医学公开数据库查询与登记

references/gnomad.md

Overview

gnomAD aggregates exome and genome sequencing data to provide allele frequencies and variant annotations across diverse populations.

API Type: GraphQL

• Endpoint: https://gnomad.broadinstitute.org/api
• Method: POST with JSON body containing GraphQL query
• Auth: None required (public, unauthenticated)
• Response format: JSON (data wrapper with GraphQL structure)

Key Queries

Variant lookup by variant ID

Variant IDs use format: {chrom}-{pos}-{ref}-{alt} with 1-based positions in the build of the selected dataset. An rsID can identify multiple alleles; resolve the exact reference/alternate allele before querying. The example below is the GRCh38 G>A allele at rs11591147, not the more common G>T allele.

POST https://gnomad.broadinstitute.org/api
Content-Type: application/json

{
  "query": "{ variant(variantId: \"1-55039974-G-A\", dataset: gnomad_r4) { variant_id rsids chrom pos ref alt exome { ac an af } genome { ac an af } } }"
}

Gene lookup

{
  "query": "{ gene(gene_symbol: \"BRCA1\", reference_genome: GRCh38) { gene_id symbol chrom start stop strand } }"
}

Variants in a gene

{
  "query": "{ gene(gene_symbol: \"PCSK9\", reference_genome: GRCh38) { variants(dataset: gnomad_r4) { variant_id consequence rsids exome { ac an af } genome { ac an af } } } }"
}

Variants in a region

{
  "query": "{ region(chrom: \"1\", start: 55039447, stop: 55064852, reference_genome: GRCh38) { variants(dataset: gnomad_r4) { variant_id rsids consequence exome { ac af } genome { ac af } } } }"
}

Transcript lookup

{
  "query": "{ transcript(transcript_id: \"ENST00000357654\", reference_genome: GRCh38) { transcript_id gene_id chrom start stop strand } }"
}

Dataset values

• gnomad_r4 -- gnomAD v4 (GRCh38, latest major release)
• gnomad_r3 -- gnomAD v3.1.2 (GRCh38, genomes only)
• gnomad_r2_1 -- gnomAD v2.1.1 (GRCh37, exomes + genomes)

Population frequency fields

Within exome or genome objects, population-specific frequencies are available via populations { id ac an } (compute AF as AC/AN only when AN is positive). Ancestry-group IDs and availability vary by dataset; read the returned IDs rather than applying a fixed cross-release list.

Response example (illustrative counts, not a current frequency estimate)

{
  "data": {
    "variant": {
      "variant_id": "1-55039974-G-A",
      "rsids": ["rs11591147"],
      "chrom": "1",
      "pos": 55039974,
      "ref": "G",
      "alt": "A",
      "exome": { "ac": 1234, "an": 250000, "af": 0.004936 },
      "genome": { "ac": 456, "an": 150000, "af": 0.00304 }
    }
  }
}

Rate Limits

• No published rate limits, but aggressive querying will be throttled
• Use reasonable request pacing (~1 req/sec recommended)
• For bulk downloads, use gnomAD's Hail tables on Google Cloud or download VCFs

Notes

• The GraphQL schema is not versioned separately; it tracks the gnomAD web interface
• Use the browser's network inspector on gnomad.broadinstitute.org to discover additional query fields and structures
• Structural variants (SV) have a separate query structure (structural_variant)
• Constraint metrics (pLI, LOEUF) are available on gene queries via gnomad_constraint.
• A null exome/genome result is unavailable data, not zero frequency. Even AC=0 must be interpreted with AN, coverage, filters, dataset and ancestry. Population frequency alone does not classify clinical pathogenicity.