Summary statistics

This page includes general documentation about the summary statistics we host in the GWAS Catalog. For instructions on submitting summary statistics please see our submission documentation.

---

General

• What are summary statistics?

• Sources

• Eligibility

• Standard format

Access

• Searching our web interface

• Navigating the FTP

• Directories

  • Harmonised subdirectories

• How can I find all the harmonised data?

Formatted and harmonised files

• What are “formatted files”?

• What are "harmonised files"?

  • Harmonisation codes

Accompanying metadata

• -meta.yaml info

  • File types

---

What are summary statistics?

While associations included within the Catalog are the curated, significant (p<1e-5) and independent loci reported within a paper, summary statistics (sumstats) are the full, aggregate p-values and association data for every variant analysed in a study. These are hosted on our servers when available and can be downloaded via links on study and publication pages. For more summary statistics FAQ, please see here.

Sources

Summary statistics are acquired from a number of different sources:

• Submitted to the GWAS Catalog by an author

• Identified via link or reference in a curated paper

• Identified via a consortium or author website

• Included in a curated paper as a supplementary file

We encourage authors to submit summary statistics to the GWAS Catalog. Summary statistics for published and pre-published/unpublished GWAS can be submitted directly via the Submission page. Instructions on how to submit summary statistics can be found here.

Eligibility

Summary statistics must meet a number of criteria to be eligible for inclusion in the GWAS Catalog:

• Must include >100,000 SNPs

• Must reproduce a GWAS described in a published paper (ie. same sample metadata and trait), or is clearly marked as pre/un-published.

• Must be freely available, i.e. not only by request to author or password-protected.

• Must not require agreement to any specific terms and conditions, unless this has been agreed in advance by the author/consortium. Since 2022, we have required newly submitted sumstats to be released under CC0 - noted under ‘Terms/Licence’ on study pages. In rare cases, special licensing terms will appear in an adjacent readme.txt.

Standard format

The vast majority of files adhere to one of two standard formats, GWAS-SSF and pre-GWAS-SSF.

Since April 2023, all summary statistics will meet GWAS-SSF - the product of a series of working groups, which has been written up as a preprint. The specification of the standard can be found here with further details in the GitHub repository. Prior to V1 of this standard, a minimum acceptable format (pre-GWAS-SSF) was used, the details of which are available in Buniello and MacArthur, 2019.

Standardised data files are additionally provided in a harmonised format, with all data reported against the same genome build and alleles aligned to the forward strand. Each file has an accompanying metadata file in .yaml format, which indicates the file_type and whether or not the file has been harmonised.

We also host a small number of non-standard files with eg. fewer than 100,000 SNPs or non-standard study design (eg. CNV or gene-based). These will be flagged as ‘Non-GWAS-SSF’ in the .yaml.

Moving beyond SNPs

Since April 2026, copy number variant (CNV) and gene-based GWAS data should follow the standards described in the beyond SNPs documentation. Previously, these data were submitted to the Catalog without a standard schema, which limited downstream reuse.

Access

We store summary statistics files on our FTP site, which is updated nightly with any newly ingested data. API access to summary statistics has been deprecated, and an improved API is coming soon.

Searching via our web interface:

Links to FTP directories are provided within the main Catalog pages. Search for a specific trait, accession ID, PMID, author, etc., and look out for the summary statistics icon, denoting availability.

Per publication page, all available sumstats will be listed under the 'Full summary statistics' tab.

You can also browse our summary statistics tables for all pre- and post-publication data here: https://www.ebi.ac.uk/gwas/downloads/summary-statistics

Navigating the FTP:

Every summary statistics file is available at a path based on the GCST accession ID. For example, the summary statistics for accession GCST123456 would be available at 'ftp.ebi.ac.uk/pub/databases/gwas/summary_statistics/GCST123001-GCST124000/GCST123456/'. You can see that the directory 'GCST123456/' is nested inside 'ftp.ebi.ac.uk/pub/databases/gwas/summary_statistics/GCST123001-GCST124000/'. This parent directory holds up to 1000 GCSTs, and will only contain those GCSTs that are within the range given. These bins of one thousand are added as necessary, when new GCSTs need housing.

Directories:

Each accession ID directory contains:

• The original sumstats file .tsv

• A metadata .yaml

• A separate readme, if provided .txt (newer folders will hold this under ‘author_notes’ inside the .yaml)

• When possible, and once processed, a harmonised directory

Harmonised subdirectories:

The latest version of the harmonisation pipeline was released in 2023. All files harmonised under this version, eg. GCST90270927 contain a ‘/harmonised’ directory with the following contents:

• A harmonised version of the raw file. .h.tsv (see What are harmonised files?)

• An index file .tbi

• A log file .log

• A metadata file .yaml

For files harmonised before 2023, eg GCST90001686, the harmonised folder will contain:

• A formatted version of the raw file - with standardised headers .f.tsv (see What are "formatted files"?)

• A harmonised version of the formatted file .h.tsv

• A metadata file .yaml

In some cases, there will be no harmonised directory due to incompatibility with our pipeline. These will be marked out with ‘Non-GWAS-SSF’ under file_type in the .yaml. If the file type is GWAS-SSF or pre-GWAS-SSF and there is no harmonised file, it is likely awaiting harmonisation, please check back at a later date.

How can I find all the harmonised data?

Harmonised summary statistics are released daily and a text file listing all of the harmonised studies is available on the FTP at the following location: https://ftp.ebi.ac.uk/pub/databases/gwas/summary_statistics/harmonised_list.txt. This text file is updated on a nightly basis.

What are “formatted files”?

Prior to the GWAS Catalog accepting direct submissions, summary statistics files were formatted by scientific curators using a semi-automated process. Where the formatted files are available, they are nested within the 'harmonised' subdirectory with the ‘*.f.tsv.gz’ suffix (example). These files will not be available for newer submitted content because those datasets have already been formatted by the authors. During the formatting process data were not manipulated but the headers were converted to a consistent format across all summary statistics files.

What are "harmonised files"?

Sumstats files are harmonised using an automated pipeline, which is detailed below. By 'harmonised' we mean that all genomic position data will be reported against the same and latest genome build. Missing chromosome, base pair location and variant ID data that can be inferred, will be. Furthermore, where possible, alleles are also orientated to the forward strand.

The harmonisation process is the following:

1. Mapping variant IDs to locations (method used is reported in the ‘hm_coordinate_conversion’ field in the harmonised data file)

   1. Update base pair location value by mapping rsID using Ensembl (currently v95) (‘hm_coordinate_conversion’ = ‘rs’); or

   2. if above not possible, liftover base pair location to latest genome build (‘hm_coordinate_conversion’ = ‘lo’); or

   3. if above not possible, remove variant from file.

2. Orientation (Open Targets project)

   1. Infer the orientation of palindromic variants. Palindromic variants are alleles on the forward strand which are the same as on the reverse strand. We infer the orientations of palindromic variants according to the strand consensus.

      1. Firstly, we randomly select 10% of sites. The effect and other alleles are compared with counterpart alternative and reference alleles in the Ensembl VCF references to identify the strand of the non-palindromic variants (forward or reverse) .

      2. The forward strand consensus can be calculated by forward/(forward+reverse) or reverse/(forward+reverse). To avoid any possibility of sampling bias:

         • If the rate ≥ 0.995, the following harmonisation steps on the palindromic variants are inferred as on the forward (or reverse) strand;

         • If the rate is the range of (0.995,0.9), this rate is recalculated by all non-palindromic variants in the data. The palindromic variants can be inferred as forward (or reverse) if the recalculated rate > 0.99, otherwise palindromic variants are dropped for harmonisation;

         • If the rate ≤ 0.9, palindromic variants are dropped in the following harmonisation step.

   2. Variant harmonisation: Using chromosome, base pair location and the effect and other alleles, query each variant against the Ensembl VCF reference to harmonise as appropriate by either:

      1. keeping record as is because:

         • it is already correctly orientated

      2. orientating to reference strand:

         • reverse complement the effect and other alleles

      3. flipping the effect and other alleles

         • because the effect and other alleles are flipped in the reference

         • this also means the beta, odds ratio, 95% CI and effect allele frequency are inverted

      4. a combination of the orientating and flipping the alleles.

      5. replace with NA because:

         • There is no counterpart record in the reference VCF file.

3. Filtering and QC

   1. Variant ID is set to variant IDs found by step (2).

   2. Records without a valid value for variant ID, chromosome, base pair location and p-value are removed.

4. Harmonisation result: For each input file, harmonisation generates three outputs:

   1. The harmonisation result (*.h.tsv.gz)

      • represents harmonised mandatory columns in a specific order and other columns in the original order.

      • contains a new column named harmonisation code is assigned to each record indicating the harmonisation process that was performed (note that currently any processes involving 'Infer strand' are not being used).

      • is sorted by chr and position and compressed using bgzip

   2. An index file (*.h.tsv.gz.tbi)

      • A tabix index file of the harmonisation result for quick data retrieve purposes

   3. A report file (report.txt) summarises:

      • the reference VCF file used in harmonisation.

      • process on the palindromic variants

      • percentage of variants dropped in the qc step

      • percentage of variants that are successfully harmonised or failed

   4. A metadata YAML file (*.h.tsv.gz-meta.yaml) for the harmonised data file.

Harmonisation codes:

Code	Description of process
1	Palindromic; Infer strand; Forward strand; Alleles correct
2	Palindromic; Infer strand; Forward strand; Flipped alleles
3	Palindromic; Infer strand; Reverse strand; Alleles correct
4	Palindromic; Infer strand; Reverse strand; Flipped alleles
5	Palindromic; Assume forward strand; Alleles correct
6	Palindromic; Assume forward strand; Flipped alleles
7	Palindromic; Assume reverse strand; Alleles correct
8	Palindromic; Assume reverse strand; Flipped alleles
9	Palindromic; Drop palindromic; Not orientated
10	Forward strand; Alleles correct
11	Forward strand; Flipped alleles
12	Reverse strand; Alleles correct
13	Reverse strand; Flipped alleles
14	Required fields are not known; Not orientated
15	No matching variants in reference VCF; Not orientated
16	Multiple matching variants in reference VCF; Not orientated
17	Palindromic; Infer strand; EAF or reference VCF AF not known; Not orientated
18	Palindromic; Infer strand; EAF < specified minor allele frequency threshold; Not orientated

Harmonised data files conform to the standard but there are some additional points to mention:

1. Where possible, data are sorted by chromosome and base pair location

2. Where possible, a <data_file>.tbi file is available to enable access to a specified genomic region using tabix (link)

3. A running report from the harmonisation is available with the harmonised file on the FTP

4. A field in the data file, ‘‘hm_coordinate_conversion’’, signifies whether the position was mapped using rsID (‘rs’) lookup or liftOver (‘lo’).

There may be differences between the data available on the FTP site compared to the database where number rounding has taken place. In the database, chromosome values of 'X', 'Y' and 'MT' will appear as 23, 24 and 25, respectively.

More extensive documentation on the harmonisation pipeline can be found here.

Accompanying metadata (-meta.yaml)

Metadata are provided in a computer and human readable YAML file adjacent to the data file, suffixed with -meta.yaml. So where the data file is called GCST123456.tsv.gz, the metadata can be found with the name GCST123456.tsv.gz-meta.yaml.

Metadata definitions:

Field	Description	Data type and values	Mandatory	Example
--------------------------	--------------------------	--------------------------	--------------------------	--------------------------
# Study meta-data
gwas_id	GWAS Catalog accession ID	Text string	Yes	GCST90244057
author_notes	Additional information about this study from the author	Text string	No	File contains GWAS summary statistics from a meta-analysis of NMR metabolic traits in up to 33 cohorts.
gwas_catalog_api	GWAS catalog REST API link	Text string	Yes	https://www.ebi.ac.uk/gwas/rest/api/studies/GCST90244057
date_metadata_last_modified	The latest date that metadata YAML file was modified	date	Yes	2023-11-28
# Trait Information
trait_description	Author reported trait description	Text string (multiple possible)	Yes	Body mass index
ontology_mapping	Short form ontology terms describing the trait	Text string (multiple possible)	No	EFO_0004918
# Genotyping Information
genome_assembly	Genome assembly for the summary statistics.	GRCh/NCBI/UCSC value	Yes	GRCh37
coordinate_system	Coordinate system used for the summary statistics	Text String (1-based or 0-based)	No	1-based
genotyping_technology	Method(s) used to genotype variants in the discovery stage.	Text string (multiple possible)	Yes	Genome-wide genotyping array
imputation_panel	Panel used for imputation	Text string	No	HRC + UK10K
imputation_software	Software used for imputation	Text string	No	SHAPEIT3 + IMPUTE4
# Sample Information
sample_ancestry_category	Broad ancestry category that best describes the sample.	Text string	Yes	European
sample_ancestry	The most detailed ancestry descriptor(s) for the sample.	Text string (multiple possible)	Yes	Finnish
sample_size	Sample size	Integer	Yes	27006
ancestry_method	Method used to determine sample ancestry e.g. self-reported/genetically determined	Text string (multiple possible)	No	self-reported
case_control_study	Flag whether the study is a case-control study	Boolean	No (default is false)	true
case_count	Number of cases for case/control study	Integer	No, unless caseControlStudy is true	27006
control_count	Number of controls for case/control study	Integer	No, unless caseControlStudy is true	27006
sex	To indicate a sex-stratified analysis	M (for male), F (for female), combined or NR if unknown	No	combined
# Summary Statistic information
data_file_name	The name of the summary statistics file	Text string	Yes	GCST90244057_buildGRCh37.tsv
file_type	The format of the summary statistics file	"GWAS-SSF v1.0", "pre-GWAS-SSF", "non-GWAS-SSF"	Yes	GWAS-SSF v1.0
data_file_md5sum	The md5 checksum of the summary statistics file.	Text string	Yes	0ec56396f89edcc21a3d5a25a6fa993d
analysis_software	Software and version used for the association analysis	Text string (multiple possible)	Yes if p-values of 0 given	REGENIE
adjusted_covariates	Any covariates the GWAS is adjusted for	Text string (multiple possible)	No	sex
minor_allele_freq_lower_limit	Lowest possible effect allele frequency	Numeric	No	0.0003
# Harmonization status
is_harmonised	Description of harmonisation codes	Text string	Only given in harmonised datasets	false
is_sorted	Flag whether the file is sorted by genomic location	Boolean	Yes	false
harmonisation_reference	The genome reference file used for harmonising the summary statistics file	Text string	No	https://ftp.ensembl.org/pub/release-104/fasta/homo_sapiens/dna/

File_types:

• GWAS-SSF v1: Version 1 of the GWAS-SSF standard format.

• Pre-GWAS-SSF: The previous, minimal format requiring only chromosome, base_pair_location/rsID and p-value.

• Non-GWAS-SSF: Not conforming to any standard format. Can include gene-based, CNV and multivariate analyses where eg. effect information is not available.

Note: for very old sumstats (where the filename does not begin ‘GCST’), no metadata .yaml will be available.

For summary statistics associated with a journal publication, study metadata can also be accessed via the GWAS Catalog web interface and REST API. Metadata for unpublished submissions is not currently available via REST API but can be found in our download files.