Multi-kernel clustering with tensor fusion on Grassmann manifold for high-dimensional genomic data.

Area of Science:

• Bioinformatics
• Machine Learning
• Data Science

Background:

• High-dimensional genomic data presents significant challenges for traditional clustering algorithms due to noise and complexity.
• Existing multi-kernel clustering methods improve affinity matrix quality but struggle with error propagation and redundancy in high-dimensional settings.
• Current multi-kernel fusion strategies lack feasibility for effectively integrating diverse kernel information.

Purpose of the Study:

• To develop an advanced multi-kernel clustering method capable of overcoming the limitations of existing approaches in high-dimensional genomic data analysis.
• To introduce a novel tensor fusion strategy on Grassmann manifolds for robust kernel integration.
• To enhance clustering performance by maximizing consensus among base kernels while mitigating noise and redundancy.

Main Methods:

• Proposed a Multi-Kernel Clustering method with Tensor fusion on Grassmann manifolds (MKCTM).
• Employed tensor low-rank constraints to maximize clustering consensus and eliminate noise/redundancy from base kernels.
• Developed a unified optimization model integrating tensor learning and fusion, solved by an effective iterative algorithm.
• Fused learned base kernels on the Grassmann manifold to generate a final consensus matrix for clustering.

Main Results:

• MKCTM demonstrated superior performance compared to 12 popular baseline clustering methods across ten diverse datasets.
• The tensor fusion approach effectively reduced noise and redundancy, leading to improved clustering accuracy.
• The method successfully integrated information from multiple base kernels through Grassmann manifold fusion.

Conclusions:

• MKCTM offers a robust and effective solution for clustering high-dimensional genomic data, outperforming existing methods.
• The proposed tensor fusion strategy on Grassmann manifolds is a significant advancement in multi-kernel learning.
• The developed iterative optimization algorithm efficiently solves the integrated tensor learning and fusion model.