Measuring rank robustness in scored protein interaction networks.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Protein interaction networks (PINs) are built using confidence scores from experimental evidence.
• Thresholding these scores creates networks for analyzing biological motifs and nodes.
• Node metric analysis can be sensitive to the chosen score threshold.

Purpose of the Study:

• To evaluate the robustness of node metrics to score threshold variations in PINs.
• To identify node metrics that yield consistent rankings across different thresholds.
• To ensure reproducible biological signal extraction from protein interaction data.

Main Methods:

• Proposed three measures: rank continuity, identifiability, and instability.
• Applied measures to twenty-five node metrics across various protein interaction networks.
• Analyzed robustness across different species, data sources, and synthetic networks.

Main Results:

• Identified four robust node metrics: step-1 ego network edges, leave-one-out differences in average redundancy, average step-1 ego network edges, and natural connectivity.
• Demonstrated good agreement of robustness measures across diverse PINs.
• Found that robustness is context-specific, depending on network topology and score distribution.

Conclusions:

• Reproducible PIN analysis requires node metrics robust to confidence score threshold changes.
• Certain node metrics exhibit significant robustness, while others do not.
• Identified robust metrics offer reliable analysis across different databases and scoring procedures.