Clustering of Cancer Attributed Networks by Dynamically and Jointly Factorizing Multi-Layer Graphs

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Integrating omics data (interactome, transcriptome) is crucial for understanding cancer mechanisms.
• Current integrative analysis algorithms struggle with data complexity and heterogeneity, leading to suboptimal performance.

Purpose of the Study:

• To propose an effective and efficient algorithm, NMF-DEC, for identifying clusters by integrating interactome and transcriptome data.
• To address the challenges of data heterogeneity in omics data analysis.

Main Methods:

• Transformed integrative omics data analysis into clustering of attributed networks by treating gene expression profiles as vertex attributes.
• Constructed a gene attribute similarity network and framed it as a multi-layer network module detection problem.
• Developed NMF-DEC, which jointly factorizes similarity and interaction networks, dynamically updating the interaction network and incorporating attribute information.

Main Results:

• NMF-DEC demonstrated higher accuracy on social networks compared to baseline methods.
• The algorithm showed superior performance on cancer attributed networks, indicating its effectiveness for omics data integration.
• The proposed method provides a robust strategy for characterizing module structures in attributed networks.

Conclusions:

• NMF-DEC offers a superior approach for the integrative analysis of omics data, particularly in cancer research.
• The algorithm's ability to handle data heterogeneity and complexity enhances the identification of biological patterns.
• This method advances the field of bioinformatics by providing a more accurate and efficient tool for omics data integration.