Another related issue is the choice of prior probabilities for the various configurations. For the eQTL analysis, we used a prior probability for a cis-eQTL. A more stringent threshold may be better suited for trans-eQTLs where the variants are further away from the gene under genetic control. We also used a prior probability of for the lipid associations. Although our knowledge about this is still lacking, this estimate has been suggested in the literature in the context of GWAS [20], [31], [32]. We assigned a prior probability of for , which encodes the probability that a variant affects both traits. It has been shown that SNPs associated with complex traits are more likely to be eQTLs compared to other SNPs chosen at random from GWAS platforms [33], and a higher weighting for these SNPs has been proposed when performing Bayesian association analyses [34], [35]. Also, eQTLs have been shown to be enriched for disease-associated SNPs when a disease-relevant tissue is used [9], [36]. Our sensitivity analysis for the parameter showed broadly consistent results (Table S1). In cases where GWAS data
