A computationally efficient Bayesian seemingly unrelated regressions model for high-dimensional quantitative trait...

Area of Science:

• Genetics
• Biostatistics
• Metabolomics

Background:

• Metabolite quantitative trait loci (mQTL) analysis is crucial for understanding genotype-phenotype relationships.
• High-throughput metabolomics data, like that from the Northern Finland Birth Cohort 1966 (NFBC66), often exhibit complex correlation structures.
• Existing multivariate QTL methods frequently overlook these phenotypic correlations or rely on oversimplified assumptions.

Purpose of the Study:

• To develop a computationally efficient Bayesian model for high-dimensional multivariate QTL analysis.
• To simultaneously estimate genotype-phenotype associations and the residual dependence structure among correlated metabolites.
• To address limitations of current methods that ignore or oversimplify phenotypic correlations.

Main Methods:

• A Bayesian seemingly unrelated regressions (SUR) model was developed for high-dimensional data.
• The model incorporates cell-sparse variable selection for distinct genetic predictor-phenotype associations.
• A sparse graphical structure was used for covariance selection, leveraging covariance matrix factorization for computational efficiency.

Main Results:

• The model was applied to the NFBC66 dataset, analyzing 158 metabolites and 9000 single nucleotide polymorphisms.
• Simultaneous estimation of genotype-phenotype associations and metabolite dependence structures was achieved.
• The approach demonstrated the ability to handle complex correlation structures in metabolomics data.

Conclusions:

• The proposed Bayesian SUR model offers an efficient and flexible framework for multivariate mQTL analysis.
• This method enhances the discovery of genotype-phenotype associations in high-dimensional, correlated metabolomics data.
• The R package BayesSUR is available to facilitate the application of this methodology.