, 2013], though this is lower than a large meta-analysis of twin data [Nan et al., 2012]). In Figure 3, we show results under the assumption that MD has a similar genetic architecture to weight (red dotted line) or to height (black continuous line) (Yang et al., 2010b). We estimated the number of samples needed for an MD GWAS to have 80% power to detect at least one locus, for different disease prevalences. If MD has a genetic architecture similar to weight (red dotted line), then, for a disease prevalence of 10% (typical

of most surveys of MD), a sample size of more than 50,000 cases will be needed to detect at least one genome-wide significant hit. About 10,000 cases are needed if MD has a genetic architecture similar to height. Figure 3 also shows that disease prevalence has a big impact on power. For example, while power to detect a variant that EPZ-6438 in vitro explains 0.08% of the variance on liability to MD will be 4%, in a sample size of 10,000 cases and 10,000 this website controls, power in schizophrenia (prevalence 1%) is

approximately 50% for the same sample size. The effect of disease prevalence (shown on the vertical axis) is not linearly related to sample size. In order to find genes with a smaller sample size, we need to collect a sample that has a lower prevalence. That could be achieved in one of two ways. If MD is truly a quantitative phenotype, then the extremes of the distribution will represent a

Tryptophan synthase less prevalent form of disease. We could take disease that is so severe that it has a prevalence of 0.5% or lower, so that fewer than 20,000 cases would provide 80% power to detect at least one locus. The problem is finding the appropriate severity scale. Alternatively, we could identify rare subtypes of depression that are less prevalent and we hope represent a more homogenous condition than MD broadly defined. Ideally, such subtypes would have a different genetic architecture, veering more toward that of height than of weight, so that much smaller samples are needed. Do such heritable subtypes of MD exist? We address this question below. We start however with a review of the genetics literature to determine whether there might be rare but relatively large-effect loci that GWASs have been unable to detect. The data we have summarized so far are compatible with the hypothesis that the genetic basis of MD arises from the joint effect of very many loci of small effect, with odds ratios of much less than 1.2. However, it is also compatible with the existence of larger effect loci, under two alternative (but not incompatible) hypotheses; first, some of the heritability of MD is explained by rare relatively large-effect loci; second, larger effect sizes would be observed if more homogeneous heritable phenotypic groupings could be identified.