Mixture prior for sparse signals with dependent covariance structure.

Area of Science:

• Statistics
• Genomics
• Signal Processing

Background:

• Estimating signals with unknown sparsity and correlations is a significant challenge in statistical analysis.
• Existing methods often assume signal independence, limiting their applicability to complex, real-world data.

Purpose of the Study:

• To develop a novel estimation method for normal mean problems with unknown sparsity and correlations.
• To effectively decompose and address the common dependence and weakly dependent error terms in observed signals.
• To provide a robust method applicable to both simulated and real-world biological data.

Main Methods:

• Decomposition of the arbitrary dependent covariance matrix into common and weakly dependent error terms.
• Subtraction of common dependence to reduce signal correlations.
• Empirical Bayesian estimation of sparsity using a likelihood approach on the modified signals.

Main Results:

• The proposed method demonstrates favorable performance compared to existing independent identically distributed signal models.
• Simulations with moderate to high sparsity and diverse dependency structures validate the algorithm's effectiveness.
• Application to HapMap gene expression data yields results consistent with established findings.

Conclusions:

• The developed method offers a significant advancement in estimating signals with complex dependency structures.
• It provides a practical and effective approach for analyzing sparse and correlated data, particularly in genomics.
• The method's successful application to HapMap data highlights its real-world utility and reliability.