DFinder

DFinder identifies drug-food interactions (DFIs) using an end-to-end graph embedding approach that integrates node attribute features and topological structure features to model how food constituents influence drug pharmacodynamics and pharmacokinetics.

Key Features:

• End-to-end graph embedding: Integrates node attribute features with network topology to model complex relationships between drugs and food constituents.
• Data integration from DrugBank and PubMed: Constructs two DFI networks from the DrugBank-DFI and PubMed-DFI datasets derived from DrugBank and PubMed.
• Graph Convolution Network for topology: Uses a simplified Graph Convolution Network (GCN) to learn topological structure features within DFI networks.
• Deep neural networks for attribute extraction: Employs deep neural networks to extract and process original node attribute features for drugs and food constituents.
• Performance versus baselines: Demonstrates superior predictive performance compared with other baseline methods for identifying DFIs.

Scientific Applications:

• Pharmacology and Toxicology: Predicts interactions that can inform understanding of adverse drug reactions and therapeutic efficacy influenced by food constituents.
• Nutritional Science: Provides insights into how dietary components may affect drug metabolism and absorption relevant to personalized nutrition.
• Drug Development: Screens for potential DFIs during drug development to help anticipate interactions prior to clinical evaluation.

Methodology:

End-to-end graph embedding integrating node attribute features and topological structure features; construction of DrugBank-DFI and PubMed-DFI networks; simplified Graph Convolution Network (GCN) for topological feature learning; deep neural networks for attribute feature extraction.