Survival-Informed Multi-Omics Kernel Fusion for Cancer Subtyping.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Cancer molecular heterogeneity complicates precise subtyping and personalized treatment.
• Existing multi-omics integration methods often lack clinical relevance and kernel redundancy, limiting prognostic utility.

Purpose of the Study:

• To introduce Survival-Informed Multi-omics Kernel Fusion (SIMKF), a framework designed to uncover clinically distinct cancer subtypes.
• To address limitations in current multi-omics integration by incorporating survival information and improving prognostic accuracy.

Main Methods:

• SIMKF framework combines survival-guided kernel selection with distribution-aware fusion.
• Utilizes adaptive weighting based on maximum mean discrepancy and spectral clustering.
• Integrates survival information with multi-omics data for enhanced cancer subtyping.

Main Results:

• SIMKF significantly outperforms existing techniques across five TCGA cancer datasets.
• Identified five distinct breast cancer subtypes with significant survival differences.
• Revealed a nonlinear relationship between methylation levels and survival outcomes in breast cancer.

Conclusions:

• SIMKF provides an automated, tuning-free tool for precision oncology.
• The framework uncovers prognostic biological mechanisms and translates into clinically applicable subtyping models.
• SIMKF enhances the prognostic utility of multi-omics data for personalized cancer therapy.