Attention-based Knowledge Graph Representation Learning for Predicting Drug-drug Interactions.

Area of Science:

• Pharmacology
• Bioinformatics
• Computational Biology

Background:

• Drug-drug interactions (DDIs) are a major cause of adverse drug events.
• Accurate DDI identification is crucial for drug development and patient safety.
• Existing computational methods often fail to fully utilize biomedical knowledge graphs (KGs).

Purpose of the Study:

• To propose an attention-based KG representation learning framework, DDKG, for enhanced DDI prediction.
• To leverage rich information within KGs, including drug attributes and triple facts, for improved DDI identification.
• To develop an end-to-end model for predicting the likelihood of interactions between drug pairs.

Main Methods:

• DDKG initializes drug representations using an encoder-decoder layer with drug attributes.
• It learns drug representations by propagating information along network paths based on node embeddings and triple facts.
• Pairwise drug interaction probabilities are estimated using learned drug representations.

Main Results:

• DDKG demonstrates superior performance compared to state-of-the-art algorithms on DDI prediction.
• Experiments were conducted on two practical datasets of varying sizes.
• The framework achieved better results across multiple evaluation metrics on all tested datasets.

Conclusions:

• The proposed DDKG framework effectively utilizes biomedical KGs for accurate DDI prediction.
• DDKG offers a significant improvement over existing methods for identifying potential drug-drug interactions.
• This approach enhances drug safety and aids in the drug development process.