Bayesian Analysis of RNA-Seq Data Using a Family of Negative Binomial Models

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• RNA-Seq data analysis traditionally focuses on gene expression, exon usage, or transcript expression separately.
• Existing methods often employ independent Negative Binomial (NB) models, which may not ensure distributional consistency across different features.
• Counts following a NB distribution for one feature (e.g., exon) do not guarantee a NB distribution for others (e.g., gene/transcript).

Purpose of the Study:

• To develop a cohesive Negative Binomial (NB) modeling framework for integrated RNA-Seq data analysis.
• To address the limitations of independent models by proposing a family of NB models that unify gene, exon, and transcript analyses.
• To improve the accuracy and efficiency of RNA-Seq data interpretation by accounting for feature interdependencies.

Main Methods:

• Proposed a family of Negative Binomial (NB) models for integrated analysis of gene expression, relative exon usage, and transcript expression.
• Developed simple Gibbs sampling algorithms for posterior inference, utilizing conjugate priors for computational tractability.
• Incorporated read assignment uncertainty to enhance the estimation of relative transcript usage.

Main Results:

• The proposed integrated NB model demonstrated coherence across gene, exon, and transcript analyses.
• Gibbs sampling algorithms facilitated efficient posterior inference and simplified relative usage estimation.
• Extensive simulations validated the performance and robustness of the developed models.

Conclusions:

• The integrated NB model provides a unified and statistically sound approach for comprehensive RNA-Seq data analysis.
• This framework improves the analysis of gene expression, exon usage, and transcript expression by considering their interrelationships.
• The developed methodology offers a more accurate and efficient tool for biological interpretation of RNA-Seq data.