Cancer Subtype Recognition Based on Laplacian Rank Constrained Multiview Clustering

Area of Science:

• Bioinformatics and computational biology
• Genomics and cancer research
• Machine learning for healthcare

Background:

• Identifying cancer subtypes is crucial for effective treatment strategies.
• Existing multiview clustering methods for cancer subtyping have limitations in data fusion and require multiple steps.
• There is a need for integrated, one-step approaches to analyze multigenomic data for cancer recognition.

Purpose of the Study:

• To develop a novel one-step multiview clustering algorithm for cancer subtype recognition.
• To address the limitations of existing methods by incorporating differential data contributions and a unified clustering process.
• To improve the accuracy and efficiency of cancer subtyping using multigenomic data.

Main Methods:

• Proposed Laplacian Rank Constrained Multiview Clustering (LRCMC), a graph learning-based framework.
• Constructed individual graphs for each data type using affinity matrices.
• Integrated graphs by weighting their contributions and merging them into a consensus graph.
• Incorporated adaptive neighbors and rank constraints on the Laplacian matrix for robust clustering.

Main Results:

• LRCMC demonstrated superior performance in cancer subtype recognition compared to state-of-the-art methods.
• Experiments on benchmark and The Cancer Genome Atlas (TCGA) datasets validated the algorithm's effectiveness.
• The one-step approach simplified the cancer subtyping process and improved label generation.

Conclusions:

• LRCMC offers an effective and efficient one-step solution for cancer subtype recognition using multigenomic data.
• The graph learning framework with rank constraints enhances the integration of diverse biological data.
• This method holds promise for advancing precision oncology through improved cancer subtyping.