covEB

covEB estimates block-diagonal correlation matrices from gene expression data using an empirical Bayes framework to improve covariance estimation in small-sample and genome-wide analyses.

Key Features:

• Empirical Bayes Framework: covEB applies empirical Bayes estimation to refine covariance and correlation estimates from gene expression data.
• Block Diagonal Matrix Assumption: covEB models the covariance/correlation matrix as block diagonal to segment internally correlated blocks that are assumed independent across blocks.
• Shrinkage Parameters: covEB uses shrinkage parameters to threshold small values toward zero, reducing noise and stabilizing estimates with limited samples.
• Improved False Discovery Rate Control: covEB demonstrates improved control of false discovery rates in simulated data compared to existing methods.
• Application to Real Data (Bacillus subtilis): covEB has been applied to Bacillus subtilis gene expression datasets to detect known regulatory units and interactions.
• Utility in Network Inference: covEB provides refined covariance estimates that can be used as preprocessing for partial correlation calculation and causal or hierarchical network inference.

Scientific Applications:

• Gene Expression Analysis: estimating correlations in gene expression studies where sample sizes are limited.
• Regulatory Network Discovery: identifying correlated gene or protein groups and supporting detection of regulatory interactions.
• Pre-processing for Advanced Analyses: providing improved covariance inputs for downstream partial correlation and causal inference methods.

Methodology:

Empirical Bayes estimation of correlation/covariance matrices under a block-diagonal structural assumption with shrinkage parameters to threshold small values.