LD-prune variants while maintaining a list of SNPs

Question

[Jautis]

Hi everyone,

I have a vcf file from which I would like to retain (i) a list of known, associated SNPs and (ii) background, independent variants that are not associated with each other or with any of my associated SNPs.

I know if I just wanted independent SNPs, I could use something like plink --indep-pairwise, but that wouldn't guarantee that I keep all of my associated SNPs or that the "random" SNPs are independent from them. I could theoretically overcome that problem by then removing random SNPs that are in linkage with my associated SNPs one by one. For example, I could add associated variants to the vcf file, use vcftools to calculate r2 in the neighborhood around each associated variant, and form a list of associated background variants to remove (e.g. those with r2 > 0.5 to the associated variant). However, running this for each of the thousands of associated variants feels like an extremely inefficient approach.

Do you have any suggestions for solving this problem, or know of any tools which may help?

Thanks in advance!

Answer 1

[Jautis]

I was given a solution by a colleague -- it seems a little hacky, but worked really well!

one way to get this to work is to use --clump in plink2. The idea is you can create an artificial association file in which your eQTLs have significant pval (e.g. 1e-5) and all other variants (the pool to select background variant from) to have pval 5e-4. With this, these eQTLs will be the index SNPs and the algorithm will clump the candidate background variants which are in high LD to eQTLs. The remaining candidate background variants in your output list will be the independent background variants (they are indep to each other and indep to eQTLs) to use