We also generated prediction models trained in the DGN whole blood cohort that included trans-eQTLs (>1Mb from gene start or end or on a different chromosome) generated from linear regression (p<10−5). We tested the predictive performance of these models in the GTEx whole blood cohort. While a few genes had higher correlations between predicted and observed expression than expected by chance, the departure from the null distribution was much smaller than that for the prediction models based on local SNPs (Supplemental Figure 4), perhaps due to the low power to map trans SNPs. Based on this result, in this paper we focus primarily on results based on local SNPs.
