COPS: A novel platform for multi-omic disease subtype discovery via robust multi-objective evaluation of clustering...

Area of Science:

• Computational biology
• Bioinformatics
• Cancer research

Background:

• Multi-view clustering for complex disease subtyping often lacks stability assessment and prognostic relevance evaluation.
• Existing frameworks fail to compare data-driven versus pathway-driven clustering approaches, creating a methodological gap.

Purpose of the Study:

• To introduce the COPS R-package for robust evaluation of single and multi-omics clustering results.
• To enable comparison between data-driven and pathway-driven clustering methods.
• To assess clustering stability and prognostic relevance for complex disease subtyping.

Main Methods:

• Developed the COPS R-package integrating similarity networks, kernel methods, dimensionality reduction, and pathway knowledge.
• Applied the framework to multi-omics data (mRNA, CNV, miRNA, DNA methylation) across seven cancer types.
• Utilized cross-fold validation, Adjusted Rand Index (ARI), Cox regression for survival analysis, and Pareto efficiency for multi-objective evaluation.

Main Results:

• COPS enables robust evaluation and comparison of diverse clustering methods.
• Affinity Network Fusion, Integrative Non-negative Matrix Factorization, and Multiple Kernel K-Means demonstrated high stability and effectiveness.
• Identified patient subgroups with significantly different survival outcomes in multiple cancer types.

Conclusions:

• Multi-view clustering requires multi-criteria assessment, including stability and prognostic relevance.
• The COPS package provides a unified framework for selecting optimal clustering approaches for disease subtype discovery.
• Data- and knowledge-driven clustering methods can be effectively compared to reveal biologically meaningful subtypes.