We found that 49.1% and 57.6% of significant cis-eQTLs (UC BF>5) replicated in the UW and Merck studies, respectively (i.e., p-value<0.05 and concordant effect directions; Figure 2A–2B, Figure 3C). The lower observed replication rate for the UW study is partially attributable to the smaller sample size (60 vs 266), but may also reflect platform-dependencies. 66.7% of significant cis-eQTL associations replicated in at least one of the two replication cohorts, while 47.6% replicated in both cohorts. Cis-eQTLs that replicated in one replication study were significantly more likely to replicate in the second replication study than expected by chance (chi-squared p-value<2×10−16) and twice-replicated eQTLs had larger effect sizes than eQTLs that replicate in only one study (Wilcoxon rank-sum test p-value<2×10−16; Figure S5, examples of non-replicated cis-eQTLs displayed in Figure S7).
