Gene set analysis approaches for RNA-seq data: performance evaluation and application guideline.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• RNA sequencing (RNA-seq) is increasingly used for gene expression studies, surpassing microarrays.
• Analyzing RNA-seq data requires methods to interpret biological processes, with gene set analysis (GSA) being a key approach.
• GSA integrates prior biological knowledge through functionally related gene sets.

Purpose of the Study:

• To review and evaluate different statistical gene set analysis (GSA) approaches for RNA-seq data.
• To compare the performance of GSA methods originally developed for microarrays and those designed for RNA-seq.
• To provide guidelines for selecting optimal GSA methods based on experimental conditions.

Main Methods:

• Evaluation of competitive and self-contained GSA approaches.
• Performance assessment using simulated and real RNA-seq datasets.
• Analysis of Type I error rate, power, robustness to sample size/heterogeneity, and bias sensitivity.

Main Results:

• GSA method performance is determined by the statistical hypothesis tested, not its origin (microarray vs. RNA-seq).
• Self-contained GSA methods demonstrate superior power and robustness to sample heterogeneity compared to competitive methods.
• Competitive methods exhibit lower reproducibility and their power is influenced by the balance of up/down-regulated genes.

Conclusions:

• Self-contained GSA methods are recommended for RNA-seq studies due to better performance and reproducibility.
• The choice of GSA method should be guided by the specific experimental design and data characteristics.
• Understanding the statistical underpinnings of GSA methods is crucial for accurate biological interpretation of RNA-seq data.