Benchmarking enrichment analysis methods with the disease pathway network.

Area of Science:

• Genomics
• Bioinformatics
• Systems Biology

Background:

• Enrichment analysis (EA) is crucial for interpreting genome-scale experiments, but selecting the optimal method remains challenging.
• Previous benchmarks suffer from difficulties in assigning true pathways and use limited evaluation metrics.
• A generalized benchmark is needed to compare the diverse range of existing EA methods.

Purpose of the Study:

• To provide a generalized benchmark for evaluating widely used enrichment analysis methods.
• To introduce a novel evaluation approach combining sensitivity and specificity for balanced assessment.
• To identify superior EA methods and explore biases in current approaches.

Main Methods:

• Developed a generalized benchmark using 82 curated gene expression datasets across 26 diseases (DNA microarray and RNA-Seq).
• Introduced the Disease Pathway Network to link related Kyoto Encyclopedia of Genes and Genomes pathways for enhanced evaluation.
• Implemented a novel evaluation metric combining sensitivity and specificity, and analyzed null hypothesis bias using randomized datasets.

Main Results:

• Network Enrichment Analysis methods demonstrated superior performance compared to overlap-based methods.
• Most evaluated EA methods exhibited null hypothesis bias, producing skewed P-values.
• The benchmark provides a more sensitive and specific evaluation of pathway enrichment.

Conclusions:

• Network Enrichment Analysis represents a more effective approach for functional interpretation of gene expression data.
• Current EA methods often suffer from biases, necessitating careful interpretation of results.
• This generalized benchmark and evaluation framework advance the field of bioinformatics and genomics.