Selective Update of Relevant Eigenspaces for Integrative Clustering of Multimodal Data

Area of Science:

• Bioinformatics
• Computational Biology
• Cancer Research

Background:

• Cancer subtype discovery from multimodal omic data faces challenges with irrelevant or non-homogeneous information across modalities.
• High-dimensional omics data increases computational costs for sample clustering.

Purpose of the Study:

• To develop a novel algorithm for extracting a low-rank joint subspace from multimodal omic data for improved cancer subtype discovery.
• To address the computational expense and information heterogeneity issues in integrative cancer data analysis.

Main Methods:

• A novel algorithm that evaluates and selects relevant omics modalities to construct a joint subspace.
• Formulation of incremental singular value decomposition for multimodal data.
• Development of quantitative indices to assess subspace construction accuracy compared to Principal Component Analysis (PCA).

Main Results:

• The proposed algorithm efficiently constructs a joint subspace by integrating low-rank subspaces of individual modalities.
• Computational efficiency is demonstrated to be superior to PCA on integrated data.
• The joint subspace constructed by the algorithm shows improved clustering efficacy over existing methods on real cancer datasets.

Conclusions:

• The novel algorithm effectively addresses challenges in multimodal omic data integration for cancer subtype discovery.
• The method offers a computationally efficient and accurate approach for identifying cancer subtypes.
• This approach enhances the utility of multimodal data in cancer research and clinical applications.