The analogy with genotype imputation is relevant here. Genotype imputation uses information from a reference sample to learn how to impute genotypes at the unmeasured SNPs in the test set. Similarly, PrediXcan uses a reference dataset in which both genome variation and gene expression levels have been measured to develop prediction models for gene expression. We use these prediction models to “impute” gene expression (which is unobserved in a typical GWAS), and we do so by estimating the genetically determined component, GReX.
