PCA-constrained multi-core matrix fusion network: A novel approach for cancer subtype identification

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Cancer subtyping categorizes cancers based on molecular, clinical, and histological features for personalized medicine.
• Network fusion methods are emerging for cancer subtype identification.
• Existing methods often rely on single core matrices, limiting comprehensive similarity capture.

Purpose of the Study:

• To propose a novel cancer subtype recognition method, PCA-constrained multi-core matrix fusion network (PCA-MM-FN).
• To address the limitations of single-core matrix fusion in capturing comprehensive similarity for cancer subtyping.

Main Methods:

• Obtain three core matrices using distinct methods.
• Project core matrices into a shared subspace via Principal Component Analysis (PCA).
• Perform weighted network fusion followed by spectral clustering on the fused network.

Main Results:

• PCA-MM-FN demonstrated superior accuracy in cancer subtype identification.
• Experiments were conducted on TCGA and multi-omics benchmark datasets using mRNA expression, DNA methylation, and miRNA expression.
• The proposed method outperforms existing approaches in identifying cancer subtypes.

Conclusions:

• The PCA-MM-FN method offers enhanced accuracy for cancer subtype recognition.
• Multi-core matrix fusion provides a more comprehensive approach to capturing similarity.
• This advancement supports more precise clinical practice and personalized cancer treatment strategies.