MultiSimNeNc: A network representation learning-based module identification method by network embedding and...

Area of Science:

• Computational Biology
• Network Science
• Bioinformatics

Background:

• Gene module identification is crucial for understanding cancer. Existing graph clustering methods often lack accuracy due to limited consideration of network topology.
• Multi-omics data integration offers a comprehensive view for biological network construction.

Purpose of the Study:

• To develop a novel network-based method, MultiSimNeNc, for accurate gene module identification.
• To improve the understanding of cancer's biomolecular mechanisms at the module level.

Main Methods:

• Integrating network representation learning (NRL) with clustering algorithms.
• Utilizing graph convolution (GC) for multi-order network similarity and non-negative matrix factorization (NMF) for node characterization.
• Employing Bayesian Information Criterion (BIC) for module number prediction and Gaussian Mixture Model (GMM) for module identification.

Main Results:

• MultiSimNeNc demonstrated superior performance compared to state-of-the-art algorithms in module identification accuracy.
• The method was validated on biological networks derived from glioblastoma multi-omics data and benchmark networks.

Conclusions:

• MultiSimNeNc is an effective approach for identifying gene modules in biological networks.
• The method enhances the understanding of pathogenesis by providing accurate module-level insights.