MPVNN

MPVNN predicts cancer-specific survival risk from gene expression by integrating signaling pathway knowledge, notably the PI3K-Akt pathway, with gene mutation data to model pathway disruptions and signal-flow perturbations.

Key Features:

• Pathway Knowledge Integration: Integrates known signaling pathways, including the PI3K-Akt pathway, into the neural network architecture to incorporate prior biological structure.
• Mutation-Driven Edge Replacement: Randomly replaces known pathway edges with gene mutation data to simulate disruptions in signal flow and capture pathway structural variability across cancers.
• Interpretability: Highlights smaller sets of genes and their pathway connections within the network to provide more interpretable risk-predictive features than standard neural networks.

Scientific Applications:

• Cancer-Specific Survival Risk Prediction: Improves cancer-specific survival risk prediction from gene expression compared to other neural network models and standard survival analysis methods.
• Pathway Analysis: Identifies key genes and interactions within the PI3K-Akt pathway that contribute to risk prediction for specific cancer types.

Methodology:

Constructs a neural network architecture embedding known signaling pathways (e.g., PI3K-Akt), integrates gene expression inputs, and randomly replaces pathway edges with gene mutation data to simulate signal-flow disruptions and represent pathway alterations across cancers.