The Beta-Binomial Distribution for Estimating the Number of False Rejections in Microarray Gene Expression Studies

Area of Science:

• Bioinformatics
• Statistical Genetics
• Genomics

Background:

• Differential expression analysis commonly assumes independence of null hypotheses.
• This assumption leads to the empirical false discovery rate (eFDR) estimator.
• Ignoring correlations among non-differentially expressed genes can impact analysis accuracy.

Purpose of the Study:

• To develop a novel statistical method for differential expression analysis.
• To account for correlations among non-differentially expressed genes.
• To introduce the beta-binomial false discovery rate (bbFDR) estimator.

Main Methods:

• Modeling the number of false rejections (V) using the beta-binomial distribution.
• Deriving the beta-binomial false discovery rate (bbFDR) estimator.
• Utilizing permutations to generate observed values of V under null hypotheses.
• Fitting a beta-binomial distribution to the observed values of V.

Main Results:

• The bbFDR estimator accounts for correlations among non-differentially expressed genes.
• Simulation studies show bbFDR outperforms eFDR in specific scenarios with correlated genes.
• The method was applied to compare gene expression in soft tissue sarcoma and normal tissues.

Conclusions:

• The beta-binomial approach offers a more robust estimation of false discovery rates when gene expression data exhibit correlations.
• This method enhances the reliability of differential expression analysis in genomics.
• The bbFDR provides a valuable alternative for analyzing complex gene expression datasets.