Robust and Adaptive Non-Parametric Tests for Detecting General Distributional Shifts in Gene Expression

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• Differential expression analysis is vital in genomics.
• Existing methods often overlook variance shifts, which are important in biological processes like aging.
• There's a need for flexible methods to detect both mean and variance changes in gene expression.

Purpose of the Study:

• Introduce QRscore (Quantile Rank Score), a novel method for detecting distributional shifts in gene expression.
• Implement QRscore to specifically detect shifts in both mean and variance.
• Combine parametric and non-parametric approaches for robust differential expression analysis.

Main Methods:

• Extend the Mann-Whitney test into a flexible family of rank-based tests.
• Utilize weights from negative binomial (NB) and zero-inflated negative binomial (ZINB) models.
• Develop QRscore for detecting mean and variance shifts in gene expression data.

Main Results:

• QRscore demonstrates high statistical power and effective false discovery rate (FDR) control in simulations.
• QRscore outperforms existing methods in detecting both variance and mean shifts.
• Analysis of GTEx bulk RNA-seq data revealed numerous differentially dispersed and expressed genes across 33 tissues.
• A significant number of genes showed variance shifts without significant mean shifts.

Conclusions:

• QRscore provides a powerful and flexible approach for differential expression analysis.
• The method enhances the genome bioinformatics toolkit by addressing limitations in detecting variance changes.
• QRscore successfully identifies differentially dispersed genes, offering new insights into gene expression regulation.