Semi-supervised heterogeneous graph contrastive learning for drug-target interaction prediction.

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning

Background:

• Drug-target interactions (DTIs) are crucial for drug discovery and repositioning.
• Graph-based methods show promise for DTI prediction but struggle with limited labeled data.
• Limited data reduces the generalization ability of existing DTI prediction models.

Purpose of the Study:

• To propose a novel framework, SHGCL-DTI, for improved DTI prediction.
• To leverage self-supervised contrastive learning to overcome data limitations in DTI prediction.
• To enhance the generalization ability of DTI prediction models.

Main Methods:

• Developed SHGCL-DTI, a framework combining semi-supervised DTI prediction with graph contrastive learning.
• Generated node representations using neighbor and meta-path views.
• Defined positive and negative pairs to maximize similarity between different views for contrastive learning.
• Reconstructed the heterogeneous network to predict potential DTIs.

Main Results:

• SHGCL-DTI demonstrated significant improvements in DTI prediction across various scenarios compared to state-of-the-art methods.
• Ablation studies confirmed the contrastive learning module enhances prediction performance and generalization.
• Identified novel DTIs supported by existing biological literature.

Conclusions:

• SHGCL-DTI effectively improves drug-target interaction prediction using self-supervised contrastive learning.
• The proposed framework offers enhanced generalization capabilities, crucial for drug discovery.
• The method successfully identifies novel, biologically relevant DTIs.