DFinder: a novel end-to-end graph embedding-based method to identify drug-food interactions

Area of Science:

• Pharmacology and Cheminformatics
• Computational Biology and Bioinformatics

Background:

• Drug-food interactions (DFIs) significantly impact drug efficacy and safety.
• Existing computational methods show promise for DFI discovery but lack focus on identification.
• Challenges include limited DFI data and the complexity of representing food components.

Purpose of the Study:

• To develop an effective computational approach for identifying drug-food interactions (DFIs).
• To address the limitations of data scarcity and complex food feature representation in DFI prediction.

Main Methods:

• Constructed two DFI datasets (DrugBank-DFI, PubMed-DFI) from existing databases.
• Developed DFinder, a novel end-to-end graph embedding method for DFI identification.
• DFinder integrates node attribute features and topological structure features using graph convolution networks and deep neural networks.

Main Results:

• DFinder successfully learned representations for drugs and food constituents by combining attribute and topological features.
• The proposed method demonstrated superior performance compared to existing baseline methods in DFI identification.
• The study successfully constructed DFI networks for analysis.

Conclusions:

• DFinder provides an effective computational solution for identifying drug-food interactions.
• The method's ability to handle complex food components and limited data offers a significant advancement.
• The developed approach has the potential to aid in the discovery of novel DFIs.