Accurate Classification of Differential Expression Patterns in a Bayesian Framework With Robust Normalization for...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Empirical Bayes methods offer a framework for differential expression (DE) analysis in RNA-seq data, enabling the computation of posterior probabilities for expression patterns.
• Existing Bayesian tools like baySeq and EBSeq have limitations, particularly in normalization strategies.

Purpose of the Study:

• To compare the performance of baySeq and EBSeq with different normalization methods (default, MRN, TCC) for multi-group RNA-seq differential expression analysis.
• To evaluate the effectiveness of TCC normalization in conjunction with Bayesian DE analysis frameworks.

Main Methods:

• Utilized three-group simulation data and real RNA-seq count data for comparative analysis.
• Assessed two R packages (baySeq and EBSeq) with their default normalization settings against MRN and TCC normalization methods.
• Compared Bayesian DE methods with the generalized linear model (GLM) framework in TCC.

Main Results:

• Bayesian methods combined with TCC normalization performed comparably or better than default settings across various simulations.
• The TCC default DE pipeline (GLM-based) outperformed Bayesian methods with TCC for overall DE evaluation.
• baySeq coupled with TCC demonstrated robustness in assigning expression patterns, even with varied pattern choices.

Conclusions:

• TCC normalization enhances the performance of Empirical Bayes methods for RNA-seq DE analysis.
• Recommends using the TCC default DE pipeline for initial gene ranking and baySeq with TCC for detailed expression pattern assignment.