Multi-omic and multi-view clustering algorithms: review and cancer benchmark

Area of Science:

• Biomedical research
• Computational biology
• Data science

Background:

• High-throughput technologies generate large biomedical omics datasets.
• Clustering of single omics data is crucial for biological and medical research.
• Advancements enable multi-omic data measurement, offering potential for deeper systems-level insights.

Purpose of the Study:

• To review algorithms for multi-omics clustering.
• To discuss key challenges in applying these algorithms.
• To benchmark leading multi-omics and multi-view clustering methods.

Main Methods:

• Comprehensive review of omics-specific and generic multi-view clustering algorithms.
• Extensive benchmark using TCGA cancer data across ten cancer types.
• Systematic comparison of algorithm performance and suitability.

Main Results:

• Identified key issues in single- versus multi-omics data utilization.
• Evaluated the effectiveness of different clustering strategies.
• Demonstrated the utility of generic multi-view methods and approximated p-values for quality assessment.

Conclusions:

• Multi-omics clustering presents computational and biological challenges.
• Algorithm choice and data integration strategy significantly impact results.
• Systematic benchmarking is essential for advancing multi-omics data analysis.