MOCSS: Multi-omics data clustering and cancer subtyping via shared and specific representation learning.

Area of Science:

• Computational biology
• Cancer research
• Bioinformatics

Background:

• Cancer is a complex disease with diverse subtypes.
• Multi-omics data offers comprehensive biological insights for cancer subtyping.
• Current unsupervised methods struggle to capture shared and specific multi-omics information.

Purpose of the Study:

• To develop a novel method for cancer subtyping using multi-omics data.
• To effectively learn both shared and specific representations from multi-omics data.
• To improve the accuracy and comprehensiveness of cancer subtyping.

Main Methods:

• Proposed a novel method based on shared and specific representation learning.
• Utilized autoencoders to extract shared and specific information from each omics dataset.
• Introduced orthogonality constraints to separate shared and specific information and contrastive learning to align shared information.
• Applied clustering on the learned representations for cancer subtyping.

Main Results:

• The proposed method effectively captures shared and specific information from multi-omics data.
• Demonstrated superior performance compared to existing state-of-the-art cancer subtyping methods.
• Successfully utilized multi-omics data for accurate cancer subtyping.

Conclusions:

• The novel shared and specific representation learning method enhances cancer subtyping.
• This approach effectively integrates multi-omics data for biological discovery.
• The method shows significant potential for advancing cancer research and personalized medicine.