On the limits of active module identification.

Area of Science:

• Network and systems medicine
• Computational biology
• Bioinformatics

Background:

• Active module identification methods (AMIMs) integrate network analysis with molecular data, typically using protein-protein interaction (PPI) networks.
• Existing PPI networks face challenges including small diameters and biases, potentially limiting their utility in disease mechanism discovery.

Purpose of the Study:

• To evaluate whether widely used AMIMs genuinely benefit from established PPI networks.
• To assess the performance of AMIMs on PPI networks compared to random networks.

Main Methods:

• Systematic analysis of commonly used AMIMs.
• Comparison of module identification results on real PPI networks versus random networks with identical node degrees.

Main Results:

• Most tested AMIMs did not yield more biologically relevant disease modules on PPI networks than on random networks.
• AMIMs primarily rely on node degrees, largely ignoring the biological information within PPI network edges.
• The exception was the DOMINO algorithm, which showed improved performance.

Conclusions:

• Current AMIMs often fail to leverage the full potential of PPI networks due to inherent biases.
• There is a critical need for new algorithms that address degree bias and utilize context-specific networks for more effective disease module identification.