Drug-target interaction prediction by integrating heterogeneous information with mutual attention network

Area of Science:

• Computational biology
• Drug discovery and development
• Bioinformatics

Background:

• Accurate drug-target interaction (DTI) prediction is crucial for drug discovery.
• Existing machine learning methods struggle with novel chemical structures.
• Leveraging large-scale biological and pharmacological data offers a promising avenue for DTI prediction.

Purpose of the Study:

• To develop a deep learning model, DrugMAN, for enhanced DTI prediction.
• To integrate multiplex heterogeneous functional networks using a mutual attention network (MAN).
• To improve the adaptability and generalization of DTI prediction models.

Main Methods:

• DrugMAN employs a graph attention network-based integration algorithm.
• It learns low-dimensional features from four drug networks and seven gene/protein networks.
• A mutual attention network (MAN) captures interaction information between drug and target representations.

Main Results:

• DrugMAN outperformed existing methods (SVM, RF, DeepPurpose, DTINet, NeoDT) in DTI prediction across four scenarios.
• It demonstrated superior performance, especially in real-world applications.
• DrugMAN exhibited minimal performance degradation in cold-start scenarios, indicating strong generalization ability.

Conclusions:

• DrugMAN effectively utilizes heterogeneous information for mining drug-target interactions.
• The model shows significant potential as a tool for accelerating drug discovery and repurposing.
• Its robust generalization ability makes it suitable for diverse and challenging DTI prediction tasks.