Covariate-Adjusted Precision Matrix Estimation with an Application in Genetical Genomics.

Area of Science:

• Genetics
• Bioinformatics
• Statistical Genomics

Background:

• Understanding gene regulatory networks is crucial in genomics.
• Genetical genomics data presents challenges due to high dimensionality and confounding genetic effects.
• Existing models often fail to adequately adjust for these genetic covariates.

Purpose of the Study:

• To develop a sparse high-dimensional multivariate regression model for conditional independence in gene expression.
• To estimate the covariate-adjusted precision matrix, revealing gene dependence structures after accounting for genetic effects.
• To identify gene networks within pathways, using yeast genetical genomics data as a case study.

Main Methods:

• Introduction of a sparse high-dimensional multivariate regression model.
• Development of a covariate-adjusted precision matrix estimation method via constrained L1 minimization and linear programming.
• Theoretical analysis establishing asymptotic convergence rates and sign consistency for estimators.
• Simulation studies comparing the proposed method with standard Gaussian graphical models.

Main Results:

• The proposed method significantly improves precision matrix estimation and graphical structure selection.
• Demonstrated effectiveness in identifying gene networks, particularly in the yeast mitogen-activated protein kinase pathway.
• Asymptotic properties established for estimators, supporting their reliability in high-dimensional settings.

Conclusions:

• The developed regression model effectively identifies conditional independence relationships among genes.
• The covariate-adjusted precision matrix provides a robust view of gene expression networks.
• This approach offers a valuable tool for genetical genomics research and gene network discovery.