Effect Size Does Matter

In the last decade, several thousand genome-wide associations (GWAs) have been described, and the technique has proven to be a remarkably robust approach to the identification of common alleles that contribute to disease phenotypes.1 However, to date, few of these highly statistically significant associations have translated into mechanistic insights.2 This translational gap is a consequence of several factors, each of which is relevant to a variable extent for any specific association. Together they conspire to prevent a reproducible trajectory from genotype to phenotype. To some degree, the translational gap has been partly obscured by the tendency to associate GWA signals with neighboring genes whose presumed function makes them appealing as candidates.3 However, the very nature of the GWA approach and emerging data on the consequences of too rapid a jump to conclusions suggest that the ‘nearest gene’ and ‘nearest appealing gene’ strategies are unlikely to offer generalizable solutions.3Article see p 1969Perhaps the most obvious hurdle to establishing causality is the fact that in most instances, the effect sizes at individual loci are in the range of 1.1- to 1.2-fold changes in the prevalence of disease or the magnitude of a quantitative trait.1,2 Such small effect sizes are difficult to model in experimental systems in which typical biological noise cannot be overcome even with enormous sample sizes. In many instances in which putative biology has been inferred from the study of local genes, the published observations are a consequence of loss of function or overexpression of a candidate gene and may not in any way reflect the subtle biology of the locus.The GWA study design is an efficient means of detecting linkage disequilibrium but does not allow the definition of locus boundaries.2 Indeed, so-called fine mapping may facilitate the detection …