Robustness of differential gene expression analysis of RNA-seq

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• RNA-sequencing (RNA-seq) is a powerful tool for gene expression analysis but lacks standardization, particularly for Differential Gene Expression (DGE) methods.
• Existing DGE methodologies yield inconsistent results, hindering the clinical adoption of RNA-seq despite ongoing development.
• Robust and reproducible DGE analysis is essential for advancing RNA-seq towards precision medicine applications.

Purpose of the Study:

• To rigorously assess the robustness of five commonly used DGE models (DESeq2, voom + limma, edgeR, EBSeq, NOISeq) against sequencing alterations.
• To compare the performance of these DGE methods across different sample sizes and filtering strategies using unbiased metrics.
• To provide guidance for selecting appropriate DGE methods to improve the standardization of RNA-seq for molecular diagnostics and precision medicine.

Main Methods:

• Investigated five gene-level DGE models using controlled analysis of fixed count matrices from two breast cancer datasets.
• Assessed model robustness under varying sample sizes (full and reduced) and different filtering regimes.
• Employed unbiased metrics including relative False Discovery Rate (FDR), model output concordance, and linear regression analysis of FDR slopes across library sizes.

Main Results:

• DGE model robustness patterns were dataset-agnostic and reliable with sufficient sample sizes.
• The non-parametric NOISeq method exhibited the highest robustness, followed by edgeR, voom + limma, EBSeq, and DESeq2.
• Relative performance rankings remained consistent across different expression levels (high and low).

Conclusions:

• The choice of DGE method significantly impacts the robustness and reproducibility of RNA-seq results.
• NOISeq emerges as a highly robust option for DGE analysis, particularly valuable for clinical settings.
• Standardization of RNA-seq analysis through informed method selection is critical for advancing precision medicine.