Assessing differential expression in two-color microarrays: a resampling-based empirical Bayes approach

Area of Science:

• Genomics and Bioinformatics
• Statistical Genetics
• Computational Biology

Background:

• Microarray analysis is crucial for gene expression, SNP identification, and methylation detection.
• Existing methods like Smyth's parametric method and Significance Analysis of Microarrays (SAM) have limitations with non-normally distributed data and fold change criteria.
• Violations of normality assumptions can inflate Type I error rates in parametric methods.

Purpose of the Study:

• To introduce a novel Resampling-based empirical Bayes (REB) method for microarray data analysis.
• To address limitations of existing methods, particularly concerning non-normally distributed data and fold change thresholds.
• To evaluate the performance of REB against Smyth's method and SAM.

Main Methods:

• Development of a novel approach combining resampling with empirical Bayes methods (REB).
• Comparative analysis using simulation studies to assess sensitivities, specificities, and false discovery rates (FDR).
• Application to a preterm delivery methylation study to illustrate FDR control differences.

Main Results:

• REB demonstrates significantly higher specificity and lower FDR compared to Smyth's parametric method for non-normally distributed data.
• REB shows higher statistical power than SAM when the proportion of differentially expressed genes is large, for both normal and non-normal data.
• REB is impervious to fold change thresholds, eliminating the need for control dataset selection.

Conclusions:

• The Resampling-based empirical Bayes method offers a robust and accurate approach for microarray data analysis, especially with non-normal distributions.
• REB provides improved control over false discovery rates and enhanced statistical power.
• The REB method is generalizable and shows promise for next-generation sequencing RNA-seq data analysis.