trait (i.e. summary association statistics from GWAS) (see Figure 2). Since TWAS using predicted expression is conceptually similar to a test for non-zero genetic covariance between gene expression and trait34, it can also be performed via a two-sample Mendelian randomization from summary statistics35. TWAS using predicted expression can increase power over a standard GWAS when there exist multiple causal variants whose effect on trait is mediated through expression. TWAS also reduces the multiple hypothesis burden by testing tens of thousands of genes instead of millions of SNPs. TWAS using predicted expression typically uses individual-level transcriptome reference data to predict gene expression, but can also be performed using only summary association statistics between SNPs and gene expression, albeit with a reduction in power34. The potential power gains of TWAS are underscored by the recent identification of 71 new susceptibility genes across 28 complex traits, of which 17 have no GWAS association within 1 Mb36. However, TWAS is underpowered compared to standard GWAS when the true biological mechanism is independent of gene expression or when expression data in the most relevant tissue is not available.
