A Knowledge Graph-Enhanced Tensor Factorisation Model for Discovering Drug Targets.

Area of Science:

• Computational biology
• Pharmacology
• Machine learning in drug discovery

Background:

• Drug discovery is lengthy and costly, with high attrition rates.
• Machine learning (ML) is increasingly applied to optimize drug discovery and development.
• Identifying druggable disease genes is a critical early-stage challenge.

Purpose of the Study:

• To develop an advanced ML model for predicting potential drug targets (genes/proteins) for diseases.
• To enhance the accuracy of identifying effective therapeutic targets.
• To reduce the time and cost associated with drug discovery.

Main Methods:

• A novel tensor factorization model was developed using a 3D data tensor (genes, diseases, evidence/outcomes).
• Data was sourced from Open Targets and PharmaProjects databases.
• Gene target representations were enriched using a drug discovery knowledge graph, and the model predicted clinical outcomes for novel gene-disease pairs.

Main Results:

• Incorporating knowledge graph embeddings significantly improved prediction accuracy.
• The proposed tensor factorization model trained with a dense neural network outperformed baseline ML classifiers and Bayesian methods.
• Evaluation using three strategies confirmed the model's predictive performance.

Conclusions:

• The developed framework effectively combines tensor factorization and knowledge graph representation learning for disease target identification.
• This approach offers a promising data-driven strategy to accelerate drug discovery.
• Further exploration of this combined ML methodology is warranted for future drug discovery efforts.