifies on

Public domain, as US government work.

Read from ClinVar’s own index on every lookup, through esearch and esummary. esummary gives ClinVar’s aggregate classification and the number of submissions behind it rather than the submissions themselves, so the individual rows are a third call, efetch, made once the section is on screen. Every row links to its own submission record.

Allele frequency and its ancestry breakdown, filtering allele frequency, quality-control flags, in-silico predictor scores, transcript consequences, and gene constraint

Data released under CC0 by the Broad Institute.

Read from the gnomAD v4 GraphQL API on every lookup, so the number on the card is the number on the gnomAD page it links to. The in-silico predictors and the MANE Select designation are carried in v4 only, so a GRCh37 lookup gets neither and the card says so.

Placing an HGVS description on the genome, naming a gene’s MANE Select transcript, and deciding whether two sources describe the same variant

Public domain, as US government work.

Two services on one host. Variation Services places a transcript description on the genome and maps every equally valid placement of an insertion or deletion onto one canonical form, so two registries can be compared exactly rather than by proximity — it also reports when a source asserts reference bases the genome does not have, which is how a change numbered against the wrong transcript is caught. Datasets names the MANE Select transcript for a gene, which is what a gene-and-change input is placed on; the card says which transcript it used and that MANE named it. Neither is reached by an rsID or a coordinate lookup.

Literature

Metadata reusable under the terms Europe PMC publishes.

Searched by gene combined with a quoted group of variant aliases, sorted by citations.

Trait associations reported by published genome-wide association studies

Data released under the EMBL-EBI terms of use.

Searched by rsID only — the API has no coordinate or HGVS filter — using the v2 API, whose predecessor reached its announced end of life in May 2026. Every association links to the study record it came from.

Resolving notations this tool cannot parse — Ensembl transcript and LRG HGVS

Ensembl data is released without restriction; see their disclaimer.

Called only when local parsing fails, and split across two hosts by assembly. It returns a versioned RefSeq accession, which pins the build and is checked before use.

Resolving an input to one variant: position on both builds, alleles, rsID, gene and spellings

Public domain, as US government work.

Read from dbSNP’s own index. An rsID is dbSNP’s record number, so a lookup by rsID needs no search; a coordinate is searched by position and then matched on the exact bases, because a position is a search key and several variants share one. An rsID can name several alleles, and the result says which one it is showing and what else the identifier covers.

Indexed alongside: the MeSH subjects the papers naming the variant are filed under

Public domain, as US government work.

A count of publications, never a test. MEDLINE’s curators assign each paper subject headings from one fixed list, and this counts the conditions most often filed alongside the papers that name the variant — over a bounded sample of them, which the result states. Which headings are conditions is read from MeSH’s own categories, looked up for each one rather than kept here, so a heading naming a gene, a protein or a method is not counted. A condition can be common because a pairing has been studied often and not supported: published volume is not evidence of an association.

The in-silico predictor scores gnomAD carries, shown under the clinical significance

Each score is published by its own authors; gnomAD distributes them under CC0.

Computational predictions rather than observations, and never a classification. Each value is shown exactly as gnomAD sent it, under the name its publishers use, with a link to the paper or project page that says how to read it — the scales run in different directions, so a number without its source is not readable.

ACMG evidence codes: the criteria a rule can decide from the numbers already on the card, and the criteria that need a person

Published guidance, cited and linked. The thresholds are the publishers’ own and are quoted rather than restated.

The one section on the card that is not read from a registry. It adds no request and no host: every number it uses — the consequence, the ancestry frequencies, the predictor scores — is already on the card, and each code is a published threshold applied to one of them, with a link to the publication the threshold came from. It reports six of the guideline’s twenty-eight criteria and lists the other twenty-two with what each would take, and it never combines them: the combining rules assume every applicable criterion has been weighed, and whether a variant is pathogenic, benign or of uncertain significance is a determination for a qualified person.