Improved variational Bayes inference for transcript expression estimation.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• RNA-sequencing (RNA-seq) quantifies transcript expression by aligning short reads to a reference genome.
• Ambiguity in read mapping due to shared transcript sequences complicates accurate expression quantification.

Purpose of the Study:

• To develop an improved variational Bayesian (VB) method for estimating transcript expression from RNA-seq data.
• To address the variance underestimation issue inherent in standard VB approximations.

Main Methods:

• Application of variational Bayesian (VB) techniques to model transcript expression as a mixture of distributions.
• Introduction of a novel approach integrating latent allocation variables out of the VB approximation.
• Comparison with existing methods using simulation studies and real RNA-seq datasets.

Main Results:

• The proposed VB method demonstrates improved computational efficiency compared to Markov chain Monte Carlo.
• The novel approach leads to a better marginal likelihood bound and more accurate posterior variance estimation.
• Enhanced performance observed in both simulated and real RNA-seq data analyses.

Conclusions:

• The refined VB method offers a more robust and accurate approach for transcript expression quantification in RNA-seq studies.
• This advancement aids in resolving read mapping ambiguities and improving the reliability of gene expression analysis.