Which GWAS SNP Database to Use, and How to Join It to Your Own Genome
If you want one GWAS SNP database, use the NHGRI-EBI GWAS Catalog. It is the curated record of published genome-wide association studies, with one row per SNP-trait association, mapped to Experimental Factor Ontology (EFO) trait terms and to Ensembl genome coordinates 1. Download the full association table from the EBI FTP site, normalize your own variant calls to the same build and allele representation, and join on chromosome and position rather than on rsID. Everything else on this page is about why that join is harder than it looks and what the other databases add.
What the main databases contain
The GWAS Catalog is a hand-curated extraction from the literature. Each row carries the following fields.
- the reported SNP
- the mapped chromosome and position
- the risk or effect allele
- the p-value
- an odds ratio or beta
- the study accession (a
GCSTidentifier)
The file to start with is gwas-catalog-associations_ontology-annotated.tsv under ftp.ebi.ac.uk/pub/databases/gwas/releases/latest/. It is a few hundred thousand rows and loads into pandas or DuckDB in seconds.
The Catalog also hosts full summary statistics for a large fraction of studies, which is a different and much larger resource. Those live under /pub/databases/gwas/summary_statistics/, bucketed by accession range, and many have a harmonised/ subdirectory. Harmonized files have been oriented to the reference genome and given consistent column names, which saves you the worst part of the work. Expect tens of millions of rows per study and plan on tabix indexing rather than loading into memory.
Three other resources are worth knowing. PheGenI comes from NCBI and joins GWAS results to dbSNP, dbGaP, Gene, and OMIM. It is useful when you want the genomic context around a hit rather than the association alone 2. GWASdb collected both genome-wide significant and moderately significant associations and annotated them with regulatory and functional predictions, which makes it a reasonable place to look for candidate mechanisms 3. GWASdb’s second release added trait ontology mapping and sub-significant variants across thousands of studies 4. For variants that fall in microRNA genes or their target sites, miRNASNP integrates GWAS associations with miRNA expression and binding predictions 5.
For interactive work, Open Targets Genetics and GWAS ATLAS are better front ends than the raw Catalog, because they precompute fine-mapping and gene assignment. We still recommend pulling the Catalog TSV for anything you intend to reproduce, because you control the version.
Getting your own variants into joinable form
Start from a VCF or gVCF from whole-genome sequencing. Array genotypes work too, but a genome gives you the low-frequency variants that arrays miss entirely, and sequencing-based association analysis has measurably better power and resolution than imputed array data 6.
Three normalization steps do most of the work:
# 1. split multiallelics and left-align against the reference
bcftools norm -m -any -f GRCh38.fa -Oz -o norm.vcf.gz sample.vcf.gz
# 2. attach current dbSNP rsIDs by position and allele
bcftools annotate -a dbsnp156.vcf.gz -c ID -Oz -o rsid.vcf.gz norm.vcf.gz
# 3. index
bcftools index -t rsid.vcf.gz
If your calls are on GRCh37 and the Catalog release is on GRCh38, lift over before joining. Use CrossMap or picard LiftoverVcf with the UCSC chain file, and count the rejects. A few percent loss is normal, and the losses are not random: segmental duplications and inverted regions fail more often. Check the Catalog release notes for the build in use rather than assuming.
Join on position and allele, and not on rsID
rsIDs are unstable. They get merged, split, and retired between dbSNP builds. Published papers cite whatever ID existed at the time. A join on the SNPS column alone will silently drop real matches and occasionally create false ones.
The join key we use is (chrom, pos, sorted(ref, alt)) on GRCh38. Then reconcile the effect allele separately. The Catalog encodes it in STRONGEST SNP-RISK ALLELE, formatted as rs1801133-T. Parse the suffix and compare it to your REF and ALT. A suffix of ? means the risk allele was never reported, and those rows cannot be scored at all.
The failure mode that ruins more analyses than any other is strand ambiguity. For A/T and C/G variants you cannot tell from the alleles alone whether the source study used the forward or reverse strand. Either drop them or resolve them by comparing allele frequency to a reference panel, and only when the minor allele frequency is far from 0.5. We drop them by default and note how many were dropped.
What counts as a hit
The convention is p < 5 × 10⁻⁸, which is a Bonferroni correction for roughly one million independent common variants across the European genome. That threshold was designed for common SNPs on arrays. As sequencing pushes into low-frequency and rare variants, the effective number of independent tests rises, and the appropriate threshold becomes more stringent 7. Treat 5 × 10⁻⁸ as a floor for common variants and expect rarer variants to need more evidence.
Also read the effect size before the p-value. Most Catalog entries for complex traits report odds ratios between 1.05 and 1.25. A variant can be associated with overwhelming statistical confidence and still shift your risk by an amount you would never notice. The modern era of GWAS produced thousands of such reproducible small effects rather than a handful of large ones 8.
A separate caution: the GWAS Catalog is a research resource. It is not a clinical variant database. Pathogenic single-gene findings belong to ClinVar and to clinical laboratory reports. If something in your data looks medically consequential, take it to a clinician or genetic counselor for confirmatory testing.
Beyond single SNPs
Individual Catalog rows are a poor summary of polygenic traits. Two better uses of the same data:
Polygenic scores. Pull a published scoring file from the PGS Catalog and compute it with plink2 --score file.txt 1 2 3 header cols=+scsums. Use --q-score-range if you want to see how the score behaves as you relax the p-value cutoff. Scores derived in one ancestry group transfer poorly to others, and the absolute value of a score is meaningless without a matched reference distribution.
Gene and pathway aggregation. Map associated variants to genes, then test whether gene sets are enriched. MAGMA and similar tools handle the linkage disequilibrium structure that naive gene-set tests get wrong 9. This is where WGS data pays off, because you can aggregate rare coding variants within a gene instead of relying on a single tagging SNP.
Questions people also ask
Is the GWAS Catalog free to download in bulk?
Yes. The association TSV, the study metadata, and the summary statistics are all on the public EBI FTP site with no registration. The full association file is small enough to keep in version control alongside your analysis code.
Why does my rsID appear in a paper but not in the Catalog file?
The Catalog curates associations that meet its inclusion criteria, mainly genome-wide significance in a study with genome-wide coverage. Sub-threshold associations are excluded. GWASdb deliberately included moderately significant variants, so it is the better place to look for those 4.
What is the difference between the association file and summary statistics?
The association file has one row per curated significant hit. Summary statistics have one row per tested variant in the study, significant or not. That is what you need for polygenic scoring, fine-mapping, or genetic correlation.
How do I know which trait an EFO term refers to?
Resolve the MAPPED_TRAIT_URI against the EFO ontology through the EBI Ontology Lookup Service. Trait labels in free text are inconsistent across papers, so the ontology term is the reliable key for grouping studies 1.
Footnotes
-
Danielle Welter, Jacqueline MacArthur, Joannella Morales, et al. The NHGRI GWAS Catalog, a curated resource of SNP-trait associations. Nucleic Acids Research, 2013. https://doi.org/10.1093/nar/gkt1229 ↩ ↩2
-
Erin M Ramos, Douglas Hoffman, Heather A Junkins, et al. Phenotype–Genotype Integrator (PheGenI): synthesizing genome-wide association study (GWAS) data with existing genomic resources. European Journal of Human Genetics, 2013. https://doi.org/10.1038/ejhg.2013.96 ↩
-
Mulin Jun Li, Zipeng Liu, Panwen Wang, et al. GWASdb v2: an update database for human genetic variants identified by genome-wide association studies. Nucleic Acids Research, 2015. https://doi.org/10.1093/nar/gkv1317 ↩
-
Mulin Jun Li, Panwen Wang, Xiaorong Liu, et al. GWASdb: a database for human genetic variants identified by genome-wide association studies. Nucleic Acids Research, 2011. https://doi.org/10.1093/nar/gkr1182 ↩ ↩2
-
Jing Gong, Chunjie Liu, Wei Liu, et al. An update of miRNASNP database for better SNP selection by GWAS data, miRNA expression and online tools. Database, 2015. https://doi.org/10.1093/database/bav029 ↩
-
Julia Höglund, Nima Rafati, Mathias Rask-Andersen, et al. Improved power and precision with whole genome sequencing data in genome-wide association studies of inflammatory biomarkers. Scientific Reports, 2019. https://doi.org/10.1038/s41598-019-53111-7 ↩
-
João Fadista, Alisa K Manning, Jose C Florez, et al. The (in)famous GWAS P-value threshold revisited and updated for low-frequency variants. European Journal of Human Genetics, 2016. https://doi.org/10.1038/ejhg.2015.269 ↩
-
Shiro Ikegawa. A Short History of the Genome-Wide Association Study: Where We Were and Where We Are Going. Genomics & Informatics, 2012. https://doi.org/10.5808/gi.2012.10.4.220 ↩
-
Marquitta J. White, Brian L. Yaspan, Olivia J. Veatch, et al. Strategies for Pathway Analysis Using GWAS and WGS Data. Current Protocols in Human Genetics, 2018. https://doi.org/10.1002/cphg.79 ↩
Pothos builds longitudinal molecular profiles of individuals: whole-genome sequencing, single-cell RNA sequencing, and plasma proteomics, delivered as a detailed report, every data file, and stored cells. Build your profile.