Comparative evaluation of gene set analysis approaches for RNA-Seq data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• RNA-Seq is now the standard for high-throughput gene expression studies, surpassing microarrays.
• A primary application of RNA-Seq is identifying differentially expressed genes between conditions.
• Gene Set Analysis (GSA) is crucial for interpreting RNA-Seq results, but its limitations with RNA-Seq data are unclear.

Purpose of the Study:

• To evaluate existing Gene Set Analysis (GSA) methods for RNA-Seq data.
• To compare the performance of multivariate and gene-level GSA approaches.
• To identify optimal GSA strategies for accurate pathway analysis in RNA-Seq.

Main Methods:

• Thorough evaluation of popular multivariate and gene-level GSA approaches.
• Utilized simulated and real RNA-Seq datasets for analysis.
• Applied multivariate non-parametric tests and univariate tests for gene-level P-values.

Main Results:

• Multivariate GSA test performance (Type I error, power) is independent of normalization methods.
• Multivariate GSA tests are unbiased regarding pathway size, gene characteristics, or differential expression percentages.
• Gene-level GSA tests' performance is significantly impacted by P-value combination methods, introducing biases.

Conclusions:

• Self-contained non-parametric multivariate tests are crucial for reliable differential pathway detection in RNA-Seq.
• Gene-level GSA tests should be avoided due to high Type I error rates and inherent biases.
• The study highlights the need for robust GSA methods tailored to RNA-Seq data characteristics.