Online Program

Multivariate Association Analysis with Correlated Traits in Families or Distantly Related Individuals (303988)

Keywords: Multivariate Multiple Linear Regression, Feasible Generalized Least Square, Seemingly unrelated Regression, Multiple phenotypes; Family data; Distantly Related Individuals,Genetic Similarity Matrix,UK Biobank data, Score Test

Multiple genome-wide association studies have reported variants that affect multiple traits, demonstrating evidence of pleiotropy or shared genetic basis for multiple phenotypes. Joint analysis of the correlated phenotypes can improve power to detect those pleiotropic variants. However, the implementation of many such joint analysis techniques can be computationally intensive at a genome-wide level. These multivariate techniques are even harder to implement in families or in distantly related individuals for an additional mode of dependency due to the genetic similarity. A traditional approach in this regard would be to fit a Multivariate Multiple Linear Regression model with a complex covariance structure which would make it computationally very demanding on a genome-wide scale. We reasonably modify the traditional covariance structure to develop a robust and computationally efficient test based on two fundamental statistical concepts namely, Feasible Generalized Least Square and Seemingly Unrelated Regression. compare our method with the existing approaches through extensive simulations and also apply our method on the UK Biobank data with four anthropometric traits.