Multi-Omics Data Fusion via a Joint Kernel Learning Model for Cancer Subtype Discovery and Essential Gene...

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Cancer heterogeneity presents challenges in diagnosis and treatment.
• Identifying cancer subtypes is crucial for personalized medicine.
• Genetic and epigenetic differences exist among cancer subtypes.

Purpose of the Study:

• To develop a computational method for identifying cancer subtypes.
• To analyze multi-omics data for subtype discovery.
• To provide a reference for clinical diagnosis and treatment.

Main Methods:

• Collected seven cancer datasets including gene expression, isoform expression, and DNA methylation data.
• Applied kernel principal component analysis (PCA) and Gaussian kernel function for feature extraction and kernel matrix fusion.
• Utilized spectral clustering for cancer subtype identification.

Main Results:

• Successfully clustered cancer subtypes using fused kernel matrices.
• Validated clustering performance using Cox regression, survival analysis, Rand index (RI), and adjusted RI (ARI).
• Identified differentially expressed genes in lung and liver cancer subtypes, including specific gene markers.

Conclusions:

• The proposed method effectively identifies cancer subtypes from multi-omics data.
• Differential gene expression analysis highlights subtype-specific markers for lung and liver cancers.
• This approach supports personalized cancer treatment strategies.