Biomarker detection and categorization in ribonucleic acid sequencing meta-analysis using Bayesian hierarchical...

Area of Science:

• Bioinformatics
• Genomics
• Statistical Genetics

Background:

• Meta-analysis of transcriptomic studies, particularly RNA-sequencing (RNA-seq), enhances statistical power for detecting differentially expressed genes.
• Existing two-stage meta-analysis methods for RNA-seq data, which combine summary statistics, can lose power for genes with low expression or short length.
• The increasing availability of public RNA-seq datasets necessitates robust methods for integrated analysis.

Purpose of the Study:

• To develop a novel, full Bayesian hierarchical model for RNA-sequencing meta-analysis.
• To improve the sensitivity and accuracy of detecting differentially expressed genes compared to traditional methods.
• To facilitate biological interpretation by categorizing biomarkers based on differential expression patterns across studies.

Main Methods:

• Proposed a full Bayesian hierarchical model (BayesMetaSeq) that directly models RNA-seq count data.
• Integrated information across genes and studies to enhance statistical power.
• Incorporated latent variables and a Dirichlet process mixture (DPM) prior to model heterogeneity and categorize biomarkers.

Main Results:

• BayesMetaSeq demonstrated improved sensitivity and accuracy in detecting differentially expressed genes through simulations.
• The method effectively integrated information across multiple RNA-seq studies.
• A real-data application on HIV-1 transgenic rat brains yielded significant biological findings.

Conclusions:

• BayesMetaSeq offers a powerful and accurate approach for RNA-sequencing meta-analysis.
• The model's ability to handle cross-study heterogeneity and categorize biomarkers enhances biological insight.
• This method advances the analysis of large-scale transcriptomic data for biomarker discovery.