QRscore

QRscore provides a flexible family of non-parametric two-sample and K-sample tests for detecting distributional shifts in gene expression, including shifts in mean and variance, by extending the Mann-Whitney test into rank-based tests with weights derived from negative binomial and zero-inflated negative binomial models.

Key Features:

• Flexible rank-based testing framework: Extends the Mann-Whitney test into a family of quantile rank score tests capable of detecting general distributional shifts beyond centrality changes.
• Mean and variance shift detection: Implements tests specifically designed to detect both shifts in mean and shifts in variance in gene expression data.
• Parametric-non-parametric hybrid weighting: Uses weights designed from negative binomial and zero-inflated negative binomial models to combine strengths of parametric and non-parametric approaches.
• False discovery rate control: Achieves high statistical power while controlling the false discovery rate in differential expression analysis.
• Multi-sample support: Provides both two-sample tests and K-sample tests for comparing multiple groups.

Scientific Applications:

• Differential gene expression analysis: Identifies differentially expressed genes across conditions in bulk RNA-seq data, including genes with significant variance shifts but non-significant mean shifts.
• Differential dispersion analysis: Detects differentially dispersed genes characterized by variance shifts in gene expression.
• Tissue comparison studies: Enables identification of differentially expressed and differentially dispersed genes across multiple tissues, as demonstrated with GTEx project data spanning 33 tissues.
• Aging and cellular signaling research: Supports detection of variance shifts documented in studies of cellular signaling pathways and aging processes.

Methodology:

QRscore is based on theoretical work around non-parametric tests, spacing statistics, and local asymptotic normality. The method extends the Mann-Whitney test into a flexible family of rank-based tests using quantile rank scores with weights derived from negative binomial and zero-inflated negative binomial models.