A unified drug-target interaction prediction framework based on knowledge graph and recommendation system.

Area of Science:

• Pharmacology and Bioinformatics
• Computational Drug Discovery

Background:

• Drug-target interaction (DTI) prediction is crucial for drug development, virtual screening, and repurposing.
• Existing DTI prediction methods struggle with sparse data and the cold start problem, limiting their effectiveness.

Purpose of the Study:

• To develop a unified framework, KGE_NFM, for accurate and robust DTI prediction.
• To address the limitations of existing DTI prediction methods, particularly the cold start problem.

Main Methods:

• KGE_NFM combines knowledge graph (KG) embeddings with Neural Factorization Machines (NFM).
• The framework learns low-dimensional representations of entities within a KG.
• Multimodal information is integrated using NFM for enhanced prediction.

Main Results:

• KGE_NFM demonstrated accurate and robust DTI predictions across four benchmark datasets.
• The framework showed particular strength in handling the cold start problem for proteins.
• Evaluations were conducted under three realistic prediction scenarios.

Conclusions:

• KGE_NFM offers a valuable approach for integrating KG and recommendation system techniques.
• The unified framework facilitates novel drug-target interaction discovery.
• This method provides insights for advancing computational drug discovery.