SeOMLR: one-step multi-view latent representation with self-weighted ensemble learning for multi-omics cancer...

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Accurate cancer subtyping is crucial for effective treatment due to molecular heterogeneity.
• Existing multi-omics integration methods often prioritize cross-omics consistency over intra-omics specificity.
• Traditional two-step clustering methods can lead to information loss and unstable cancer subtypes.

Purpose of the Study:

• To develop a novel method, seOMLR, for improved cancer subtyping using multi-omics data.
• To address limitations of existing methods by enhancing both specificity and consistency in data integration.
• To provide a robust computational framework for cancer subtyping research.

Main Methods:

• Proposed seOMLR, a one-step multi-view latent representation method with self-weighted ensemble learning.
• Employed relaxed exclusivity constraints and consistency regularization within a sparse low-rank self-representation framework.
• Utilized spectral rotation for discrete clustering structure extraction and joint iterative optimization for fusion and clustering.

Main Results:

• seOMLR effectively balances specificity and consistency in multi-omics data integration for cancer subtyping.
• The self-weighted ensemble strategy adaptively incorporates prior subtyping information, enhancing learning.
• Experiments on simulated and TCGA cancer datasets demonstrated superior performance compared to existing methods.

Conclusions:

• seOMLR offers an efficient and accurate approach to multi-omics data fusion for cancer subtyping.
• The method overcomes information loss and instability issues associated with traditional two-step clustering.
• seOMLR provides valuable computational support for advancing cancer subtyping research.