For each SNP, a P-value is obtained by comparing the genotype frequencies between the cases and controls using the Pearson's chi-square test with two degrees of freedom. Extending the work of Yu et al. [18], we use an adaptive rank truncated product method. The L P-values of the L SNPs mapped to a gene are sorted from smallest to largest: p(1) ≤ p(2) ≤ ... ≤ p(L), with p(l) being the lth smallest P-value. We use to combine the first K P-values, where K is the truncation point. Permuting the phenotypes and computing the statistic in permutated data allows us to assess the overall significance of the K SNPs. In the permutation procedure, we permute the phenotype values N times to obtain N permutated datasets. For the nth permutated dataset, we denote the resulting P-values as , and calculate the corresponding . Then the P-value for evaluating W(K) is calculated by . To maximize the association of the subset of SNPs and the trait, all possible values of K are calculated and the one with the smallest P-value is chosen. The corresponding SNPs are used to represent the gene.
