population's sample size, all significant detected effects are relatively large; the range of Spearman's rho, the correlation coefficient, is 0.338–0.919 for the normalized and PCA-corrected data, and 0.337–0.933 for the ‘REDUCED data’ (Table S4). There is substantial overlap between genes detected from the normalized and PCA corrected data with that of the REDUCED data (Table S5). Of the genes with significant cis- associations in the REDUCED data analysis, 70–77% of the genes are novel, i.e., were not identified as having a significant cis- association in the normalized and PCA-corrected data analysis, though the vast majority of these significant p-values were close to significance thresholds in the PCA corrected data analysis (Figure S6). The additional cis-eQTLs detected in the REDUCED analysis are likely due to the increased sensitivity of the analysis (see [27]). Of the 22 to 35 percent of the cis-eQTLs that did not replicate in the REDUCED analysis, a portion were likely artifactual associations in the first analysis, or else weak effects with low power for replication (as evidenced by the lowest replication in the analysis of the smallest population, the MEX). Together, these results demonstrate that by applying dimension reduction to the expression data, we do not introduce
