Comparing Bayesian-Based Reconstruction Strategies in Topology-Based Pathway Enrichment Analysis

Area of Science:

• Bioinformatics
• Systems Biology
• Computational Biology

Background:

• High-throughput omics technologies generate vast genomic data.
• Pathway enrichment analysis (PEA) extracts biological insights.
• Topology-based pathway analysis (TPA) enhances PEA by incorporating pathway structure.

Purpose of the Study:

• Compare Bayesian network (BN) reconstruction strategies (BPA, BNrich, PROPS, Clipper, Ensemble) for TPA.
• Evaluate their performance in pathway enrichment analysis and tumor/non-tumor classification using gene expression data.
• Analyze differences in BN structure reconstruction and cyclic structure removal strategies.

Main Methods:

• Applied five BN reconstruction methods (BPA, BNrich, PROPS, Clipper, Ensemble) to gene expression data.
• Performed pathway enrichment analysis (PEA) and tumor vs. non-tumor classification.
• Examined reconstructed Bayesian network (BN) structures, particularly the JAK-STAT pathway, and cyclic structure removal strategies.

Main Results:

• All methods achieved high accuracy (AUC > 0.95) in distinguishing tumor from non-tumor samples.
• Pathway rankings varied significantly due to differences in BN structures arising from distinct cyclic structure removal strategies.
• BNrich, utilizing expert intervention for loop removal, generated BNs that best align with biological knowledge.

Conclusions:

• Different BN reconstruction strategies yield distinct pathway rankings despite similar classification performance.
• BNrich's approach to handling cyclic structures offers superior biological plausibility.
• This comparative analysis provides guidance for selecting appropriate TPA methods for specific biological data analysis tasks.