clust.perturb

clust.perturb evaluates the robustness of graph-based clusters in protein-protein interaction networks by measuring cluster stability under random edge perturbations to identify reproducible protein complexes.

Key Features:

• Robustness Testing: Introduces random perturbations by rewiring a portion of network edges to test the robustness of graph-based clustering algorithms on interactomes.
• Quantification Using Jaccard Index: Quantifies cluster robustness using the maximum Jaccard index to measure similarity between clusters before and after perturbation.
• Predictive Power for Real-World Stability: Identifies clusters whose reproducibility under perturbation predicts their likelihood of consistent reclustering across different experiments.
• Application Across Domains: Applies to protein-protein interaction networks (interactomes) and to a range of graph-based clustering algorithms used in other domains.

Scientific Applications:

• Protein Complex Identification: Distinguishes stable multi-member protein complexes from spurious associations within interactomes.
• Algorithm Evaluation: Provides a framework for evaluating the robustness of different clustering algorithms against network noise.

Methodology:

Introduce controlled perturbations by randomly rewiring network edges, reassess cluster configurations, and quantify cluster similarity using the maximum Jaccard index.