Drug-target prediction through self supervised learning with dual task ensemble approach.

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Drug-target interaction (DTI) prediction is crucial for pharmaceutical research, aiding in virtual screening, identifying new uses for existing drugs, and predicting side effects.
• Current methods often require extensive manual annotations or struggle with robustness.
• Heterogeneous biological networks offer a rich source of information by integrating drug, gene, and disease data.

Purpose of the Study:

• To develop a robust computational model for accurate DTI prediction.
• To leverage self-supervised learning (SSL) for efficient embedding extraction without manual annotations.
• To enhance the reliability of Graph Neural Networks (GNNs) through ensemble learning for DTI prediction.

Main Methods:

• A novel GNN-based architecture was designed, incorporating a task-based module and an ensemble module.
• Self-supervised learning (SSL) was employed for feature embedding extraction, utilizing pretext tasks based on structural or similarity information.
• Ensemble learning was integrated into the GNN framework to improve robustness against non-robustness issues.

Main Results:

• The proposed ensemble module demonstrated strong performance in DTI link prediction.
• Achieved a mean Area Under the Receiver Operating Characteristic curve (AUCROC) of 0.960 in cold start scenarios.
• Achieved a mean AUCROC of 0.970 in warm start scenarios, with minimal deviation, indicating high accuracy and reliability.

Conclusions:

• The developed GNN architecture with an ensemble module effectively predicts drug-target interactions.
• The model shows significant promise for accelerating drug discovery and development by accurately identifying potential drug-target relationships.
• The approach provides a robust and efficient method for DTI prediction, particularly beneficial in scenarios with limited or no prior interaction data (cold start).