Efficient two-stage genome-wide association designs based on false positive report probabilities

Despite recent advances, very-high-throughput (VHT) technologies capable of genotyping hundreds of thousands of SNPs in individual samples remain prohibitively expensive for the large studies necessary to screen substantial sections of the genome for variants with modest effects on disease risk. This paper presents a two-stage strategy, where a portion of available samples are genotyped with VHT technology, and a small number of the most promising variants are genotyped with standard high-throughput techniques in the remaining samples as an independent replication study. The sample sizes in the first and second stages and the corresponding significance levels are chosen to limit False Positive Report Probability (FPRP), while maximizing the number of Expected True Positives (ETPs). (The FPRP is the conditional probability that a marker is not truly associated with disease, given the a significant test for disease-marker association.) For a fixed budget, the two-stage strategy has greater power (a larger number of ETPs) than the single-stage strategy (where all subjects are genotyped using expensive VHT technology). Furthermore, concentrating on the FPRP leads to considerable savings relative to strategies designed to control the family-wise error (e.g. Bonferonni correction). The FPRP and number of ETPs can also accommodate researchers' prior beliefs about the number of causal loci and the magnitude of their effects. The expected number of false positives does not change if the true number and effects of causal loci differs from the specified prior (although the false discovery rate will vary), thus limiting the absolute amount of resources spent chasing "false leads."