The presence of SNPs in expression probes had a larger effect on reproducibility at extremely high thresholds (Figure 3C). For example, the replication rate for cis-eQTLs with BF>5 is not significantly affected by the presence of SNPs in probes (p-value = 0.189); however, replication rate for cis-eQTLs thresholded at BF>10 is significantly affected by probe SNPs (p-value = 0.0354; 65.6% with, 74.9% without SNP) and replication rate is significantly associated with an interaction between probe SNPs and eQTL significance (logistic regression BF-SNP interaction p-value = 0.0224). These results suggest the proportion of non-reproducible cis-eQTLs increases with eQTL significance such that, for eQTLs with BF>10, ∼27% of the non-replication rate can be explained by the presence of hybridization artifacts caused by known polymorphisms. To investigate the potential confounding role of unannotated polymorphisms in eQTL ascertainment, we re-sequenced 15 expression probes for genes that had large discrepancies in correlation measurements between the UW and UC studies that did not overlap a known SNP (9 probes with strong UW correlation but low UC correlation, 6 of the converse; Table S2). We found that
