Variational inference for detecting differential translation in ribosome profiling studies

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• Translational efficiency regulates protein synthesis.
• Ribosome profiling (Ribo-seq) and RNA-sequencing (RNA-seq) measure total and actively translated transcripts.
• Current Ribo-seq analysis methods have limitations in handling paired experimental designs.

Purpose of the Study:

• To develop an improved statistical model for Ribo-seq data analysis.
• To account for the paired nature of Ribo-seq and RNA-seq experiments.
• To provide a user-friendly software tool for analyzing translational efficiency.

Main Methods:

• A hierarchical Bayesian generalized linear mixed effects model was proposed.
• The model incorporates random effects for paired samples.
• A novel variational Bayesian algorithm was implemented in the "riboVI" software tool.

Main Results:

• Simulations showed "riboVI" outperforms existing methods in ranking differentially translated genes.
• "riboVI" effectively controls the false discovery rate.
• Analysis of real Ribo-seq data revealed novel insights into virus-host interactions, including changes in hormone signaling.

Conclusions:

• The proposed hierarchical Bayesian model and "riboVI" software offer a superior approach for Ribo-seq data analysis.
• This method enhances the ability to detect biologically significant changes in translational efficiency.
• The findings provide new biological insights into virus-host interactions.