We developed a new approach to identify genes whose expression is significantly associated to complex traits in individuals without directly measured expression levels (Methods). We leveraged a relatively small set of reference individuals for whom both gene expression and genetic variation (single nucleotide polymorphisms, SNPs) have been measured to impute the cis-genetic component of expression into a much larger set of phenotyped individuals from their SNP genotype data (Figure 1). The imputed expression can be viewed as a linear model of genotypes with weights based on the correlation between SNPs and gene expression in the training data while accounting for linkage disequilibrium (LD) among SNPs. We then correlated the imputed gene expression to the trait to perform a transcriptome-wide association study (TWAS) and identify significant expression-trait associations (Methods). Work in parallel to ours has also proposed to find expression-trait associations through imputation of gene expression when GWAS at an individual level is available16(see Discussion). However, a critical limitation is that large-scale GWAS data are typically only publicly available at the level of summary association statistics (e.g. individual SNP effect sizes)2–4.
