Simultaneous clustering of multiview biomedical data using manifold optimization.

Area of Science:

• Computational biology
• Data science
• Bioinformatics

Background:

• Multiview clustering aims to find patterns across multiple data sources.
• Existing methods often focus on either consistent or differential clusters, not both simultaneously.
• Identifying both types of clusters is crucial for a comprehensive understanding of complex datasets.

Purpose of the Study:

• To develop a novel method for simultaneous clustering of multiview data.
• To address the limitation of existing methods that only identify either consistent or differential clusters.
• To enable a more complete analysis of multiview datasets by capturing both shared and unique cluster structures.

Main Methods:

• A manifold optimization approach is proposed for simultaneous multiview clustering.
• The binary optimization model is relaxed to a real-valued problem on the Stiefel manifold.
• A line-search algorithm on the manifold is employed to solve the optimization problem.

Main Results:

• The proposed method demonstrates competitive performance against state-of-the-art algorithms for consistent clusters.
• The method significantly outperforms existing approaches when differential clusters are present.
• Application to TCGA cancer datasets successfully identified both consistent and differential clusters, revealing biologically relevant insights and potential mechanisms for cancer development.

Conclusions:

• The developed manifold optimization method effectively performs simultaneous multiview clustering.
• This approach provides a powerful tool for cancer stratification and differential cluster identification.
• The findings highlight the utility of identifying both consistent and differential clusters for understanding complex biological systems and disease mechanisms.