CAGCL: Predicting Short- and Long-Term Breast Cancer Survival With Cross-Modal Attention and Graph Contrastive...

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Accurate breast cancer prognosis is essential for guiding treatment strategies and supporting patient recovery.
• Leveraging diverse datasets can improve the precision of survival prediction models.

Purpose of the Study:

• To develop a novel predictive model for forecasting breast cancer prognosis.
• To enhance breast cancer survival prediction by integrating multi-modal data.

Main Methods:

• Utilized the TCGA Database, incorporating clinical records, copy number variation, gene expressions, DNA methylation, microRNA sequencing, and whole slide images.
• Developed a model (CAGCL) employing graph contrastive learning with cross-modality attention to integrate six distinct data modalities.
• Applied a cross-attention framework to graph contrastive learning-extracted features for binary classification of short-term and long-term survivors (five-year threshold).

Main Results:

• The proposed CAGCL model demonstrated superior performance over baseline and state-of-the-art models.
• Achieved high performance metrics: 0.932 accuracy, 0.954 sensitivity, 0.958 precision, 0.956 F1 score, and 0.948 AUC.
• Effectively predicted breast cancer survival likelihood using integrated multi-modal data.

Conclusions:

• The CAGCL model offers a highly effective approach for breast cancer survival prediction.
• Integrating diverse data sources through graph contrastive learning and cross-attention significantly improves prognostic accuracy.
• This model can aid clinicians in making more informed treatment decisions for breast cancer patients.