Drug repurposing and prediction of multiple interaction types via graph embedding.

Area of Science:

• Computational drug discovery
• Bioinformatics
• Network pharmacology

Background:

• Identifying drug-target interactions (DTIs) is crucial for targeted drug discovery and repurposing.
• Delineating the specific type of drug interaction is essential for understanding therapeutic outcomes.

Purpose of the Study:

• To develop a computational approach for predicting novel drug-target interactions (DTIs).
• To predict the type of interaction induced by drugs on their targets.
• To facilitate drug repurposing by identifying new therapeutic applications for existing drugs.

Main Methods:

• A heterogeneous graph mining approach integrating drug-drug and protein-protein similarity networks with drug-disease and protein-disease associations.
• Node embedding principles to map the three-layer heterogeneous graph into low-dimensional vectors.
• A multi-label, multi-class classification task using gradient boosted trees to predict DTIs and their interaction types.

Main Results:

• The DT2Vec+ model demonstrated promising results in predicting the type of DTIs.
• The methodology successfully integrated and mapped drug-target-disease associations into dense vectors.
• A comprehensive analysis predicted the degree and type of interaction for unknown DTIs, including applications to cancer biomarkers.

Conclusions:

• DT2Vec+ is a novel approach for predicting DTIs across six interaction types.
• The method effectively leverages heterogeneous graph information and node embedding for DTI prediction.
• This work advances drug repurposing by providing a robust tool for identifying drug-target interactions and their modes of action.