Our results show that to provide an accurate answer to the colocalisation problem, high-density genotyping and/or accurate use of imputation techniques are key. The quality of the imputation is another important parameter. Indeed, while the variance of the regression coefficient can be estimated solely on the basis of the minor allele frequency for typed SNPs and sample size (and the case control ratio in the case of a binary outcome) [17], [28], this ignores the uncertainty due to imputation. Filtering out poorly imputed SNPs partially addresses this problem, with the drawback that it may exclude the causal variant(s). Hence, providing estimates of the variance of the MLE, together with the effect estimates, will result in greater accuracy. This additional option is available on the coloc package in R (http://cran.r-project.org/web/packages/coloc).
