Subtype-WESLR: identifying cancer subtype with weighted ensemble sparse latent representation of multi-view data

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Cancer subtyping is crucial for personalized medicine in heterogeneous cancer patients.
• High-throughput technologies generate multi-omics data for cancer subtyping.
• Existing computational methods yield diverse subtypes even from the same data, presenting a challenge for integration.

Purpose of the Study:

• To develop a novel computational method for accurate and reliable cancer subtyping.
• To effectively integrate information from distinct subtypes identified by various methods.
• To improve the precision of cancer subtyping using heterogeneous omics data.

Main Methods:

• A weighted ensemble sparse latent representation (subtype-WESLR) method was proposed.
• It fuses base clustering from distinct methods as prior knowledge.
• Projects sample feature profiles to a common latent subspace, maintaining local structure and consistency.

Main Results:

• subtype-WESLR demonstrated superior performance compared to competing methods.
• Experiments were conducted on synthetic and eight public multi-view datasets from The Cancer Genome Atlas.
• The method effectively utilizes integrated base clustering for more precise subtype identification.

Conclusions:

• The proposed subtype-WESLR method offers a robust approach for cancer subtyping.
• Integrating diverse subtyping results enhances accuracy and reliability.
• This advancement holds potential for refining personalized cancer treatment strategies.