Comparative analysis of statistical and deep learning-based multi-omics integration for breast cancer subtype...

Area of Science:

• Oncology
• Bioinformatics
• Genomics

Background:

• Breast cancer (BC) is a leading cause of cancer mortality worldwide.
• BC subtype heterogeneity complicates molecular understanding, diagnosis, and treatment.
• Multi-omics integration shows promise for enhancing BC subtype identification, but methods require evaluation.

Purpose of the Study:

• To compare statistical (MOFA+) and deep learning (MOGCN) multi-omics integration approaches for breast cancer subtyping.
• To evaluate feature selection effectiveness and biological relevance for BC subtypes.
• To identify optimal methods for advancing personalized breast cancer medicine.

Main Methods:

• Integrated host transcriptomics, epigenomics, and shotgun microbiome data from 960 BC patient samples.
• Compared MOFA+ (statistical) and MOGCN (deep learning) for multi-omics integration.
• Assessed feature discrimination using linear/nonlinear models and analyzed pathway relevance.

Main Results:

• MOFA+ outperformed MOGCN in feature selection, achieving a higher F1 score (0.75) in nonlinear classification.
• MOFA+ identified 121 relevant pathways, compared to 100 identified by MOGCN.
• Key pathways like Fc gamma R-mediated phagocytosis and SNARE pathway were implicated, offering insights into immune response and tumor progression.

Conclusions:

• MOFA+ is a superior unsupervised tool for feature selection in breast cancer subtyping.
• Multi-omics integration holds significant potential for improving BC subtype prediction.
• The findings provide critical insights for developing personalized medicine strategies for breast cancer.