Layer-Specific Modules Detection in Cancer Multi-Layer Networks.

Area of Science:

• Network Science
• Computational Biology
• Data Mining

Background:

• Multi-layer networks model complex systems with diverse interactions, surpassing traditional single-layer approaches.
• Clustering multi-layer networks is challenging due to the need to balance intra-cluster connectivity and inter-layer relationships.
• Existing layer-specific clustering algorithms suffer from low accuracy and network perturbation sensitivity.

Purpose of the Study:

• To propose a novel algorithm, Layer-Specific Nonnegative Matrix Factorization (LSNMF), for identifying layer-specific modules in multi-layer networks.
• To address the limitations of current algorithms in accuracy and sensitivity.
• To enhance the characterization and modeling of layer-specific module structures.

Main Methods:

• Nonnegative Matrix Factorization (NMF) is employed to extract vertex features from multi-layer networks.
• Vertex features are decomposed into common and specific components.
• An orthogonality constraint is applied to specific components to ensure feature distinctiveness.

Main Results:

• LSNMF significantly outperforms state-of-the-art methods across various evaluation metrics.
• The algorithm efficiently extracts stage-specific modules.
• Identified modules show enrichment of known biological functions and correlation with patient survival time.

Conclusions:

• LSNMF offers a robust and accurate approach for layer-specific module detection in multi-layer networks.
• The method's ability to capture specific vertex features enhances structural characterization.
• LSNMF provides valuable biological insights, linking network modules to functional enrichment and clinical outcomes.