Gwas Catalog

Integrating NHGRI-EBI GWAS Catalog with ENCODE Regulatory Data

When to Use

• User wants to intersect ENCODE regulatory elements with GWAS-associated variants
• User asks about "GWAS", "genome-wide association", "disease variants", or "trait-associated SNPs"
• User needs to find which GWAS hits overlap enhancers, promoters, or TF binding sites
• User wants to prioritize GWAS loci by functional annotation from ENCODE data
• Example queries: "find GWAS variants in my H3K27ac peaks", "which diabetes GWAS hits overlap pancreas enhancers?", "annotate GWAS loci with ENCODE regulatory marks"

Connect genome-wide association study findings with ENCODE functional annotations to identify which regulatory elements harbor disease-associated variants and prioritize causal mechanisms for non-coding GWAS hits.

Scientific Rationale

The question: "Which of the disease-associated variants from GWAS fall within active regulatory elements, and what can ENCODE tell us about their functional impact?"

The GWAS Catalog (maintained by NHGRI-EBI) contains over 500,000 variant-trait associations from 6,000+ publications. The central challenge of post-GWAS analysis is that >90% of these associations point to non-coding regions of the genome. ENCODE provides the essential functional annotation layer: if a GWAS variant falls within an active enhancer in disease-relevant tissue, that enhancer becomes a candidate causal mechanism.

This was first demonstrated systematically by Maurano et al. (2012, Science), who showed that disease-associated variants are enriched in DNase I hypersensitive sites (DHSs), and that the cell-type specificity of the DHS predicts the relevant disease tissue. This foundational insight drives the entire GWAS-ENCODE integration framework.

Scale of the Problem

• GWAS Catalog: 500,000+ associations, 100,000+ unique variants, 5,000+ traits
• ENCODE cCREs: 926,535 regulatory elements covering 7.9% of the genome
• Overlap expectation: ~8% of random variants would overlap a cCRE by chance
• Observed enrichment: GWAS variants show 2-5x enrichment in regulatory elements (higher for tissue-matched elements)

Key Literature

• Sollis et al. 2023 "The NHGRI-EBI GWAS Catalog: knowledgebase and deposition resource" (Nucleic Acids Research). The current GWAS Catalog publication describing the REST API, summary statistics hosting, and expanded annotation pipeline. DOI: 10.1093/nar/gkac1010
• Buniello et al. 2019 "The NHGRI-EBI GWAS Catalog of published genome-wide association studies, targeted arrays and summary statistics 2019" (Nucleic Acids Research, ~3,500 citations). The widely-cited GWAS Catalog reference describing curation standards and the move to EFO ontology for traits. DOI: 10.1093/nar/gky1120
• Maurano et al. 2012 "Systematic localization of common disease-associated variation in regulatory DNA" (Science, ~3,000 citations). The foundational demonstration that GWAS variants concentrate in DNase I hypersensitive sites, with cell-type-specific enrichment predicting disease-relevant tissues. Enabled de novo identification of pathogenic cell types from variant sets. DOI: 10.1126/science.1222794
• ENCODE Project Consortium 2020 (Nature, ~1,656 citations). Registry of 926,535 human cCREs that provides the regulatory annotation layer for GWAS variant interpretation. DOI: 10.1038/s41586-020-2493-4
• Finucane et al. 2015 (Nature Genetics, ~2,253 citations). Stratified LD Score Regression (S-LDSC) for partitioning heritability into ENCODE-defined functional categories. DOI: 10.1038/ng.3404
• Nasser et al. 2021 (Nature, ~468 citations). ABC model linked 5,036 GWAS signals to 2,249 genes using ENCODE data. DOI: 10.1038/s41586-021-03446-x

GWAS Catalog REST API Reference

Base URL: https://www.ebi.ac.uk/gwas/rest/api

No authentication required. Responses are JSON (HAL format).

Key Endpoints

Pagination

All list endpoints support pagination:

• ?page=0&size=20 (default page size is 20, max is 500)

Bulk Downloads

For genome-wide analysis, use the GWAS Catalog downloads (faster than API):

• All associations: https://www.ebi.ac.uk/gwas/api/search/downloads/full
• Alternative: https://www.ebi.ac.uk/gwas/docs/file-downloads
• Format: TSV with columns for variant, trait, p-value, OR/beta, study, etc.

Step 1: Define the Disease/Trait and Relevant Tissues

Query GWAS Catalog for a Trait

import requests

# Search by trait name
trait = "type 2 diabetes"
url = "https://www.ebi.ac.uk/gwas/rest/api/studies"
params = {"diseaseTrait": trait}
response = requests.get(url, params=params)
studies = response.json()["_embedded"]["studies"]

print(f"Found {len(studies)} GWAS studies for '{trait}'")
for study in studies[:5]:
    print(f"  {study['accessionId']}: {study['publicationInfo']['title'][:80]}...")

Use EFO IDs for Standardized Trait Queries

The Experimental Factor Ontology (EFO) standardizes trait names:

# Query by EFO ID (more precise)
efo_id = "EFO_0001360"
url = f"https://www.ebi.ac.uk/gwas/rest/api/efoTraits/{efo_id}/associations"
params = {"size": 500}
response = requests.get(url, params=params)

Map Trait to ENCODE Tissues

Following the Maurano 2012 framework — disease-associated variants are enriched in tissue-specific regulatory elements:

# Check ENCODE data availability for disease-relevant tissue
encode_get_facets(organ="pancreas")
encode_get_facets(organ="liver")

Step 2: Retrieve GWAS Variants

Get Associations for a Variant

def get_gwas_associations(rs_id):
    """Get all GWAS associations for a variant."""
    url = f"https://www.ebi.ac.uk/gwas/rest/api/singleNucleotidePolymorphisms/{rs_id}/associations"
    response = requests.get(url)
    if response.status_code == 200:
        return response.json()["_embedded"]["associations"]
    return []

# Example: rs7903146 (strongest T2D variant, in TCF7L2)
associations = get_gwas_associations("rs7903146")
for assoc in associations:
    trait = assoc["efoTraits"][0]["trait"] if assoc["efoTraits"] else "Unknown"
    pval = assoc["pvalue"]
    print(f"  Trait: {trait}, p-value: {pval}")

Get All Variants for a Study

def get_study_associations(study_id):
    """Get all associations from a GWAS study."""
    url = f"https://www.ebi.ac.uk/gwas/rest/api/studies/{study_id}/associations"
    params = {"size": 500}
    response = requests.get(url, params=params)
    return response.json()["_embedded"]["associations"]

# Example: Mahajan et al. 2018 T2D GWAS
associations = get_study_associations("GCST006867")

Extract Variant Coordinates

GWAS Catalog provides variant locations. Extract for BED format:

def associations_to_bed(associations):
    """Convert GWAS Catalog associations to BED format lines."""
    bed_lines = []
    for assoc in associations:
        for locus in assoc.get("loci", []):
            for gene in locus.get("strongestRiskAlleles", []):
                rs_id = gene.get("riskAlleleName", "").split("-")[0]
        # Get location from the SNP endpoint
        snp_url = f"https://www.ebi.ac.uk/gwas/rest/api/singleNucleotidePolymorphisms/{rs_id}"
        snp_resp = requests.get(snp_url)
        if snp_resp.status_code == 200:
            snp = snp_resp.json()
            for loc in snp.get("locations", []):
                chrom = f"chr{loc['chromosomeName']}"
                pos = int(loc['chromosomePosition'])
                bed_lines.append(f"{chrom}\t{pos-1}\t{pos}\t{rs_id}")
    return bed_lines

Step 3: LD Expansion

Critical: GWAS lead SNPs are NOT necessarily causal. The causal variant is often a different SNP in linkage disequilibrium (LD).

When to Expand

LD Expansion Tools

• LDlink (NIH): https://ldlink.nih.gov/?tab=ldproxy — Web and API
  • API: https://ldlink.nih.gov/LDlinkRest/ldproxy?var={rsId}&pop={population}&r2_d=r2&token={token}
  • Requires free API token
• 1000 Genomes: Reference panel for LD calculation
• PLINK: plink --ld-snp {rsId} --ld-window-r2 0.8

Population-Specific LD

Always use the appropriate ancestry population:

Step 4: Intersect GWAS Variants with ENCODE Peaks

Retrieve ENCODE Peak Files

# Chromatin accessibility
encode_search_experiments(assay_title="ATAC-seq", organ="pancreas", biosample_type="tissue")
encode_search_experiments(assay_title="DNase-seq", organ="pancreas", biosample_type="tissue")

# Active regulatory marks
encode_search_experiments(assay_title="Histone ChIP-seq", target="H3K27ac", organ="pancreas")
encode_search_experiments(assay_title="Histone ChIP-seq", target="H3K4me3", organ="pancreas")

# TF binding
encode_search_experiments(assay_title="TF ChIP-seq", organ="pancreas")

# Get IDR thresholded peak files (GRCh38)
encode_list_files(
    experiment_accession="ENCSR...",
    file_format="bed",
    output_type="IDR thresholded peaks",
    assembly="GRCh38",
    preferred_default=True
)

bedtools Intersection

# Create BED file from GWAS variants (with LD proxies)
# gwas_variants_ld.bed: chr start end rsId pvalue trait

# Intersect with ENCODE peaks
bedtools intersect \
    -a gwas_variants_ld.bed \
    -b encode_h3k27ac_peaks.bed \
    -wa -wb \
    > gwas_in_enhancers.bed

bedtools intersect \
    -a gwas_variants_ld.bed \
    -b encode_atac_peaks.bed \
    -wa -wb \
    > gwas_in_accessible.bed

bedtools intersect \
    -a gwas_variants_ld.bed \
    -b encode_ctcf_peaks.bed \
    -wa -wb \
    > gwas_in_ctcf.bed

Enrichment Calculation

# Simple enrichment: observed vs. expected overlap
total_variants = 1000
variants_in_peaks = 150
genome_fraction_in_peaks = 0.02  # 2% of genome covered by peaks

expected = total_variants * genome_fraction_in_peaks
enrichment = variants_in_peaks / expected
print(f"Enrichment: {enrichment:.1f}x (observed={variants_in_peaks}, expected={expected:.0f})")

For rigorous enrichment testing, use S-LDSC (Finucane et al. 2015) or GARFIELD (Iotchkova et al. 2019) as described in the variant-annotation skill.

Step 5: Classify Regulatory Impact

For each GWAS variant in an ENCODE peak, classify the regulatory mechanism:

The Maurano 2012 Framework

1. Identify the DHS/ATAC peak — confirms the region is accessible in disease tissue
2. Classify the element type — H3K27ac (enhancer/promoter), H3K4me3 (promoter), CTCF (insulator)
3. Determine tissue specificity — tissue-specific elements are more likely causal
4. Link to target gene — Hi-C, ABC model, eQTL colocalization

Regulatory Impact Table

Step 6: Variant-to-Gene Linking

The nearest gene to a GWAS variant is the correct target only ~50-60% of the time. Use ENCODE data for better gene assignment:

ABC Model (Nasser et al. 2021)

Activity-By-Contact predictions link enhancers to genes using ENCODE ATAC-seq + H3K27ac + Hi-C:

• Pre-computed for 131 cell types
• Download from: https://www.engreitzlab.org/abc/
• Linked 5,036 GWAS signals to 2,249 target genes

Hi-C Contact Maps

encode_search_experiments(assay_title="Hi-C", organ="pancreas")

Check whether the variant-containing enhancer physically contacts a gene promoter.

eQTL Colocalization

Cross-reference with GTEx eQTLs to identify the gene regulated by the GWAS variant. See gtex-expression skill for API details.

Step 7: Log Provenance

encode_track_experiment(accession="ENCSR...", notes="GWAS-ENCODE overlay for T2D")

encode_log_derived_file(
    file_path="/path/to/gwas_encode_intersection.tsv",
    source_accessions=["ENCSR...", "ENCSR...", "ENCSR..."],
    description="Intersection of T2D GWAS variants (Mahajan 2018, LD r2>0.8 EUR) with ENCODE H3K27ac and ATAC-seq peaks in pancreas",
    file_type="variant_annotation",
    tool_used="bedtools intersect + GWAS Catalog REST API + LDlink",
    parameters="GRCh38, genome-wide significant (p<5e-8), IDR thresholded peaks, EUR LD r2>0.8"
)

encode_link_reference(
    experiment_accession="ENCSR...",
    reference_type="pmid",
    reference_id="30297969",
    description="Mahajan et al. 2018 T2D GWAS providing variant set"
)

Pitfalls and Caveats

1. Lead SNP Is Not Causal Variant

The GWAS Catalog reports lead SNPs (strongest statistical signal per locus). The causal variant is often a different SNP in LD. Always expand to LD proxies (r2 >= 0.8) before intersecting with ENCODE peaks. Fine-mapped credible sets (SuSiE, FINEMAP) bypass this issue.

2. P-Value Thresholds

Genome-wide significance is p < 5 x 10^-8. Suggestive significance (p < 1 x 10^-5) may be included in some GWAS Catalog entries. For regulatory annotation, focus on genome-wide significant variants unless the user specifically requests suggestive hits. Sub-threshold variants have higher false-positive rates.

3. Population Specificity

LD patterns differ substantially by ancestry. A lead SNP in a European GWAS may not tag the same LD block in African or East Asian populations. Always use population-matched LD reference panels. Trans-ethnic fine-mapping can narrow credible sets by exploiting LD differences.

4. Trait Ontology Mapping

The GWAS Catalog uses EFO (Experimental Factor Ontology) for trait standardization. Free-text trait names in the API may not match exactly. Use EFO IDs for precise queries. The mapping between disease traits and ENCODE tissue/biosample terms requires manual curation.

5. Assembly Consistency

The GWAS Catalog provides coordinates in GRCh38. Older GWAS studies may report hg19/GRCh37 positions. ENCODE peaks should be in GRCh38. Always verify assembly match. Use UCSC liftOver or CrossMap for coordinate conversion if needed. NEVER mix assemblies.

Walkthrough: Connecting GWAS Variants to ENCODE Regulatory Elements

Goal: Identify GWAS-significant variants that fall within ENCODE-defined regulatory elements to prioritize causal non-coding variants for a disease of interest. Context: Type 2 diabetes has hundreds of GWAS hits, most in non-coding regions. ENCODE maps which of these overlap active regulatory elements.

Step 1: Find ENCODE regulatory data for pancreas

encode_search_experiments(assay_title="ATAC-seq", organ="pancreas", organism="Homo sapiens")

Expected output:

{
  "total": 6,
  "results": [
    {"accession": "ENCSR456PAN", "assay_title": "ATAC-seq", "biosample_summary": "pancreas", "status": "released"},
    {"accession": "ENCSR457ISL", "assay_title": "ATAC-seq", "biosample_summary": "islet of Langerhans", "status": "released"}
  ]
}

Interpretation: Both whole pancreas and islet-specific ATAC-seq available. Islet data is more relevant for T2D (beta cell disease).

Step 2: Download regulatory peak files

encode_list_files(accession="ENCSR457ISL", file_format="bed", output_type="IDR thresholded peaks", assembly="GRCh38")

Expected output:

{
  "files": [
    {"accession": "ENCFF200ISL", "output_type": "IDR thresholded peaks", "file_format": "bed narrowPeak", "file_size_mb": 0.9}
  ]
}

Step 3: Query GWAS Catalog for T2D associations

Using GWAS Catalog REST API (via skill guidance):

GET https://www.ebi.ac.uk/gwas/rest/api/associations?efoTrait=EFO_0001360&pvalueFilter=5e-8

Expected key fields per association:

{
  "riskFrequency": "0.30",
  "pvalue": 2.4e-12,
  "snps": [{"rsId": "rs7903146", "chromosomeName": "10", "chromosomePosition": 114758349}],
  "efoTraits": [{"trait": "type 2 diabetes mellitus"}]
}

Step 4: Intersect GWAS variants with ENCODE peaks

bedtools intersect -a t2d_gwas_variants.bed -b ENCFF200ISL.bed -wa -wb > t2d_regulatory_variants.bed

Interpretation: GWAS variants overlapping islet ATAC-seq peaks are strong candidates for causal regulatory variants. Variants in islet-specific (but not other tissue) peaks suggest tissue-restricted regulatory mechanisms for T2D.

Step 5: Prioritize candidates with multi-evidence support

Rank variants by evidence layers:

1. GWAS significant (p < 5×10⁻⁸) ✓
2. In islet open chromatin (ATAC-seq peak) ✓
3. Near islet-expressed gene (GTEx) — check via gtex-expression
4. In H3K27ac-marked enhancer — check via histone-aggregation
5. ClinVar annotation — check via clinvar-annotation

Integration with downstream skills

• ENCODE peaks from → regulatory-elements define regions to intersect with GWAS hits
• Overlapping variants feed into → variant-annotation for functional prediction
• Population frequencies from → gnomad-variants contextualize GWAS allele frequencies
• Expression effects checked via → gtex-expression eQTL data
• Clinical significance from → clinvar-annotation adds clinical interpretation

Code Examples

1. Survey ENCODE data for GWAS-relevant tissues

encode_get_facets(facet_field="organ", assay_title="ATAC-seq", organism="Homo sapiens")

Expected output:

{
  "facets": {
    "organ": {"brain": 32, "heart": 18, "liver": 14, "pancreas": 6, "blood": 25}
  }
}

2. Find histone mark data to layer on GWAS regions

encode_search_experiments(assay_title="Histone ChIP-seq", organ="pancreas", target="H3K27ac", organism="Homo sapiens")

Expected output:

{
  "total": 4,
  "results": [
    {"accession": "ENCSR300ACE", "assay_title": "Histone ChIP-seq", "target": "H3K27ac", "biosample_summary": "islet of Langerhans"}
  ]
}

3. Track experiments used for GWAS annotation

encode_track_experiment(accession="ENCSR457ISL", notes="Islet ATAC-seq for T2D GWAS variant annotation")

Expected output:

{
  "status": "tracked",
  "accession": "ENCSR457ISL",
  "notes": "Islet ATAC-seq for T2D GWAS variant annotation"
}

Integration

Presenting Results

Summary Table

Key Numbers to Report

• Total GWAS variants analyzed (lead + LD proxies)
• Fraction overlapping ENCODE regulatory elements (by element type)
• Enrichment over background (observed/expected)
• Number of tissue-specific elements harboring variants
• Variant-to-gene links established and method used
• Diseases/traits represented

Related Skills

• variant-annotation — Full ENCODE variant annotation and prioritization workflow
• clinvar-annotation — Clinical significance of variants in ENCODE peaks
• gnomad-variants — Population frequency for GWAS variants
• disease-research — Disease-focused ENCODE analysis workflows
• regulatory-elements — Characterizing ENCODE regulatory elements at GWAS loci
• gtex-expression — eQTL colocalization and expression context for GWAS genes
• publication-trust — Verify literature claims backing analytical decisions

For the request: "$ARGUMENTS"