,
7 tweets,
2 min read
Read on Twitter
Our MPAT (Multiple Phenotype Association Test) using GWAS summary statistics is out in JASA. It addresses the question which multiple phenotype PC or a combination of phenotype PCs should be used to boost GWAS power using multiple phenotypes (1/n) tandfonline.com/doi/full/10.10…
(2/n) Which phenotype PC gives the best power for association analysis of multiple phenotypes? the first phenotype PC, the last phenotype PC, or a combination of phenotype PCs? Which combination? Is the multivariate Wald Test most powerful ?
(3/n)One does not need the original data to perform multiple phenotype association analysis. One can estimate the correlation of multiple phenotypes using the empirical correlation of individual phenotype GWAS summary statistics across all the SNPs or a subset of LD pruned SNPs
(4/n) The optimal phenotype PC for association analysis is the one with the smallest principal angle: the angle between its eigen vector direction and the true effect size direction. The first phenotype PC often has the worst power, the last phenotype PCs often have more power
(5/n) As the true effect size direction is unknown, MPAT proposes an omnibus combination of multiple phenotype PC based tests to provide robust power for multiple phenotype association analysis using individual phenotype GWAS summary statistics
(6/n) The multivariate Wald Test is the same is the phenotype PC based test using a quadratic combination of phenotype PCs weighted using the inverse of their eigen values , and is not always most powerful.
(7/n) The uniformly most powerful multiple phenotype association test does not exist. An omnibus test is needed.
