A marginal mixture model for selecting differentially expressed genes across two types of tissue samples

Area of Science:

• Bioinformatics
• Statistical Genetics
• Computational Biology

Background:

• Bayesian hierarchical models are widely used for identifying differentially expressed genes.
• Existing models assume constant marginal means and variances across gene clusters and tissue types.
• Model selection for Bayesian hierarchical models in microarray data analysis is challenging.

Purpose of the Study:

• To develop a flexible marginal mixture model for gene expression data.
• To address limitations of existing Bayesian hierarchical models regarding marginal distributions.
• To create an improved method for selecting differentially expressed genes across tissue types.

Main Methods:

• Proposed a marginal mixture model approximating gene profile distributions with three-component multivariate Normal distributions.
• Each component allows for differing marginal mean vectors and covariance matrices.
• Derived a gene selection method based on the proposed mixture model.

Main Results:

• The proposed gene selection method demonstrated good performance on a real microarray dataset.
• On simulated datasets, the method consistently achieved superior performance based on class agreement indices.
• Outperformed several other gene selection methods across different mixture model simulations.

Conclusions:

• The marginal mixture model offers a flexible alternative to traditional Bayesian hierarchical models for gene expression analysis.
• The derived gene selection method is effective and robust for identifying differentially expressed genes.
• This approach enhances the accuracy and reliability of gene expression profiling studies.