A Bayesian extension of the hypergeometric test for functional enrichment analysis

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Functional enrichment analysis interprets high-throughput data, often using Gene Ontology (GO) terms.
• The hypergeometric P-value is a common but limited method for identifying enriched gene sets.
• Existing methods neglect biological dependencies, are subject to size constraints, and have redundant information.

Purpose of the Study:

• To propose a novel Bayesian approach for functional enrichment analysis.
• To address the limitations of the hypergeometric P-value in gene set analysis.
• To incorporate biological dependencies and avoid size constraints in enrichment detection.

Main Methods:

• A Bayesian approach utilizing the non-central hypergeometric model.
• Incorporation of GO dependence structure via priors on non-centrality parameters.
• Inference based on posterior probabilities, avoiding overlapping information and size constraints.

Main Results:

• The proposed method detects moderate, consistent enrichment signals.
• It identifies sets of closely related, biologically meaningful functional terms.
• Simulation studies and a real-data application demonstrate the method's efficacy.

Conclusions:

• The Bayesian approach offers a more robust and biologically relevant alternative to traditional methods.
• This method enhances the interpretation of high-throughput data by identifying coherent functional gene sets.
• It provides theoretical insights into assumption and implementation for various analytical methods.