Statistical test for QQ-plot comparison of GWAS summary statistics

Question

[f-rasmussen]

I haver performed a GWAS and I want to know if there are significantly more significant p-values than I would expect by random chance. Normally this is visualized with a QQ-plot but I want to have a p-value to support this claim. For that reason I am looking for a way to say that the observed and expected p-value distributions are significantly different. I only have access to summary data, so permutation tests are not possible.

Does anyone know of a method to do this test?

My take is using the Kolmogorov-Smirnoff test by using the observed and theoretical distribution. I am however, in doubt about how to properly define the theoretical p-value-distribution.

p_exp <- runif(9e6, 1/9e6, 1)

In this case there will approx. alpha * N significant p-values, with significance threshold alpha and Ntests.

Answer 1

[devon.orourke]

Have you tried a permutation test with PLINK? See here. I've managed to wrangle the --assoc test with perm and mperm options for data with millions of SNPs. Perhaps it could work for you too?

[f-rasmussen]

I forgot to inlcude an important piece of information.... I only have access to summary data. Another and much more relevant reason to not do permutation.