Variable Selection and Joint Estimation of Mean and Covariance Models with an Application to eQTL Data

Area of Science:

• Genomics
• Bioinformatics
• Systems Biology

Background:

• Gene regulatory relationships are complex and difficult to determine due to genetic and epigenetic factors.
• Existing models often struggle to capture intricate dependencies in genomic data.

Purpose of the Study:

• To develop a robust statistical model for elucidating conditional gene dependencies.
• To introduce a variable selection algorithm for identifying key regulatory interactions and hub genes.

Main Methods:

• Utilized a joint mean and constant covariance model (JMCCM) with modified Cholesky decomposition for precision matrix parametrization.
• Developed a variable selection algorithm combining generalized cross-validation (GCV) and Bayesian Information Criterion (BIC) with Rao and Wald statistics.
• Applied the model to miRNA and single nucleotide polymorphism (SNP) data from yeast (eQTL data).

Main Results:

• Achieved sparse estimation of the precision matrix, effectively representing gene networks.
• Identified significant hub genes consistent with prior biological evidence.
• Simulation studies confirmed the model's efficiency in identifying true underlying networks.
• The yeast eQTL data analysis constructed gene networks that reproduced known biological pathways, including the cell cycle pathway.

Conclusions:

• The proposed JMCCM and variable selection algorithm provide an effective approach for inferring gene regulatory networks.
• The method facilitates the discovery of biologically relevant gene interactions and hub genes.
• Demonstrated the utility of the model in analyzing real-world genomic data, yielding clinically and biologically meaningful insights.