CancerOmicsNet

CancerOmicsNet predicts the effectiveness of kinase inhibitors on cancer cell lines by integrating biological networks, genomic information, inhibitor profiling, and gene-disease associations into a graph neural network to model system-level interactions.

Key Features:

• Graph Neural Network Architecture: Employs a graph neural network with attention propagation mechanisms to capture intricate relationships within heterogeneous biological data.
• Integration of Heterogeneous Data: Integrates biological networks, genomic information, inhibitor profiling, and gene-disease associations into a unified graph structure.
• High Predictive Accuracy: Achieves an area under the receiver operating characteristics (ROC) curve of 0.83 for predicting kinase inhibitor responses.
• Generalization to Unseen Data: Generalizes across different cancer cell lines and inhibitors to enable predictions beyond training data.
• Cross-Validation at Tissue Level: Uses tissue-level cross-validation to evaluate and ensure robust predictive performance across tissue types.

Scientific Applications:

• Predicting kinase inhibitor responses: Predicts therapeutic effects of kinase inhibitors on different cancer cell lines.
• System-level interpretation of cancer processes: Supports understanding of interactions among genes, drugs, and biological networks in cancer by integrating diverse data sources.
• Prioritizing therapeutic candidates: Enables identification and prioritization of promising kinase inhibitors for further experimental validation.

Methodology:

Constructs a graph-based representation with nodes representing genes or drugs and edges denoting relationships, then applies a graph neural network with attention propagation mechanisms to learn representations for predicting kinase inhibitor effects, with performance assessed via tissue-level cross-validation.