Examination of the interplay of the factors influencing eQTL replication revealed several interesting trends. As mentioned above, replication probability was significantly associated with SNP to TSS distance, but this association decreases with increasing cis-eQTL significance (distance×BF interaction logistic regression p-value = 3.98×10−5). Thus, location information can help to differentiate real from false positive correlations of modest effect, but is less important for very strong correlations. We constructed stepwise multivariate logistic regression models, restricted to associations with BF>5, and confirmed that BF (logistic regression chi-squared p-value = 7.32×10−3), SNP to TSS distance (p-value = 2.33×10−3), gene expression (p-value = 0.0230), gene expression CV (p-value = 1.33×10−4), and probe SNP×BF interaction (p-value = 0.0207) all have significant effects on the cis-eQTL replication rate. In contrast, SNP minor allele frequency, SNP type (imputed or direct), and genotype variance do not substantially influence replication rate (p-values>0.5).
