GGMs

GGMs reconstruct Gaussian graphical models from quantitative profile data (e.g., transcripts, metabolites, proteins) to infer molecular regulatory networks by estimating 'shrunk' covariance matrices and testing 'shrunk' partial correlations.

Key Features:

• Shrinkage-Based Covariance Estimation: Employs shrinkage-based covariance estimators to produce an invertible 'shrunk' covariance matrix when sample size is smaller than the number of variables.
• Geometric Reformulation and Probability Density Inclusion: Introduces a geometric reformulation of shrinkage-based GGMs and a novel probability density that explicitly incorporates the shrinkage parameter.
• Exact Significance Testing for 'Shrunk' Partial Correlations: Provides an exact hypothesis testing method for assessing significance of 'shrunk' partial correlations (edges) despite the modified probability density from shrinkage.
• Computational Efficiency and Accuracy: Demonstrates accuracy comparable to Monte Carlo (an unbiased non-parametric method) across shrinkage values while offering improved computational efficiency.

Scientific Applications:

• Network Reconstruction: Reconstructs molecular regulatory networks from gene expression data to reveal interactions among transcripts, metabolites, and proteins.
• Type I Error Control: Ensures accurate control of Type I error rates when inferring network edges from 'shrunk' partial correlations.
• Comparative Performance and Validation: Outperforms existing methods such as the R package GeneNet in empirical evaluations and is validated on gene expression datasets including stress response in Escherichia coli and influenza infection effects in Mus musculus.

Methodology:

Applies a geometric reformulation of shrinkage-based GGMs to integrate the shrinkage parameter into a probability density and enables exact hypothesis testing for 'shrunk' partial correlations.