Classifying breast cancer using multi-view graph neural network based on multi-omics data.

Area of Science:

• Computational Biology
• Bioinformatics
• Oncology

Background:

• Cancer grading and subtyping are crucial for prognosis and treatment, yet current prediction methods often rely on traditional machine learning and single omics data.
• Integrating diverse omics data offers a more comprehensive approach to understanding cancer heterogeneity.
• Deep learning presents an opportunity to enhance multi-omics data integration for improved cancer classification.

Purpose of the Study:

• To propose a novel deep learning algorithm for predicting cancer differentiation and subtypes by integrating multi-omics data.
• To develop a multi-view graph neural network (MVGNN) model for enhanced cancer classification.
• To evaluate the performance of the MVGNN model against existing methods.

Main Methods:

• A multi-view graph neural network (MVGNN) framework was developed, incorporating graph convolutional network (GCN) and attention modules.
• Feature selection was performed on three types of omics data using chi-square and mRMR methods.
• Weighted patient similarity networks were constructed and utilized for GCN training, followed by attention-based multi-omics data integration.

Main Results:

• The MVGNN model demonstrated strong performance in cancer classification prediction.
• Comparative experiments showed the MVGNN outperforming traditional machine learning and other multi-omics integration methods.
• Performance was evaluated using 5-fold cross-validation, with analyses conducted on single, dual, and triple omics data.

Conclusions:

• The proposed MVGNN model is effective for cancer classification prediction using integrated multi-omics data.
• This deep learning approach offers a promising advancement in cancer differentiation and subtype analysis.
• The study highlights the potential of MVGNN in improving prognostic and therapeutic strategies through accurate cancer subtyping.