BridgeDPI

BridgeDPI predicts drug–protein interactions by integrating network-level information and molecular properties through virtual nodes and a Graph Neural Network to improve DPI prediction.

Key Features:

• Virtual Nodes Integration: Introduces virtual nodes that bridge drugs and proteins to construct a learnable drug-protein association network.
• Graph Neural Network (GNN) Utilization: Employs a Graph Neural Network to propagate information across the drug-protein association network and capture direct and indirect interactions.
• Combination of Methods: Combines network-based (guilt-by-association) and learning-based approaches using existing interaction databases to leverage complementary information.
• Supervised Optimization: Optimizes the drug-protein association network using supervised signals derived from downstream DPI prediction tasks.

Scientific Applications:

• Drug–Protein Interaction Prediction: Predicts DPIs on real-world datasets, including BindingDB, C.ELEGANS, HUMAN, and DUD-E.
• Therapeutic Target Prioritization: Prioritizes potential therapeutic targets by improving the accuracy of DPI predictions.
• Support for Drug Discovery: Supports discovery and development of novel drugs by enabling more accurate screening of drug–protein associations.

Methodology:

Constructs a learnable drug-protein association network using virtual nodes, optimizes this network with supervised signals from DPI prediction tasks, and applies a Graph Neural Network while combining network-based and learning-based information from existing interaction databases.