GraphGR

GraphGR predicts the therapeutic effects of kinase inhibitors on cancer cell lines by applying graph neural networks to integrated biological networks, genomic profiles, inhibitor profiling, and gene-disease associations.

Key Features:

• Integration of heterogeneous data: Integrates biological networks, genomic profiles, inhibitor profiling, and gene-disease associations into unified graphs for each cell-line–drug combination.
• Attention propagation mechanisms: Employs attention propagation mechanisms within the graph neural network architecture to prioritize relevant features and interactions.
• Graph reduction protocol: Utilizes a graph reduction protocol that leverages topological data and biological knowledge to reduce network complexity while retaining critical information.
• Tissue-level cross-validation and performance: Evaluated with cross-validation at the tissue level, achieving an area under the receiver operating characteristic curve (AUC) of 0.83 on the presented dataset.
• Generalization and literature validation: Demonstrates generalization across diverse cancer cell lines and kinase inhibitors with predictions validated against biomedical literature.

Scientific Applications:

• Oncology research: Predicts responses of cancer cell lines to specific kinase inhibitors to support hypothesis generation in preclinical studies.
• Precision medicine: Informs personalized treatment strategy research by predicting drug efficacy across cell-line–drug combinations using multi-modal biological data.

Methodology:

Constructs a graph for each cell-line–drug instance by integrating heterogeneous data sources, applies a graph neural network enhanced with attention propagation mechanisms to predict binary drug response outcomes, and reduces network complexity using a topology- and biology-informed graph reduction protocol; performance is assessed via tissue-level cross-validation (AUC 0.83).