Subset Testing and Analysis of Multiple Phenotypes (STAMP)

TL;DR

STAMP is a novel mixture model-based method for region-specific meta-analysis of multiple phenotypes across GWAS, effectively identifying associated phenotype subsets and distinguishing true signals from heterogeneity.

Contribution

It introduces a flexible, mixture model framework for multi-phenotype meta-analysis, improving power in detecting true associations when up to 50% of outcomes are related.

Findings

STAMP outperforms standard methods when ≤50% of phenotypes are associated.

It effectively identifies subsets of associated phenotypes.

Demonstrated on cancer risk and eQTL data from TCGA.

Abstract

Meta-analysis of multiple genome-wide association studies (GWAS) is effective for detecting single or multi marker associations with complex traits. We develop a flexible procedure ("STAMP") based on mixture models to perform region based meta-analysis of different phenotypes using data from different GWAS and identify subsets of associated phenotypes. Our model framework helps distinguish true associations from between-study heterogeneity. As a measure of association we compute for each phenotype the posterior probability that the genetic region under investigation is truly associated. Extensive simulations show that STAMP is more powerful than standard approaches for meta analyses when the proportion of truly associated outcomes is $\leq$ 50\%. For other settings, the power of STAMP is similar to that of existing methods. We illustrate our method on two examples, the association of a region on chromosome 9p21 with risk of fourteen cancers, and the associations of expression of quantitative traits loci (eQTLs) from two genetic regions with their cis-SNPs measured in seventeen tissue types using data from The Cancer Genome Atlas (TCGA).