It is also worthy to point out that the number of selected SNPs varies between genes. This is because we used permutations to decide both the number and the set of SNPs to represent each gene. The permutation of phenotypes and recalculation of statistical values for about half a million SNPs and thousands of subjects is computationally expensive. To seek a balance between the computational complexity and not losing too much information from SNPs, we set a nominal significance threshold chose only SNPs with smaller P-value for pathway analysis. To further reduce computation, we recommend using an upper limit for the number of representative SNPs for each gene.
