NOTE v1.2 is draft and not yet implemented. Existing tools are working to v1.0
Specifying a format to store GWAS summary data is necessary to aid with data sharing and tool development. Using the VCF format can fulfil the following requirements
- It uses a pre-existing, well known and well defined format
- Aligning against the reference genome and handling various difficulties such as indels, build differences and multi-allelic variants has been solved by the htslib library.
- Many tools exist that can be used for manipulation
- The file format is relatively small
- Indexing makes looking up by chromosome and position extremely fast
- Indexing time is very fast
- We can treat each GWAS as a distinct unit rather than storing everything in a database which is less nimble
- We can store multiple GWAS datasets in a single file by using one sample column for each GWAS
- It is easy to export the data into other tabular formats
- Initial tests indicate it could translate directly to distributed databases that sit on top of vcf e.g https://github.com/GenomicsDB/GenomicsDB
- gwas2vcf - open source software to convert plain text GWAS summary statistics to VCF
- gwas2vcfweb - open source front/back-end for gwas2vcf running at http://vcf.mrcieu.ac.uk/
- bcftools - can be used to manipulate, align with genome references, query, as it is in standard vcf format
- R/gwasvcf - wrapper around bioconductor/VariantAnnotation package for performing natural GWAS queries in R. Includes LD proxy functionality
- pygwasvcftools - wrapper around pysam package for performing natural GWAS queries in python
- R/TwoSampleMR - can use GWAS vcf files directly for summary data based Mendelian randomization analysis
- ldsc - a fork of the LD score regression programme that allows reading in data directly from GWAS vcf format.
The VCF format specification is available from hts-specs.
A broad overview:
- the header describes the harmonisation and the trait(s)/study/studies in the file.
- each row of the body contains information about a variant
- variant information (e.g. chromosome, id, position, alleles, annotations) are represented in the first few columns
- the sample section (which would normally represent the genotype information of one column per individual) is used to store GWAS summary data. Each sample column represents one study/trait.
The VCF header defines fields found in the body including the trait field which contain information about the GWAS study.
An example is given below. It has the following main meta data sections:
- INFO - describes the annotations included for the variants
- FORMAT - describes the fields available for the GWAS summary data (e.g. ES = effect size etc)
- study - describes metadata about the study i.e. publication
- trait - describes metadata about the trait i.e. phenotype that is tested for association with the variant
- contig - descriptions of the chromosomes
- bcftools - list of commands used to create the file
##fileformat=VCFv4.2
##fileurl=https://github.com/MRCIEU/gwas-vcf-specification/blob/master/example.vcf.txt
##filedate="07/09/2020"
##gwasformat=GWAS-VCFv1.2
##source=Gwas2VCFv1.2.0
##reference=ftp://ftp.broadinstitute.org/bundle/b37/human_g1k_v37.fasta.gz
##contig=<ID=1,length=249250621,assembly=GRCh37.p13>
##contig=<ID=2,length=243199373,assembly=GRCh37.p13>
##contig=<ID=3,length=198022430,assembly=GRCh37.p13>
##contig=<ID=4,length=191154276,assembly=GRCh37.p13>
##contig=<ID=5,length=180915260,assembly=GRCh37.p13>
##contig=<ID=6,length=171115067,assembly=GRCh37.p13>
##contig=<ID=7,length=159138663,assembly=GRCh37.p13>
##contig=<ID=8,length=146364022,assembly=GRCh37.p13>
##contig=<ID=9,length=141213431,assembly=GRCh37.p13>
##contig=<ID=10,length=135534747,assembly=GRCh37.p13>
##contig=<ID=11,length=135006516,assembly=GRCh37.p13>
##contig=<ID=12,length=133851895,assembly=GRCh37.p13>
##contig=<ID=13,length=115169878,assembly=GRCh37.p13>
##contig=<ID=14,length=107349540,assembly=GRCh37.p13>
##contig=<ID=15,length=102531392,assembly=GRCh37.p13>
##contig=<ID=16,length=90354753,assembly=GRCh37.p13>
##contig=<ID=17,length=81195210,assembly=GRCh37.p13>
##contig=<ID=18,length=78077248,assembly=GRCh37.p13>
##contig=<ID=19,length=59128983,assembly=GRCh37.p13>
##contig=<ID=20,length=63025520,assembly=GRCh37.p13>
##contig=<ID=21,length=48129895,assembly=GRCh37.p13>
##contig=<ID=22,length=51304566,assembly=GRCh37.p13>
##contig=<ID=X,length=155270560,assembly=GRCh37.p13>
##contig=<ID=Y,length=59373566,assembly=GRCh37.p13>
##FILTER=<ID=PASS,Description="All filters passed">
##INFO=<ID=RSID,Number=1,Type=String,Description="dbSNP identifier",Source="https://ftp.ncbi.nih.gov/snp/latest_release/VCF/GCF_000001405.25.gz",Version="153">
##FORMAT=<ID=NS,Number=A,Type=Float,Description="Variant-specific number of samples/individuals with called genotypes used to test association with specified trait">
##FORMAT=<ID=EZ,Number=A,Type=Float,Description="Z-score provided if it was used to derive the ES and SE fields">
##FORMAT=<ID=SI,Number=A,Type=Float,Description="Accuracy score of summary association statistics imputation">
##FORMAT=<ID=NC,Number=A,Type=Float,Description="Variant-specific number of cases used to estimate genetic effect (binary traits only)">
##FORMAT=<ID=ES,Number=A,Type=Float,Description="Effect size estimate relative to the alternative allele">
##FORMAT=<ID=SE,Number=A,Type=Float,Description="Standard error of effect size estimate">
##FORMAT=<ID=LP,Number=A,Type=Float,Description="-log10 p-value for effect estimate">
##FORMAT=<ID=AF,Number=A,Type=Float,Description="Alternative allele frequency in trait subset">
##FORMAT=<ID=AC,Number=A,Type=Float,Description="Alternative allele count in the trait subset">
##study=<ID="PMID:12345678",Source="PubMed">
##trait=<ID=EFO0004340,Description="Body mass index",Source="EFO",Version="3.14.0",Type="continuous",Test="linear",Unit="SD",Population="European",TotalSamples=461460,TotalVariants=9851866,VariantsNotRead=0,HarmonisedVariants=9851866,VariantsNotHarmonised=0,SwitchedAlleles=9851866,FileUrl="https://gwas.mrcieu.ac.uk/files/ukb-b-19953/ukb-b-19953.vcf.gz",FileDate="24/04/2020">
##trait=<ID=EFO0001360,Description="type II diabetes mellitus",Source="EFO",Version="3.14.0",Type="binary",Test="logistic",Population="European",TotalSamples=462933,TotalCases=2972,TotalVariants=9851866,VariantsNotRead=0,HarmonisedVariants=9851866,VariantsNotHarmonised=0,SwitchedAlleles=9851866,FileUrl="https://gwas.mrcieu.ac.uk/files/ukb-b-13806/ukb-b-13806.vcf.gz",FileDate="24/04/2020">
| #CHROM | POS | ID | REF | ALT | QUAL | FILTER | INFO | FORMAT | EFO0004340 | EFO0001360 |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 49298 | . | T | C | . | PASS | RSID=rs10399793 | NS:NC:ES:SE:LP:AF:AC | 463005:0:0.00103892:0.0034984:0.113509:0.613764:568351 | 463005:2972:9.098e-05:0.000294716:0.119186:0.613764:568351 |
| 1 | 49298 | . | T | A | . | PASS | RSID=rs10399793 | NS:NC:ES:SE:LP:AF:AC | 463005:0:0.00214602:0.00346583:0.267606:0.011:4630 | 463005:2972:0.000102689:0.00029197:0.136677:0.012:4630 |
| 1 | 91536 | . | GTC | G | . | PASS | RSID=rs6702460 | NS:NC:ES:SE:LP:AF:AC | 463005:0:0.00410514:0.0034125:0.638272:0.456845:423042 | 463005:2972:0.000329732:0.000287485:0.60206:0.456851:423042 |
| 1 | 534192 | . | C | T | . | PASS | RSID=rs6680723 | NS:NC:ES:SE:LP:AF:AC | 463005:0:0.000334321:0.0038979:0.0315171:0.24094:223131 | 463005:2972:0.000106473:0.000328379:0.124939:0.24096:223131 |
| 1 | 706368 | . | A | AAA | . | PASS | RSID=rs12029736 | NS:NC:ES:SE:LP:AF:AC | 463005:0:-0.00030371:0.00241981:0.0457575:0.515705:477487 | 463005:2972:7.16085e-06:0.000203854:0.0132283:0.51565:477487 |
The first row represents a biallelic variant (rs10399793). The reference allele (T) is always the non-effect allele and must match the reference genome sequence. The alternative allele (C) is always the effect allele and often (but not always) the minor allele. The final column contains the effect size (ES), standard error (SE), P value on -log10 scale (LP), study allele frequency (AF) and sample size (NS). Some fields are optional others required, refer to the header and section 2 (below) for details.
| Field | Description | Required |
|---|---|---|
| NS | Variant-specific number of samples/individuals with called genotypes used to test association with specified trait | FALSE |
| EZ | Z-score provided if it was used to derive the ES and SE fields | FALSE |
| SI | Accuracy score of association statistics imputation | FALSE |
| NC | Variant-specific number of cases used to estimate genetic effect (binary traits only) | FALSE |
| ES | Effect size estimate relative to the alternative allele | TRUE |
| SE | Standard error of effect size estimate | TRUE |
| LP | -log10 p-value for effect estimate | TRUE |
| AF | Alternative allele frequency in trait subset | FALSE |
| AC | Alternative allele count in the trait subset | FALSE |
Genomic positions with more than one variant should be stored on individual rows as shown above (rs10399793)
Lyon, M.S., Andrews, S.J., Elsworth, B. et al. The variant call format provides efficient and robust storage of GWAS summary statistics. Genome Biol 22, 32 (2021). https://doi.org/10.1186/s13059-020-02248-0
The Variant Call Format (VCF) Version 4.2 Specification. (2020).
Available from: https://samtools.github.io/hts-specs/VCFv4.2.pdf