Efficient and Accurate Multiple-Phenotype Regression Method for High Dimensional Data Considering Population Structure

Abstract

A typical genome-wide association study tests correlation between a single phenotype and each genotype one at a time. However, single-phenotype analysis might miss unmeasured aspects of complex biological networks. Analyzing many phenotypes simultaneously may increase the power to capture these unmeasured aspects and detect more variants. Several multivariate approaches aim to detect variants related to more than one phenotype, but these current approaches do not consider the effects of population structure. As a result, these approaches may result in a significant amount of false positive identifications. Here, we introduce a new methodology, referred to as GAMMA for generalized analysis of molecular variance for mixed-model analysis, which is capable of simultaneously analyzing many phenotypes and correcting for population structure. In a simulated study using data implanted with true genetic effects, GAMMA accurately identifies these true effects without producing false positives induced by population structure. In simulations with this data, GAMMA is an improvement over other methods which either fail to detect true effects or produce many false positive identifications. We further apply our method to genetic studies of yeast and gut microbiome from mice and show that GAMMA identifies several variants that are likely to have true biological mechanisms.

Reconstruction of networks with direct and indirect genetic effects

“…This approximation could probably be improved if we could obtain good estimates of and , and set . Such estimates however require fitting an MTM for |S| + 1 traits, which for large S is statistically and computationally challenging, unless pairwise or other approximations are applied (Furlotte and Eskin 2015; Joo et al 2016). Moreover, it seems unclear how should be incorporated in the tests.…”

Powerful Genetic Association Analysis for Common or Rare Variants with High-Dimensional Structured Traits

“…To calculate the exact critical value of a DKAT under a given significance level, we need to study its distribution under the null hypothesis of no association. Two current standard approaches of calculating the null distribution of a genetic association test statistic are permutation-based resampling methods (Hua and Ghosh 2015;Pan et al 2015;Joo et al 2016;Kim et al 2016) and large sample-based asymptotic methods (Kwee et al 2008;Wu et al 2010Wu et al , 2011Broadaway et al 2016;Wu and Pankow 2016). However, both methods have potential limitations.…”

“…Among existing multivariate-trait association tests, both multiple testing-adjusted univariate trait methods (Yang et al 2010;van der Sluis et al 2013) and dimension reduction-based methods (Klei et al 2008;Ferreira and Purcell 2009) can be limited with high-dimensional traits. Other multivariate traits-single SNP association testing methods (O'Reilly et al 2012;Joo et al 2016) suffer from power loss when there are systematic but weak marginal effects for each SNP. To make the comparison fair, we focus on existing methods that test association between multivariate traits and multiple SNPs/rare variants.…”

“…However, results based on dimensionreduction methods are difficult to interpret and lose power when the weights for collapsing the multiple traits are imperfect (Aschard et al 2014). The final category is the broadest and is based on multivariate-regression methods, which often assume a parametric model for the relationships between multiple traits and a single SNP (O'Reilly et al 2012;Zhou and Stephens 2014;Wu and Pankow 2015;Joo et al 2016;Ray et al 2016;Schaid et al 2016). The specific modeling strategies underlying each approach vary with some approaches using classical mixed models and others using alternative strategies, e.g., MultiPhen (O'Reilly et al 2012), which uses ordinal regression to regress a SNP on multiple traits.…”

“…Besides being able to incorporate high-dimensional structured traits in genetic association analysis, another major contribution of DKAT is that we introduce a new test design for genetic association testing. Currently, two most popular Pvalue calculation methods for genetic association analysis is either based on large-sample asymptotic theory (Wu et al 2010(Wu et al , 2011Broadaway et al 2016;Wu and Pankow 2016) or via permutations (Hua and Ghosh 2015;Pan et al 2015;Joo et al 2016;Kim et al 2016). However, the large-sample asymptotic theory-based P-value calculation can lead to a conservative test with accumulated estimation error (Lee et al 2012;Chen et al 2016), as in studies with small samples or high-dimensional traits.…”