MOTCS: A Cancer Subtype Classification and Key Biomarker Recognition Model Based on Multi-Omics Data Integration of...

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Cancer exhibits high heterogeneity, necessitating personalized treatment strategies.
• Multi-omics data (genome, transcriptome, epigenome) integration offers a comprehensive view of tumor molecular features.
• Driver genes are crucial for tumorigenesis and improving model interpretability when combined with multi-omics data.

Purpose of the Study:

• To develop an advanced computational model for precise cancer subtype classification.
• To investigate the impact of integrating driver gene information with multi-omics data for improved classification.
• To enhance the interpretability of cancer subtype classification models.

Main Methods:

• Feature extraction and selection for driver and non-driver genes across multiple omics datasets.
• Utilizing transformer encoders for feature representation learning of both gene types.
• Concatenating omics-specific feature embeddings and employing MLP classifiers.
• Integrating cross-omics feature representations via view association discovery networks.

Main Results:

• The proposed MOTCS model demonstrated superior performance across four independent cancer datasets compared to existing methods.
• Incorporating driver gene features significantly improved the classification accuracy of MOTCS.
• MOTCS effectively achieved precise cancer subtype classification by leveraging mined feature embeddings.

Conclusions:

• MOTCS provides a robust framework for cancer subtype classification by integrating multi-omics data and driver gene information.
• The model's performance highlights the importance of driver genes in understanding cancer complexity and heterogeneity.
• This approach advances personalized medicine by enabling more accurate identification of cancer subtypes.