Besides philosophical differences and some gain in power of Bayesian procedures, testing many hypotheses will inevitably increase the probability of errors, both type I and II, and relying solely on statistical methods is not sufficient. It is becoming clear that by controlling the probability of the type I error we may be ignoring the majority of the biologically important findings [95] and we need methods to be able to look at associations with less stringent P-values or posterior probabilities. To this end, we introduced a Bayesian procedure for analysis of GWAS that uses a hierarchical set of filters to reduce the false positive rate without imposing unnecessary stringent thresholds [96]. This procedure leverages the patterns of linkage disequilibrium in the human genome to accept as significant only those associations that are supported by associations of SNPs in the same LD blocks, and our experimental evaluation suggests that these filters help reducing the false positive rate by 50% [96]. Several authors have proposed strategies that leverage properties of the human genome or knowledge of disease mechanisms and pathways more likely to be involved in the disease to prioritized genes [97].
