Choice of library size normalization and statistical methods for differential gene expression analysis in balanced...

Area of Science:

• Bioinformatics
• Genomics
• Statistical Analysis

Background:

• High-throughput RNA sequencing (RNA-seq) is crucial in molecular biology.
• Lack of standardized normalization and statistical methods hinders RNA-seq data interpretation and reproducibility.
• This study addresses the need for optimal analytical methods in RNA-seq data analysis.

Purpose of the Study:

• To compare the performance of the UQ-pgQ2 normalization method against RLE, TMM, and UQ methods.
• To evaluate different normalization and statistical test combinations for differential gene expression analysis.
• To assess the impact of sample size and sequencing read depth on analysis outcomes.

Main Methods:

• Comparison of UQ-pgQ2, RLE, TMM, and UQ normalization methods.
• Evaluation of Wald test (DESeq2) and exact/QL F-Test (edgeR) statistical tests.
• Analysis using MAQC RNA-seq datasets and simulated data with varying sample sizes and read depths.

Main Results:

• UQ-pgQ2 normalization with an exact test demonstrated superior power and specificity for small sample replicates.
• For larger sample sizes, the Wald test outperformed the exact test, while the QL F-test showed the best performance across tested sample sizes (5, 10, 15).
• RLE, TMM, and UQ methods showed similar performance, and read depth had minimal impact on differential gene expression analysis.

Conclusions:

• UQ-pgQ2 combined with exact or QL F-tests is recommended for controlling false positives, especially with small sample sizes.
• For large sample sizes, UQ-pgQ2 with a QL F-test is preferable for type I error control in intra-group analysis.
• The choice of normalization and statistical method significantly impacts RNA-seq differential expression analysis results, with sample size being a key factor.