Variance component score test for time-course gene set analysis of longitudinal RNA-seq data

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• RNA-sequencing (RNA-seq) is replacing microarrays for gene expression measurement.
• RNA-seq data, measured as counts, require adapted statistical tools due to inherent heteroscedasticity.
• Nonparametric regression has been proposed to model RNA-seq counts as continuous variables.

Purpose of the Study:

• To propose tcgsaseq, a principled, model-free, and efficient method for detecting longitudinal changes in predefined RNA-seq gene sets.
• To identify gene sets with expression that varies over time using a novel variance component score test.

Main Methods:

• tcgsaseq employs a variance component score test that accounts for covariates and heteroscedasticity without assuming parametric distributions for transformed counts.
• The method is computationally efficient, featuring a simple test statistic form and limiting distribution.
• A permutation-based version is available for small sample sizes.

Main Results:

• tcgsaseq demonstrates excellent statistical properties, including enhanced stability and power compared to ROAST, edgeR, and DESeq2.
• Existing methods like ROAST, edgeR, and DESeq2 may fail to control type I error rates in certain realistic scenarios.
• The method was validated using simulated data and two real biological datasets.

Conclusions:

• tcgsaseq provides a robust and powerful tool for analyzing longitudinal RNA-seq data.
• The proposed method addresses limitations of current state-of-the-art techniques in gene set analysis.
• The tcgsaseq method is publicly available as an R package for the research community.