Attention-Based Multi-NMF Deep Neural Network with Multimodality Data for Breast Cancer Prognosis Model

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Accurate prognostic prediction for breast cancer is vital to prevent overtreatment and guide clinical decisions.
• Gene expression profiles and clinical data are key biological markers for cancer prediction.
• Integrating diverse data modalities can enhance the accuracy of predictive models.

Purpose of the Study:

• To develop an advanced model for precise prognostic prediction in breast cancer patients.
• To integrate clinical data with gene expression data for improved predictive performance.
• To introduce a novel approach for breast cancer survival analysis.

Main Methods:

• Proposed an end-to-end model named Attention-based Multi-NMF DNN (AMND).
• Utilized Multiple Nonnegative Matrix Factorization (Multi-NMF) algorithms to extract features from gene expression data.
• Integrated clinical data and employed an Attention mechanism for enhanced feature selection.

Main Results:

• The AMND model demonstrated superior predictive performance compared to models using single data modalities (gene or clinical).
• AMND outperformed models relying on single Nonnegative Matrix Factorization (NMF) algorithms for feature extraction.
• The model's performance was competitive with, or better than, previously reported breast cancer prognostic models.

Conclusions:

• The proposed AMND model offers a robust strategy for breast cancer prognostic prediction.
• Integrating multi-modal data (clinical and gene expression) significantly improves prediction accuracy.
• The AMND framework is adaptable for survival prediction in other cancer types.