A Novel Attention-Mechanism Based Cox Survival Model by Exploiting Pan-Cancer Empirical Genomic Information

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Accurate cancer prognosis is crucial for treatment planning and early diagnosis.
• Deep learning shows promise in biomedical applications but faces challenges with high-dimensional cancer transcriptome data, including overfitting and limited training samples.
• Existing deep learning survival analysis models struggle with the complexities of cancer data.

Purpose of the Study:

• To introduce a novel deep learning framework, SAVAE-Cox, for survival analysis of high-dimensional cancer transcriptome data.
• To address challenges like overfitting and improve the accuracy of cancer prognosis.
• To leverage adversarial transfer learning and attention mechanisms for enhanced model performance.

Main Methods:

• Developed SAVAE-Cox, a framework integrating an attention mechanism and adversarial transfer learning.
• Trained and evaluated the model on 16 types of TCGA cancer RNA-seq datasets.
• Compared SAVAE-Cox against state-of-the-art survival analysis models, including Cox-ph, Cox-lasso, Cox-ridge, Cox-nnet, and VAECox.

Main Results:

• SAVAE-Cox demonstrated superior performance compared to existing models, achieving a higher concordance index.
• The model effectively handled high-dimensional transcriptome data and addressed overfitting issues.
• Feature analysis experiments indicated the model's utility in identifying cancer-related genes and biological functions.

Conclusions:

• SAVAE-Cox offers a robust and accurate solution for survival analysis in high-dimensional cancer transcriptome data.
• The framework's innovative approach, combining attention and adversarial transfer learning, enhances prognostic capabilities.
• SAVAE-Cox aids in the discovery of crucial cancer biomarkers and biological pathways.