An Integrated Method Based on Wasserstein Distance and Graph for Cancer Subtype Discovery

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Cancer pathogenesis is complex, involving multiple omics levels.
• Accurate cancer subtype identification is crucial for effective treatment.
• Existing methods may not fully capture multi-omics data complexity.

Purpose of the Study:

• To propose a novel computational method for multi-omics cancer subtype identification.
• To enhance the accuracy of distinguishing cancer subtypes using integrated omics data.
• To provide a robust framework for personalized cancer therapy.

Main Methods:

• Developed a novel method, Wasserstein distance and graph autoencoder for multi-omics (WVGMO).
• Employed a variational autoencoder measured by Wasserstein distance (WVAE) for spatial information extraction.
• Utilized a graph autoencoder (GAE) to preserve topological and feature information.
• Applied k-means clustering for final cancer subtype identification.

Main Results:

• WVGMO demonstrated strong performance across seven different cancer types.
• The method effectively integrated four types of omics data from TCGA.
• Results showed WVGMO performed comparably or superiorly to existing advanced methods.

Conclusions:

• WVGMO offers a powerful approach for multi-omics cancer subtype identification.
• The method's ability to retain topological and feature information is key to its success.
• This approach holds promise for advancing precision oncology and treatment strategies.