Network-based integrative clustering of multiple types of genomic data using non-negative matrix factorization

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Identifying novel molecular subtypes of diseases is crucial for targeted therapies.
• Integrative clustering of multi-omics data reveals complex biological patterns.
• Existing methods may struggle with high-dimensional and noisy biological datasets.

Purpose of the Study:

• To introduce a novel integrative network-based clustering method, non-negative matrix factorization (nNMF).
• To apply nNMF for identifying latent subtype structures in interrelated multi-omics datasets.
• To evaluate nNMF's performance against existing clustering techniques.

Main Methods:

• nNMF utilizes consensus matrices from non-negative matrix factorization (NMF) on individual data types to form patient-sample networks.
• Multiple networks are combined into a comprehensive network structure, optimizing relationship strengths.
• Spectral clustering is applied to the final network to determine cluster groups.

Main Results:

• nNMF demonstrates competitive performance compared to previous methods, particularly excelling when signal-to-noise ratio is low.
• The method was successfully applied to simulated data and The Cancer Genome Atlas datasets (glioblastoma, glioma, head and neck cancer).
• nNMF effectively identifies latent subtype structures in multi-omics data.

Conclusions:

• The proposed nNMF method offers a powerful, non-parametric approach for disease subtyping using multi-omics data.
• nNMF facilitates the discovery of novel molecular subtypes for further association studies.
• The R program for nNMF will be made available to researchers.