Discovering protein drug targets using knowledge graph embeddings.

Area of Science:

• Bioinformatics
• Computational Biology
• Drug Discovery

Background:

• Computational drug-target interaction (DTI) prediction offers insights into drug mechanisms and effects.
• Existing DTI models have limitations in drug processing, proteome coverage, and high false positive rates.

Purpose of the Study:

• To develop a novel computational approach for predicting drug target proteins.
• To address limitations of existing DTI prediction models.

Main Methods:

• Formulated DTI prediction as a link prediction problem in knowledge graphs.
• Utilized biomedical knowledge bases to construct a comprehensive drug-target knowledge graph.
• Developed and applied a knowledge graph embedding model, TriModel, to learn drug and target representations.

Main Results:

• The TriModel approach demonstrated superior performance in predicting drug-target interactions.
• Outperformed five existing models in standard benchmark tests, achieving higher area under ROC and precision-recall curves.
• Successfully inferred candidate drug-target interactions based on learned embeddings.

Conclusions:

• The proposed knowledge graph embedding approach offers a robust and accurate method for DTI prediction.
• This novel method enhances drug discovery by improving prediction accuracy and coverage.
• The developed models and data are publicly available for further research.