An expectation and maximization algorithm for multivariate genome-wide association studies (EMmvGWAS)

Area of Science:

• Genetics
• Biostatistics
• Computational Biology

Background:

• Genome-wide association studies (GWAS) traditionally analyze one quantitative trait at a time.
• Joint analysis of multiple traits can enhance statistical power and reveal shared genetic architecture.
• Existing multivariate GWAS methods face significant computational challenges.

Purpose of the Study:

• To present EMmvGWAS, an efficient R-based computational framework for multivariate GWAS.
• To reduce computational time for analyzing multiple traits without compromising statistical power.
• To facilitate the identification of pleiotropic effects and shared genetic underpinnings of complex traits.

Main Methods:

• Utilizes an Expectation-Maximization (EM) algorithm to estimate genetic and environmental covariance matrices.
• Employs a semi-exact method by treating the ratio of covariance matrices as a constant for efficient genome-wide scanning.
• Enables closed-form solutions for marker effects and residual covariance matrices.

Main Results:

• EMmvGWAS significantly reduces computational time for multivariate GWAS.
• The method demonstrates scalability and robust performance in simulation studies and real datasets (rice, mice, human).
• The framework supports multivariate analyses and can accommodate univariate analyses.

Conclusions:

• EMmvGWAS provides an efficient and powerful approach for multivariate GWAS.
• The R package is publicly available, promoting wider adoption in genetic research.
• This method aids in uncovering complex trait genetic architecture and pleiotropic effects.