Classifying breast cancer subtypes on multi-omics data via sparse canonical correlation analysis and deep learning.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate breast cancer subtype classification is essential for effective clinical diagnosis and treatment.
• Early breast cancer symptoms are often subtle, necessitating advanced diagnostic methods.
• High-throughput sequencing generates vast multi-omics data, offering potential for improved subtype identification.

Purpose of the Study:

• To develop a computational method for integrating multi-omics data to enhance breast cancer subtype classification.
• To address the challenge of identifying associations between different omics data for predictive modeling.
• To improve the accuracy of breast cancer subtype prediction by leveraging multi-omics data integration.

Main Methods:

• Proposed a differential sparse canonical correlation analysis network (DSCCN) for breast cancer subtype classification.
• Performed differential analysis on multi-omics expression data to identify differentially expressed (DE) genes.
• Employed sparse canonical correlation analysis (SCCA) to find correlations between multi-omics DE genes and used multi-task deep learning for prediction.

Main Results:

• DSCCN effectively identifies differentially expressed genes across multiple omics datasets.
• Sparse canonical correlation analysis (SCCA) successfully mined highly correlated features between these genes.
• Multi-task deep learning models trained on correlated DE genes accurately predicted breast cancer subtypes, addressing data heterogeneity.

Conclusions:

• The proposed DSCCN method demonstrates superior accuracy in classifying breast cancer subtypes compared to existing approaches.
• Mining associations among multi-omics data is a powerful strategy for improving breast cancer subtype prediction.
• DSCCN offers a promising tool for leveraging complex biological data in cancer research and clinical applications.