P-value distribution in case-control association studies

There is accumulating evidence that common traits are due to fairly large numbers of SNPs, each with rather little contribution. In such a situation, single-locus methods may suffer from low power, especially when sample size is small. In this study, we propose a method to estimate the number of SNPs that confer susceptibility to complex disease. We compute p-values for all SNPs and test whether there are significantly more small p-values than expected. For any given split point between 0 and 1, we count the numbers of p-values smaller and larger than the split point, and estimate the corresponding expected numbers of p-values. Then we compute the goodness-of-fit chi-square and the corresponding z-score. We define the maximal z-score across all split points as our test statistic, whose significance level is evaluated by permutation tests. By analyzing two real datasets, we demonstrate that our method is promising in case-control association studies.