Wow, these p-values are tiny! Fig 1 has heaps of SNPs at -log10 > 50, that's crazy. Could be an effect of their relatively strange decision to use a simple mode

Wow, these p-values are tiny! Fig 1 has heaps of SNPs at -log10 > 50, that's crazy. Could be an effect of their relatively strange decision to use a simple model for GWAS - they used a straightforward linear regression with a more stringent p-value cutoff (P<5*10^-8), which I guess explains these ridiculously tiny p-values. I wonder how the results would have looked like with a more complex model (like an mixed linear model as implemented in GAPIT) or with a different strategy to adjust for multiple testing (all of those methods in PLINK)