Kernel Fusion Method for Detecting Cancer Subtypes via Selecting Relevant Expression Data.

Area of Science:

• Bioinformatics
• Genomics
• Cancer Research

Background:

• Cancer is a heterogeneous disease with diverse subtypes.
• Early diagnosis of cancer subtypes is crucial for effective patient treatment.
• The Cancer Genome Atlas (TCGA) provides multi-omics data for cancer research.

Purpose of the Study:

• To develop a novel method for uncovering cancer subtypes.
• To integrate gene, miRNA, and isoform expression data for subtype discovery.
• To identify key genes associated with specific cancer subtypes.

Main Methods:

• Extracted a novel multi-omics dataset from TCGA.
• Utilized LASSO regression for gene selection to reduce noise.
• Constructed similarity kernels using Chebyshev distance for each data type.
• Fused similarity matrices using Similarity Kernel Fusion (SKF).
• Applied spectral clustering to the fused similarity matrix.

Main Results:

• The proposed method demonstrated superior performance (better P-value in Cox model) compared to existing methods on multiple cancer datasets.
• Identified key genes highly expressed in specific cancer subtypes, including HSPA2A, RNASE1, CLIC6, IFITM1 for breast cancer, and C4BPA, SESN3, IRS1 for lung cancer.
• Generated distinct survival curves, highlighting the prognostic significance of identified genes.

Conclusions:

• The developed method effectively identifies cancer subtypes by integrating multi-omics data.
• The identified key genes can serve as potential biomarkers for early diagnosis and targeted therapy.
• This approach advances our understanding of cancer heterogeneity and aids in personalized medicine.