Exploring drug-target interaction prediction on cold-start scenarios via meta-learning-based graph transformer

Area of Science:

• Computational biology
• Pharmacology
• Artificial intelligence in drug discovery

Background:

• Drug-target interaction (DTI) prediction is crucial for drug discovery.
• Existing computational methods struggle with the cold-start problem due to reliance on prior interaction data.
• New drugs or targets with limited data pose a significant challenge for current DTI prediction models.

Purpose of the Study:

• To address the cold-start problem in drug-target interaction prediction.
• To develop a computational method capable of predicting DTIs for novel drugs and targets.
• To enhance the adaptability of DTI prediction models to scenarios with scarce interaction information.

Main Methods:

• Proposed MGDTI (Meta-learning-based Graph Transformer for Drug-Target Interaction prediction).
• Utilized drug-drug and target-target similarity to augment limited interaction data.
• Employed meta-learning for adaptability to cold-start tasks.
• Leveraged graph transformers to capture long-range dependencies and prevent over-smoothing.

Main Results:

• MGDTI demonstrated effectiveness in DTI prediction under cold-start scenarios.
• The method successfully mitigated the scarcity of interaction data using similarity information.
• Meta-learning enabled MGDTI to adapt to new drugs and targets with no prior interactions.

Conclusions:

• MGDTI offers a promising solution for the cold-start problem in DTI prediction.
• The integration of meta-learning and graph transformers enhances prediction accuracy for novel entities.
• This approach advances computational drug discovery by enabling predictions in data-scarce environments.