Toward a gold standard for benchmarking gene set enrichment analysis

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene set enrichment analysis (GSEA) is crucial for high-throughput gene expression data.
• Current GSEA method evaluations are limited due to a lack of gold standards.
• Existing assessments often rely on selected datasets and qualitative biological reasoning.

Purpose of the Study:

• To establish a reproducible benchmarking framework for GSEA methods.
• To evaluate GSEA methods based on applicability, gene set prioritization, and detection of relevant biological processes.
• To provide practical guidance for GSEA in the context of RNA-seq data analysis.

Main Methods:

• Developed an extensible framework for reproducible GSEA method benchmarking.
• Created a compendium of 75 expression datasets across 42 human diseases (microarray and RNA-seq).
• Associated each dataset with curated Gene Ontology (GO)/KEGG relevance rankings.

Main Results:

• Assessed 10 major GSEA methods, revealing significant differences in runtime and RNA-seq applicability.
• Identified variations in the fraction of enriched gene sets based on the null hypothesis.
• Quantified the recovery of predefined relevance rankings, offering insights into method performance.
• Provided recommendations for applying microarray-based methods to RNA-seq data.

Conclusions:

• The developed framework enables robust and reproducible GSEA method evaluation.
• Practical recommendations are offered for RNA-seq data analysis and interpreting GSEA results.
• Guidance is provided on selecting methods for prioritizing gene sets with high biological relevance.