Significance testing of top vs. random SNPs

Question

[am29]

After performing GWAS, I calculated the percentage of phenotypic variance (SNP-based heritability) for top SNPs and random SNPs using GREML (GCTA). The variance of random SNPs was calculated for 3 different groups of randomly chosen SNPs and then averaged to get the final result. Top SNPs explained much more variance then random (10% > 0.2%). I want to check if the difference is statistically significant.

Which significance test should I use for this?

[dthorbur]

I suspect there is a more appropriate way, but you could do a permutation analysis, randomly sampling n SNPs 1000 times and generate a one-tailed p-value (where n is the number of SNPs in your top SNPs group). If your variance is above 990 of the random permutations then your p-value is 0.01.

I would also recommend using a circular genomic permutation analysis which accounts for genomic proximity and structure of outliers when conducting a permutation analysis as described in Cabrera et al. 2012.