Cancer subtype discovery and biomarker identification via a new robust network clustering algorithm

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Understanding cancer phenotypic changes and subtypes is crucial for effective treatment.
• Microarray gene expression data offers insights but can contain outliers, complicating subtype discovery.
• Existing methods may struggle with heterogeneous subtypes and identifying interdependent biomarkers.

Purpose of the Study:

• To develop a robust, outlier-resistant method for discovering cancer subtypes based on gene expression.
• To elucidate cluster-specific network structures and identify biologically relevant cancer biomarkers.
• To address the limitations of current gene expression-based cancer subtyping approaches.

Main Methods:

• Proposed a penalized model-based Student's t clustering with unconstrained covariance (PMT-UC).
• Utilized an adaptive LASSO penalty for robust biomarker identification and network reconstruction.
• Employed the expectation-maximization algorithm with graphical lasso for model fitting.
• Implemented a network-based gene selection criterion to identify subnetworks as biomarkers.

Main Results:

• PMT-UC demonstrated effectiveness and robustness in discovering cancer subtypes on simulated and real datasets.
• The method successfully identified cluster-specific networks and relevant cancer biomarkers.
• Biomarkers were identified as subnetworks, including those with low individual discriminative power but central network roles.
• Learned biologically significant correlations among genes, aiding in understanding underlying pathways.

Conclusions:

• PMT-UC offers a robust and effective approach for cancer subtype discovery from gene expression data.
• The method successfully integrates subtype identification, network analysis, and biomarker discovery.
• PMT-UC can identify novel, biologically significant cancer biomarkers and their network context.