Drug-target interaction predication via multi-channel graph neural networks.

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Drug-target interaction (DTI) prediction is crucial for drug discovery.
• Existing methods often struggle with discrete or manual feature representations.
• Current deep learning approaches for DTI prediction may not fully integrate topological structure and semantic information.

Purpose of the Study:

• To propose a novel deep learning model, DTI-MGNN, for enhanced drug-target interaction prediction.
• To address limitations in current methods by fusing topological structure and semantic information in drug-protein (DPP) representation learning.
• To improve the learning of DPP node representations by considering differential influences of neighboring nodes.

Main Methods:

• Developed DTI-MGNN, a model utilizing multi-channel graph convolutional networks and graph attention.
• Employed two independent graph attention networks to learn distinct node interactions within topology and feature graphs.
• Integrated a graph convolutional network with shared weights to capture common information across both graph types.

Main Results:

• The DTI-MGNN model successfully combines topological structure and semantic features for improved DPP representation learning.
• Achieved state-of-the-art performance on public datasets for DTI prediction.
• Demonstrated high accuracy in identifying DTIs, with an area under the receiver operating characteristic curve of 0.9665.

Conclusions:

• DTI-MGNN offers a significant advancement in DTI prediction by effectively integrating diverse data representations.
• The model's ability to capture complex node interactions enhances its predictive power.
• This approach holds promise for accelerating the drug discovery pipeline through more accurate target identification.