mrcieu / gwas-vcf-specification Goto Github PK

Hi,

How do you handle harmonizing case/control GWAS where sumstats are reported as OR and 95%CI? Is there established method to convert this to ES and SE?

Thanks,

Vince

with thanks to @jielab for the suggestion

Hello
I am writing because while using the MRCIEU open GWAS project, I came across a GWAS for ALS and on the ieu webpage https://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST005647/ it states that this gwas has around 39 million SNPs. However, when I downloaded and converted to .txt and reformatted using unique(sumstats_file) in R, there were only about 9 million left. 30 million rows of SNPs got removed! I checked and the rows/SNPs that got removed were duplicates.
I then went to the webpage provided by the paper and downloaded their sumstats file and it contained about 10 million SNPs, roughly corresponding to the unique 9 million+ from the .VCF.
This is quite problematic, for all sorts of reasons and especially when multiple comparisons are made. I am quite curious to why this happened.
On the ieu website, there is no particular issues contact, so I am choosing to report it here. Hopefully it will get answered. Thanks very much.

Hi,
Can I ask how the example vcf was generated? I was trying to recreate using gwas2vcf but have been unable to do so. I have a single study with summary stats for multiple outcomes and would like to store the data together - any advice?

Cheers,
Nick

https://github.com/MRCIEU/gwas-vcf-specification/blob/1209f37adb9ca1e60d45519dac8fd75990edd798/example.vcf.txt#L43C1-L45C1

We should make explicit that duplicated variants (same CHROM:POS:REF:ALT) are not allowed, as per VCF spec. The VCF spec does not make this explicit (even though it can be deduced) and it creates confusion among end users. Indeed, malformed/invalid VCFs are often generated with duplicate variants, even though they would fail validation using EBI's https://github.com/EBIvariation/vcf-validator

Tests:

• Extract single rsid
• Interval queries
• pval query
• multi-trait analysis

In most tools, ES is treated like a β rather than an odds ratio (OR). Additionally, if an OR is used, the results will be incorrect. Because of this, could you add to the documentation and specifications that ES should be β-like?

@explodecomputer do you agree?

Hi,

I performed two-sample Mendelian randomization analyses for several traits through OpenGWAS database, and now I'm trying the new method CAUSE (Causal Analysis Using Summary Effect Estimates, PMID: 32451458), which needs to format GWAS summary data of exposure and outcome traits first. I have two questions:

(1) How to extract several specific columns (e.g. beta, se, A1, A2) from the downloaded GWAS-VCF files to a tabular form?
(2) Some of the GWAS-VCF files were not found (e.g. https://gwas.mrcieu.ac.uk/files/finn-a-I9_HYPTENS/finn-a-I9_HYPTENS.vcf.gz), where can I download them?

Many thanks,
Hellen

Hello,

would it be possible for the more IT than GWAS crowd to have a fool-proof table along the lines:

beta_col => ES
pval_col => LP
se_col => SE
imp_info_col => SI

It gets more complicated further one goes from the obvious.

Thanks a lot for your help.
DK

PS I am aware that automating the JSON creation from the TSV column names and some dictionary is a rather tall order, but would it be possible to have some CSV/JSON file with usable by gwas2vcf TSV column names encountered in the wild?