Multi-phenotype Transcriptome-wide Association Study (TWAS) Tests Using Summary Statistics

摘要

There is a great interest in multi-trait genetic association tests, which have been well developed for single SNPs. Here, we extend these methods to multi-SNP TWAS tests to improve power of detecting genes associated with phenotypes regulated through similar pathways. We show that the TWAS test statistic for multiple phenotypes has the same form as the single SNP statistic, replacing the Z-score vector from single SNP tests for multiple traits with Z-scores from TWAS. Thus, established methods for combining single-SNP test statistics across multiple traits can be easily extended to the TWAS case, including SUM, Wald, and ASSET tests. We evaluated several such methods in simulation under different alternatives (different covariance and effect sizes among the phenotypes). Out tests have proper Type I error (when SNPs are not associated with any of the traits). Our results suggest that conducting TWAS with multiple phenotypes jointly improves the power of TWAS. The simulation showed improvement in power compared to individual non-combined tests and a simple combined test with Bonferroni correction. However, we observed no uniform supreme multi-trait method, since the power of each method varies across different alternatives. The Wald test was near-optimal in most of scenarios. We then jointly analyzed the TWAS results from the Global Lipids Genetics Consortium of 4 traits (LDL, HDL, TG, and total cholesterol) with cross-tissue weights built with sparse canonical correlation analysis on GTEx gene expression data. The joint analysis identified additional trait-associated genes and provided new information into the gene regulation architecture for these traits.