Deep subspace fusion based on integrated self-supervision for cancer subtype identification

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-throughput technologies generate complex multi-omics data essential for cancer research.
• Integrating multi-omics data presents challenges due to data heterogeneity and noise.
• Existing integration methods often rely on unsupervised learning due to limited labeled data.

Purpose of the Study:

• To introduce a novel framework, Deep Subspace Fusion based on Integrated Self-supervision (DSFIS), for cancer subtype recognition.
• To enhance the extraction of valuable information from multi-omics data for improved patient stratification.
• To address limitations of unsupervised methods in multi-omics data integration.

Main Methods:

• Developed DSFIS, a framework utilizing autoencoders with a self-representation layer.
• Integrated self-supervision to guide autoencoders in generating representative sample subspace structures.
• Compared DSFIS against eight state-of-the-art multi-omics data integration approaches.

Main Results:

• DSFIS effectively identified cancer subtypes based on multi-omics data.
• The framework achieved superior performance in survival prognosis analysis compared to other algorithms.
• DSFIS demonstrated enhanced clinical correlation analysis, indicating its potential for personalized medicine.

Conclusions:

• DSFIS offers a powerful approach for integrating multi-omics data for cancer subtype identification.
• The self-supervision mechanism in DSFIS effectively captures patient similarities and differences.
• DSFIS shows significant potential for advancing cancer research and clinical applications through multi-omics data analysis.