Multi-View Kernel Learning for Identification of Disease Genes.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Gene expression data and protein-protein interaction (PPI) networks are crucial for understanding gene function.
• Both data types reveal co-functional gene groupings, aligning with multi-view learning principles.
• Existing methods may not fully leverage the complementary information from heterogeneous data sources.

Purpose of the Study:

• To develop a novel multi-view kernel learning algorithm (DiGId) for accurate disease gene identification.
• To learn a consensus kernel that integrates heterogeneous gene expression and PPI network data.
• To identify potential disease genes by uncovering inherent cluster structures within the integrated data.

Main Methods:

• Proposed a novel multi-view kernel learning approach to learn a consensus kernel.
• Incorporated low-rank constraints on the multi-view kernel for effective partitioning into clusters.
• Developed a method to quantify the importance of individual data views (gene expression and PPI networks).

Main Results:

• The DiGId algorithm successfully identified potential disease genes by leveraging a learned joint cluster structure.
• The consensus kernel effectively captured heterogeneous information from individual views.
• Extensive analysis on cancer-related datasets demonstrated the algorithm's effectiveness.

Conclusions:

• Multi-view kernel learning provides a powerful framework for integrating heterogeneous biological data.
• DiGId offers an effective approach for disease gene identification by combining gene expression and PPI network information.
• The proposed method highlights the utility of capturing inherent cluster structures across different data views.