DGSS: A Dynamic Interaction Graph Neural Network with Specific Substructure Awareness for Drug Synergy Prediction.

Area of Science:

• Computational biology
• Pharmacology
• Oncology

Background:

• Combination therapy is crucial for treating complex diseases like cancer, but monotherapy faces toxicity and resistance.
• Current computational methods struggle to model cell-specific drug responses and dynamic drug-cell interactions, hindering accurate synergy prediction.

Purpose of the Study:

• To develop a novel computational framework, DGSS (Dynamic Interaction Graph Neural Network with Cell-Specific Drug Substructure Awareness), for predicting synergistic drug pairs.
• To explicitly capture cell-specific drug substructures and dynamic drug-cell interactions for enhanced synergy prediction.

Main Methods:

• Developed DGSS, incorporating a hierarchical attention mechanism to identify cell-line-specific drug substructures by correlating molecular subgraphs with genomic features.
• Implemented a dynamic graph network to model evolving cell-line states during drug exposure, capturing context-dependent interactions.

Main Results:

• DGSS demonstrated robust performance across 12 datasets and three partitioning strategies, consistently outperforming state-of-the-art baseline models.
• Achieved high performance on the Loewe Synergy dataset with AUROC of 96.0% and AUPRC of 85.5%, indicating strong predictive accuracy and stability.

Conclusions:

• DGSS effectively bridges molecular substructure dynamics with cellular context, advancing precision in drug synergy prediction.
• The proposed data-driven framework offers a promising approach for optimizing combination therapies in personalized oncology.