Empirical Bayes method for reducing false discovery rates of correlation matrices with block diagonal structure

Area of Science:

• Genomics
• Systems Biology
• Bioinformatics

Background:

• Correlation matrices are crucial for understanding biological regulatory and signaling networks.
• Small experimental sample sizes and genome-wide scale computations pose challenges for accurate correlation matrix estimation.
• Existing methods struggle with accuracy and computational demands in inferring biological networks.

Purpose of the Study:

• To develop an improved method for estimating covariance matrices in gene expression data.
• To address limitations of existing methods in handling small sample sizes and computational complexity.
• To enhance the accuracy of biological network inference.

Main Methods:

• Developed an empirical Bayes approach assuming a block diagonal form for the covariance matrix.
• Applied the method to simulated gene expression data.
• Validated the approach on a real dataset from Bacillus subtilis.

Main Results:

• The empirical Bayes method demonstrated lower false discovery rates compared to existing techniques on simulated data.
• The method successfully identified known regulatory units and interactions in the Bacillus subtilis dataset.
• Achieved significant covariance detection and controlled false discovery rates even with small sample sizes (n=10).

Conclusions:

• The proposed method offers superior performance over existing approaches for estimating covariances and controlling false discovery rates, especially in low-sample scenarios.
• The method facilitates the discovery of potential regulatory networks.
• It serves as a valuable pre-processing step for inferring causal and hierarchical network structures.