Interestingly, significant cis-eQTLs were no more likely to replicate when analyses were restricted to probes targeting the same exon (chi-squared p-value = 0.759), demonstrating that most non-replicating eQTLs (in our study design) can not be accounted for by differential splicing or isoform usage. Similarly, replication was not improved when analyses were restricted to gene expression measurements derived from more than one expression probe (chi-squared p-value = 0.919). Additionally, the minor allele frequency of the associated SNP did not have a significant effect on replication rate (logistic regression chi-squared p-value = 0.600; Figure S10), and eQTLs at imputed SNPs replicated at similar rates to directly genotyped SNPs (logistic regression chi-squared p-value = 0.574; Figure S10). Uncertainty at imputed SNPs does not appear to have a significant effect on cis-eQTL replication rate, as the ratio of observed to expected genotype variance was not associated with replication rate in any of the three sample sets (logistic regression chi-squared p-values all >0.152; Figure S12).
