moBRCA-net: a breast cancer subtype classification framework based on multi-omics attention neural networks.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Breast cancer heterogeneity necessitates accurate subtyping for effective treatment.
• Current subtyping relies on single-omics data, limiting comprehensive patient views.
• Multi-omics data integration offers a holistic approach but faces high dimensionality challenges.

Purpose of the Study:

• To develop an interpretable deep learning framework for breast cancer subtype classification using multi-omics data.
• To address limitations of existing deep learning approaches in multi-omics integration.

Main Methods:

• Developed moBRCA-net, a deep learning framework integrating gene expression, DNA methylation, and microRNA datasets.
• Incorporated a self-attention module to capture feature importance within each omics dataset.
• Transformed features based on learned importance for subtype prediction.

Main Results:

• moBRCA-net demonstrated significantly enhanced performance compared to existing methods.
• Validated the effectiveness of multi-omics integration for improved breast cancer classification.
• Confirmed the utility of omics-level attention in capturing feature relevance.

Conclusions:

• moBRCA-net provides an effective and interpretable approach for breast cancer subtyping.
• The framework highlights the power of integrating diverse omics data with attention mechanisms.
• The moBRCA-net tool is publicly available for research use.