Robustness evaluations of pathway activity inference methods on gene expression data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-throughput technologies generate vast gene expression data.
• Pathway analysis methods estimate pathway activities from this data.
• Existing methods are categorized as non-Topology-Based (non-TB) or Pathway Topology-Based (PTB).

Purpose of the Study:

• To systematically assess and compare the robustness of pathway activity inference methods.
• To evaluate the performance of seven widely used methods across diverse cancer datasets.
• To compare the robustness of pathway activity, risk-active pathways, and genes identified by these methods.

Main Methods:

• Comprehensive robustness evaluations of seven pathway activity inference methods.
• Utilized six cancer gene expression datasets for analysis.
• Two assessments were performed: pathway activity robustness and risk-active pathway/gene robustness.

Main Results:

• Mean reproducibility power generally decreased with increased pathway selections for most methods.
• Entropy-based Directed Random Walk (e-DRW) demonstrated the highest reproducibility power across all datasets.
• No single method provided satisfactory results for risk-active pathway and gene identification across all datasets.

Conclusions:

• Pathway Topology-Based (PTB) methods generally exhibit greater reproducibility power than non-TB methods.
• PTB methods show potential in identifying more informative pathways and genes.
• Further research is needed to improve the robustness of risk-active pathway and gene identification across datasets.