A comparison of methods for differential expression analysis of RNA-seq data

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Differential gene expression analysis is key to understanding phenotypic variation.
• DNA microarrays were traditionally used, but high-throughput sequencing of cDNA (RNA-seq) is emerging as a powerful alternative.
• The decreasing cost of sequencing is driving increased adoption of RNA-seq for gene expression studies.

Purpose of the Study:

• To compare eleven software packages for differential gene expression analysis of RNA-seq data.
• To evaluate the performance of these methods using both simulated and real RNA-seq datasets.
• To identify optimal methods for analyzing RNA-seq data, considering varying experimental conditions.

Main Methods:

• An extensive comparison of eleven freely available R-based software packages was conducted.
• Methods were evaluated using a matrix of read counts per genomic feature across multiple samples.
• Both simulated and real RNA-seq data were utilized for method assessment.

Main Results:

• All evaluated methods encountered challenges with very small sample sizes, necessitating cautious interpretation of results.
• For experiments with larger sample sizes, methods incorporating a variance-stabilizing transformation with the 'limma' approach demonstrated strong performance.
• The nonparametric SAMseq method also showed favorable results across various conditions for larger sample sizes.

Conclusions:

• Small sample sizes in RNA-seq experiments pose significant challenges for all current differential expression analysis methods.
• For robust differential expression analysis with adequate sample sizes, combining variance stabilization with 'limma' or using SAMseq are recommended.
• Future research should focus on developing methods that are more robust to small sample sizes in RNA-seq studies.