CNN-Cox

CNN-Cox predicts patient survival from cancer genomics data using a convolutional neural network integrated with cascaded Wx feature selection for prognosis-related tasks.

Key Features:

• Convolutional Neural Network Framework: A CNN architecture extracts hierarchical features from high-throughput cancer genomics data to characterize subtypes and inform survival prediction.
• Prognosis-Related Feature Selection (Cascaded Wx): A cascaded Wx feature selection process screens and prioritizes prognostic genes to reduce dimensionality and mitigate overfitting in high-throughput sequencing data.
• Efficiency and Performance: A reduced-parameter CNN design provides robust survival prediction performance across multiple cancer types in The Cancer Genome Atlas (TCGA) cohort, including bladder carcinoma, head and neck squamous cell carcinoma, kidney renal cell carcinoma, brain low-grade glioma, lung adenocarcinoma (LUAD), lung squamous cell carcinoma, and skin cutaneous melanoma.
• Interpretability: Identifies potential prognostic gene signatures and hub genes (e.g., ANLN, RACGAP1, KIF4A among 13 hub genes in LUAD) with supporting protein-protein interaction network analysis.

Scientific Applications:

• Survival prediction: Predicts patient survival outcomes and prognosis from cancer genomics datasets.
• Biomarker discovery: Identifies biologically meaningful prognostic gene features and hub genes that may inform therapeutic target research.
• Cross-cancer prognostic analysis: Enables comparative prognostic analyses across multiple TCGA cancer types listed above.

Methodology:

Implements a convolutional neural network to extract hierarchical features, applies cascaded Wx for feature selection to screen prognostic genes and reduce dimensionality, and uses protein-protein interaction network analysis to support biological interpretation, with evaluation on TCGA cohorts.