An Improved Deep Learning Algorithm for Breast Cancer Survival Prediction Based on Multi-Omics Data.

Area of Science:

• Computational biology
• Oncology
• Genomics

Background:

• Breast cancer remains a significant global health challenge, particularly for women.
• Accurate survival prediction is crucial for effective clinical decision-making and personalized treatment strategies.
• Multi-omics data holds potential for enhancing breast cancer prognosis.

Purpose of the Study:

• To develop an interpretable and effective deep learning model for breast cancer survival prediction.
• To leverage multi-omics data, including clinical, copy number alteration (CNA), and gene expression data.
• To integrate advanced deep learning architectures with robust feature selection methods.

Main Methods:

• A novel deep learning model combining Bi-directional Long Short-Term Memory (BiLSTM) and Convolutional Neural Network (CNN) architectures was proposed.
• Minimum Redundancy Maximum Relevance (MRMR) feature selection was integrated to identify key predictive features.
• The model was validated on two large-scale datasets: METABRIC (n=1980) and TCGA-BRCA (n=1080).

Main Results:

• The proposed model demonstrated superior performance over existing algorithms, achieving high AUC-ROC and accuracy.
• The integrated BiLSTM and CNN architectures effectively captured temporal and spatial patterns in the data.
• Exceptional accuracy was reported: 98% on the METABRIC dataset and 96% on the TCGA dataset.

Conclusions:

• The combined BiLSTM, CNN, and MRMR framework provides an accurate and interpretable method for breast cancer survival prediction.
• This approach offers valuable insights for oncologists, potentially improving patient management.
• The model shows promise for broader applications within the field of oncology.