Bivariate Gaussian Mixture Model for GWAS summary statistics

Bivariate Gaussian Mixture Model (GMM) is a novel statistical technique that jointly models GWAS summary statistics from two phenotypes as a mixture of causal, pleiotropic and null components. The model can be used as an... [ view full abstract ]

Bivariate Gaussian Mixture Model (GMM) is a novel statistical technique that jointly models GWAS summary statistics from two phenotypes as a mixture of causal, pleiotropic and null components. The model can be used as an enrichment test to detect presence of shared genetic architecture between the phenotypes, as well a tool to discover specific loci that simultaneously affect both phenotypes. The model estimates true proportion of causal SNPs after controlling for LD structure, provides per-SNP posterior probability of being causal together with corresponding local false discovery rate, and estimates posterior effect sizes. In this presentation we will present basic concepts behind Bivariate GMM, apply in to Schizophrenia, Bipolar Disorder and other phenotypes, and discuss future applications.