A mixture model approach to multiple testing for the genetic analysis of gene expression

Area of Science:

• Genetics
• Statistical genetics
• Bioinformatics

Background:

• Genome-wide association studies (GWAS) generate vast amounts of data, leading to challenges in multiple testing correction.
• The false discovery rate (FDR) is a common metric for controlling errors in such studies.
• Accurate estimation of FDR and related metrics is crucial for reliable genetic findings.

Purpose of the Study:

• To propose a novel finite mixture model for estimating local FDR (lFDR), FDR, and false non-discovery rate (FNR).
• To apply this model to variance-component linkage analysis.
• To provide a robust method for handling multiple testing in dense genetic marker studies.

Main Methods:

• Development of a parametric finite mixture model.
• Empirical estimation of the null distribution.
• Application to variance-component linkage analysis.
• Utilizing microarray expression profiles from the Genetic Analysis Workshop 15 (GAW15) Problem 1 dataset.

Main Results:

• The proposed model effectively estimates lFDR, FDR, and FNR.
• Demonstrated empirical estimation of the null distribution.
• Successful application on a real-world genetic dataset.

Conclusions:

• The finite mixture model offers an improved approach for estimating FDR and related criteria in genetic studies.
• This method enhances the reliability of findings from dense genome-wide marker analyses.
• The approach is particularly valuable for complex genetic analyses like variance-component linkage.