To address this question, we used Illumina MetaboChip data and imputed the genotyped regions using the Minimac software ([19] and Methods). We then selected only the subset of variants present in the Illumina 660K genotyping array. We simulated data under the assumption of a shared causal variant, with 4,000 individuals in the biomarker dataset. We then computed the PP4 statistic with and without restricting the SNP set to the Illumina 660K Chip SNPs (Figure 4). We also considered two different scenarios, with the causal SNP included/not included in the Illumina 660W panel (Figures S1 and S2 for more exhaustive simulations).
