A network-assisted co-clustering algorithm to discover cancer subtypes based on gene expression.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Understanding cancer heterogeneity is crucial and relies on identifying distinct subtypes.
• Current methods often use generic clustering on gene expression data, neglecting gene interaction networks.
• Integrating molecular interaction networks can improve the identification of biologically relevant cancer subtypes.

Purpose of the Study:

• To develop a novel method for cancer subtype identification that incorporates gene network information.
• To improve the accuracy and biological relevance of cancer subtype clustering.
• To address the limitations of existing methods in capturing network-level interactions.

Main Methods:

• Developed a network-assisted co-clustering for the identification of cancer subtypes (NCIS) algorithm.
• Incorporated gene network information by assigning weights to genes based on network impact.
• Utilized a semi-nonnegative matrix tri-factorization-based weighted co-clustering algorithm.
• Validated NCIS on simulated data and The Cancer Genome Atlas (TCGA) datasets for Breast Cancer and Glioblastoma Multiforme.

Main Results:

• NCIS effectively groups samples and genes into biologically meaningful clusters.
• Demonstrated improved separation of patient samples into clinically distinct subtypes.
• Achieved higher accuracy and noise tolerance on simulated datasets compared to consensus hierarchical clustering.
• NCIS successfully integrated gene network information for enhanced cancer subtype identification.

Conclusions:

• The weighted co-clustering approach in NCIS offers a novel way to integrate gene network data.
• NCIS can identify cancer subtypes obscured by heterogeneity using high-dimensional gene expression data.
• This tool aids in a more comprehensive understanding of cancer subtypes and molecular perturbations.