Identifying differentially expressed genes in meta-analysis via Bayesian model-based clustering

Area of Science:

• Bioinformatics
• Statistical Genetics
• Computational Biology

Background:

• Gene expression meta-analysis combines data from multiple studies to increase statistical power.
• Identifying differentially expressed genes (DEGs) is crucial for understanding biological processes and disease mechanisms.

Purpose of the Study:

• To propose a Bayesian model-based clustering approach for robust identification of DEGs in meta-analysis.
• To compare the performance of the proposed Bayesian method against conventional permutation methods.

Main Methods:

• A Bayesian hierarchical model was employed to integrate information across studies.
• A mixture prior was utilized to distinguish between DEGs and non-DEGs.
• Markov chain Monte Carlo (MCMC) methods were used for posterior estimation and handling missing data.

Main Results:

• The Bayesian approach provides easily computable significance measures like Bayesian false discovery rate (FDR), local FDR, and integration-driven discovery rate (IDR).
• The proposed model-based method demonstrated superiority over permutation methods, particularly in scenarios with imbalanced differential expression (excessive under- or over-expressed genes).

Conclusions:

• The Bayesian model-based clustering offers a powerful and flexible framework for gene expression meta-analysis.
• This method enhances the accuracy of DEG identification, especially in complex datasets with skewed expression patterns.