Our method has a critical assumption. In combing P-values of SNPs in a gene we assume that the P-values are independent, although in reality some SNPs in a gene are in linkage disequilibrium (LD). When comparing the results with and without removing SNPs in strong LD, we found there is no big difference between them. However, a future direction is to relax this assumption and develop a SNP selection method that explicitly takes the LD patterns into account rather than remove SNPs in LD. It is interesting to note that Peng et al. [15] also found that ignoring LD could actually lead to better results than methods with very conservative multiple testing corrections. The permutation test we consider might partially alleviate the effect due to LD.
