Fully Bayesian mixture model for differential gene expression: simulations and model checks

Area of Science:

• Bioinformatics
• Statistical Genetics
• Computational Biology

Background:

• Differential gene expression analysis is crucial for understanding biological processes.
• Existing methods often rely on fixed proportions of differentially expressed genes.
• Accurate estimation of gene expression variability is essential.

Purpose of the Study:

• To develop a flexible Bayesian hierarchical model for detecting differentially expressed genes.
• To estimate the proportion of differentially expressed genes and mixture parameters in a fully Bayesian manner.
• To guide the selection of appropriate mixture priors using predictive model checks.

Main Methods:

• A 3-component mixture prior is formulated to classify genes.
• Gene variances are modeled as exchangeable to account for inter-gene variability.
• Bayesian estimation is employed for model parameters and proportions.

Main Results:

• The model successfully estimates the proportion of differentially expressed genes and mixture parameters.
• Accurate estimation of false discovery rates is achieved.
• Predictive model checks favored a mixture model with extra variability around zero over a point mass null.

Conclusions:

• The proposed Bayesian hierarchical model offers an improved approach to differential gene expression analysis.
• The method provides a data-driven way to select appropriate mixture priors.
• The developed R software package facilitates the application of this model in biological research.