A Bayesian model selection approach for identifying differentially expressed transcripts from RNA sequencing data.

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• Current methods for analyzing RNA sequencing data often separate transcript expression quantification from differential expression analysis.
• This separation can lead to suboptimal results and a lack of integrated inference.

Purpose of the Study:

• To develop a unified statistical framework for the joint estimation of transcript expression levels and differential expression.
• To improve the accuracy and efficiency of RNA sequencing data analysis by integrating two key tasks.

Main Methods:

• A hierarchical Bayesian model, extending the BitSeq framework, is proposed for joint estimation.
• Markov chain Monte Carlo (MCMC) sampling, including reversible jump MCMC and a collapsed Gibbs sampler, is used for posterior inference.
• A cluster representation of aligned reads enables parallel estimation for computational efficiency.

Main Results:

• The proposed model demonstrates conjugacy for fixed dimension variables, allowing analytical derivation of conditional distributions.
• The collapsed Gibbs sampler is shown to outperform the reversible jump MCMC sampler.
• Benchmarking on synthetic and real RNA sequencing data confirms the algorithm's performance against alternative methods.

Conclusions:

• Jointly estimating expression and differential expression provides a more robust and integrated approach to RNA sequencing analysis.
• The developed Bayesian framework and efficient samplers offer a valuable tool for molecular biologists and bioinformaticians.
• The source code is publicly available, facilitating wider adoption and further research.