Enhancing drug-target interaction prediction with graph representation learning and knowledge-based regularization.

Area of Science:

• Computational Biology
• Drug Discovery
• Bioinformatics

Background:

• Accurate prediction of drug-target interactions (DTIs) is vital for drug discovery and repurposing.
• Existing deep learning methods struggle to capture complex drug-target relationships and integrate biological knowledge effectively.

Purpose of the Study:

• To develop a novel framework for DTI prediction using graph neural networks and knowledge integration.
• To enhance the accuracy and interpretability of DTI prediction models.

Main Methods:

• A customized graph-based message passing scheme to learn representations from molecular structures and protein sequences.
• Knowledge-based regularization strategy to integrate domain knowledge from biomedical ontologies and databases.
• Integration of graph neural networks (GNNs) with knowledge integration for DTI prediction.

Main Results:

• Achieved an average AUC of 0.98 and an average AUPR of 0.89 on benchmark datasets, outperforming state-of-the-art methods.
• Identified salient molecular substructures and protein motifs through visualization of learned attention weights, demonstrating interpretability.
• Predicted novel DTIs for FDA-approved drugs with a high experimental confirmation rate.

Conclusions:

• The proposed framework provides a powerful and interpretable solution for DTI prediction.
• This approach has the potential to significantly accelerate the identification of new drug candidates and therapeutic targets.
• The model's ability to integrate biological knowledge enhances its practical utility in drug discovery.