PrediXcan application to a GWAS dataset consists of “imputing” the transcriptome using the weights derived from reference transcriptome datasets and correlating the GReX with the phenotype of interest using regression methods (e.g., linear, logistic, Cox) or non-parametric approaches (e.g., Spearman). (For the specific results on disease phenotypes analyzed here, we used logistic regression with disease status.) We are aware of the attenuation bias that arises because of the error in the estimation of GReX. This is a subject to be investigated in the future, but this bias does not invalidate our analysis since we only use the estimate of GReX as a discovery tool. Figure 2 summarizes the flow of the method development described above.
