DiPhiSeq: robust comparison of expression levels on RNA-Seq data with large sample sizes

Area of Science:

• Genomics and Bioinformatics
• Statistical Genetics
• Cancer Research

Background:

• RNA-Seq data analysis is crucial for identifying differentially expressed (DE) genes as biological markers.
• Existing DE gene detection methods require improvement for cancerous data due to high expression diversity in cancer samples.
• Cancerous data often exhibits greater expression variability in cancer groups compared to control groups.

Purpose of the Study:

• To develop a statistical method for detecting genes with differential expression and differential dispersion in RNA-Seq data.
• To identify novel clinical markers by analyzing gene expression diversity.
• To enhance the robustness of gene expression analysis in the presence of outliers and noise.

Main Methods:

• Proposed a novel statistical method, DiPhiSeq, for RNA-Seq data analysis.
• Employed a redescending penalty on the quasi-likelihood function for improved robustness.
• Utilized simulations and real data analysis to validate the method's performance.

Main Results:

• DiPhiSeq effectively detects genes with both different average expressions and different diversities of expressions.
• The method demonstrates superior robustness against outliers and noise compared to existing approaches.
• DiPhiSeq identified unique sets of genes, highlighting its potential for novel biomarker discovery.

Conclusions:

• DiPhiSeq offers a robust and effective approach for analyzing RNA-Seq data, particularly for complex biological samples like those from cancer patients.
• The identification of 'differentially dispersed' genes provides new avenues for clinical marker discovery.
• The DiPhiSeq R package is available on CRAN, facilitating its application in the research community.