MSH-DTI: multi-graph convolution with self-supervised embedding and heterogeneous aggregation for drug-target...

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Network pharmacology utilizes computational methods for drug-target interaction (DTI) prediction.
• Existing DTI models face limitations due to restricted feature data and simplistic information integration from multiple networks.

Purpose of the Study:

• To propose MSH-DTI, a novel deep learning model for robust DTI prediction.
• To address limitations in feature extraction and heterogeneous information integration in DTI prediction.

Main Methods:

• Utilizes self-supervised learning for comprehensive drug and target structure feature extraction.
• Employs a Heterogeneous Interaction-enhanced Feature Fusion Module for multi-graph construction.
• Applies graph convolutional networks and an attention mechanism for effective feature extraction and integration.

Main Results:

• MSH-DTI achieved high performance with AUROC of 0.9620 and AUPR of 0.9605 on the DTINet dataset.
• Outperformed existing DTI prediction models in experimental evaluations.
• Demonstrated the model's ability to focus on critical features through an attention mechanism.

Conclusions:

• MSH-DTI serves as an effective tool for discovering novel drug-target interactions.
• The model's utility is further validated by successful case studies in predicting new DTIs.