What is the recommended approach when full GWAS summary statistics are available for only part of the sample, while for another part (e.g., 23andMe participants) only the top 10,000 SNPs are provided? How should this be handled when constructing a polygenic score?
Replying to “What is the recommended approach when full GWAS su…”:
check if the heritability of your GWAS without 23andMe is still reasonable and use that. Most current methods aren’t suitable to use on just 10,000 top SNPs, though you can try an old-fashioned clumping-and-thresholding method
Replying to “What is the recommended approach when full GWAS su…”:
This is common and not ideal. Generally if you can’t access the full set we work with what we can get.
Or you can use the top 10k if they LD pruned with C&T
Replying to “What is the recommended approach when full GWAS su…”:
if you’re making clumping and thresholding scores, you could just use the sumstats from the full sample for the top 10k and restricted sample for the rest. With Bayesian methods like SbayesR, this might be an issue, though because of per-SNP sample sizes varying too much. You could try making two PGSs, C+T for top 10k, exclude those SNPs and everything in LD with them to use SBayesR for the rest
Replying to “What is the recommended approach when full GWAS su…”:
That’s an excellent question. Often we can use the available full gwas sumstats for the subset of samples.
Replying to “What is the recommended approach when full GWAS su…”:
Thanks!