BayesSUR

BayesSUR performs joint modeling of correlated multivariate phenotypes using a Bayesian seemingly unrelated regressions framework to discover high-dimensional quantitative trait loci, with particular application to metabolite QTL analysis.

Key Features:

• Bayesian Seemingly Unrelated Regressions (SUR) Model: Implements a Bayesian SUR framework for simultaneous analysis of multiple phenotypes while accommodating their correlations.
• Cell-Sparse Variable Selection: Enables association of different genetic predictors with distinct phenotype responses via cell-sparse variable selection.
• Sparse Graphical Structure for Covariance Selection: Incorporates a sparse graphical model to capture conditional dependencies among phenotypic variables for covariance selection.
• Factorization of Covariance Matrix: Utilizes factorization of the covariance matrix to reduce computational complexity in large model spaces.
• Computational Efficiency for Large-Scale Genetic Data: Employs a computationally efficient Bayesian SUR implementation tailored for large-scale genetic datasets.
• Joint Estimation of Associations and Residual Dependence: Simultaneously estimates genotype-phenotype associations and the residual dependence structure among phenotypes.

Scientific Applications:

• Metabolite QTL Mapping (NFBC66): Applied to the Northern Finland Birth Cohort 1966 (NFBC66) with 158 metabolites measured by NMR spectroscopy and genotype data of 9,000 directly genotyped SNPs.
• Multivariate QTL Mapping in High-Throughput Biomarker Studies: Suited for QTL mapping of correlated multivariate phenotypes derived from high-throughput biomarker technologies.
• Estimation of Residual Dependence among Metabolites: Enables simultaneous estimation of genotype-phenotype associations and residual metabolite dependence to elucidate inter-metabolite relationships.

Methodology:

Uses a Bayesian seemingly unrelated regressions framework with cell-sparse variable selection, a sparse graphical structure for covariance selection, factorization of the covariance matrix, and joint estimation of genotype-phenotype associations and residual dependence.