shared. Both the ranking method and proportionality testing share the drawback of having to specify a subset of SNPs to base the test on, and Wallace [14] shows that this step can generate significant biases. The main sources of bias are overestimation of effect sizes at selected SNPs (termed “Winner's curse”), and the fact that, owing to random fluctuations, the causal variant may not always be the most strongly associated one. These factors lead to rejection of colocalisation in situations where the causal SNP is in fact shared. Although this can be overcome in the case of proportionality testing by averaging over the uncertainty associated with the best SNP models [14], perhaps the greatest limitation is the requirement for individual level genotype data, which are rarely available for large scale eQTL datasets.
