Hierarchical structure-guided high-dimensional multi-view clustering.

Area of Science:

• Data Science
• Bioinformatics
• Computational Biology

Background:

• Multi-view data clustering integrates information from diverse data aspects.
• Heterogeneous data often exhibits hierarchical structures across different views.
• Existing methods may not fully capture these cross-view hierarchical relationships.

Purpose of the Study:

• To propose a novel high-dimensional multi-view clustering approach that accounts for hierarchical structures across views.
• To address the challenges posed by differing data granularities in multi-view datasets.
• To develop a robust method for uncovering complex data relationships.

Main Methods:

• A novel non-convex optimization problem formulation for hierarchical multi-view clustering.
• Application of the Alternating Direction Method of Multipliers (ADMM) for effective solution.
• Establishment of the statistical properties of the proposed clustering estimator.

Main Results:

• The proposed method demonstrates effectiveness and superiority in simulation studies.
• It successfully identifies a hierarchical clustering structure in lung adenocarcinoma data (histopathology and gene expression).
• The discovered structure significantly differs from those found by alternative clustering approaches.

Conclusions:

• The novel multi-view clustering method accurately captures hierarchical relationships within heterogeneous data.
• This approach offers enhanced insights into complex biological datasets, such as those in lung cancer.
• The method provides a valuable tool for analyzing multi-modal data with inherent hierarchical structures.