3DGT-DDI

3DGT-DDI predicts drug–drug interactions (DDIs) by integrating three-dimensional molecular graph representations with textual information using a deep learning architecture.

Key Features:

• 3D Molecular Graph Representation: Encodes drugs as three-dimensional molecular graphs incorporating atomic connectivity and positional information to capture spatial molecular structure.
• Hybrid Deep Learning Architecture: Combines a 3D graph neural network with a pre-trained text attention mechanism to integrate chemical structural information with textual biological context.
• Molecular Substructure Modeling: Analyzes molecular substructures within 3D molecular graphs to identify structural components contributing to drug–drug interactions.
• Performance Evaluation on DDIExtraction 2013: Achieves a macro F1 score of 84.48% on the DDIExtraction 2013 shared task dataset for drug–drug interaction prediction.
• Model Interpretability: Provides weight visualization on datasets such as DrugBank to identify molecular features contributing to predicted interactions.

Scientific Applications:

• Combination Therapy Analysis: Predicts potential interactions between co-administered drugs to support evaluation of combination treatment strategies.
• Adverse Drug Interaction Detection: Identifies drug pairs with potential adverse interaction effects that may impact therapeutic safety.
• Drug Development Research: Supports identification of structural determinants of drug–drug interactions to guide drug candidate design and optimization.

Methodology:

3DGT-DDI represents drugs as three-dimensional molecular graphs, processes them using a 3D graph neural network combined with a pre-trained text attention mechanism, and predicts drug–drug interactions with interpretability through weight visualization.