Integrative clustering of multi-level 'omic data based on non-negative matrix factorization algorithm.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-throughput omics data integration offers insights into disease molecular underpinnings.
• Cancer research increasingly utilizes multi-omics for identifying molecular subtypes impacting treatment.
• Tumors with similar morphology can display distinct molecular profiles, necessitating advanced classification methods.

Purpose of the Study:

• To develop an integrative approach for disease subtype classification using multiple omics data.
• To introduce intNMF, a non-negative matrix factorization-based method for integrative clustering.
• To enable the identification of novel molecular subtypes with potential clinical relevance.

Main Methods:

• Developed intNMF, an integrative clustering algorithm based on non-negative matrix factorization.
• Applied intNMF to simultaneously cluster multiple high-dimensional molecular datasets (DNA methylation, copy number alteration, gene/protein expression).
• Utilized data from The Cancer Genome Atlas (TCGA) for validation, including simulated and real patient data.

Main Results:

• intNMF effectively performs integrative clustering of diverse omics data within a single analysis.
• The method leverages information across multiple biological levels from the same individuals.
• Demonstrated advantages over model-based clustering by not requiring specific distributional assumptions for the data.

Conclusions:

• intNMF provides a robust framework for integrative disease subtype discovery.
• The approach facilitates a comprehensive understanding of molecular heterogeneity in diseases like cancer.
• This method has the potential to refine disease classification and inform personalized treatment strategies.