ContraDTI: Improved drug-target interaction prediction via multi-view contrastive learning

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning

Background:

• Drug-target interaction (DTI) identification is vital for drug discovery.
• Machine learning accelerates DTI prediction but is limited by scarce annotated data.
• Existing supervised methods struggle with limited labeled drug data.

Purpose of the Study:

• To introduce ContraDTI, a novel framework for DTI prediction using multi-view contrastive learning.
• To address the challenge of limited labeled data in DTI prediction.
• To enhance the performance of DTI prediction models in data-limited scenarios.

Main Methods:

• ContraDTI employs a multi-view contrastive learning approach.
• It utilizes the drug's molecular graph as the main view and its SMILES string as the side view.
• Two loss functions are used for main view contrast and cross-view alignment.

Main Results:

• ContraDTI significantly improves DTI prediction classification performance.
• The framework demonstrates superior performance, especially in low-data regimes.
• Experiments on single-target and multi-target DTI datasets validate its effectiveness.

Conclusions:

• ContraDTI is a powerful tool for drug-target interaction prediction.
• It effectively overcomes data scarcity challenges in DTI prediction.
• The framework offers a promising solution for data-limited drug discovery scenarios.