Efficient substructure feature encoding based on graph neural network blocks for drug-target interaction prediction.

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Drug-target interaction (DTI) prediction is vital for drug discovery.
• Graph neural networks (GNNs) show promise for drug feature encoding.
• Existing GNN methods struggle to balance local and global drug molecular graph features.

Purpose of the Study:

• To develop a novel model, GNNBlockDTI, for improved DTI prediction.
• To enhance the representation learning of drug molecules and target proteins.
• To address limitations in current GNN-based DTI prediction methods.

Main Methods:

• Proposed GNNBlockDTI model incorporating GNNBlocks for local pattern capture.
• Implemented a feature enhancement strategy with gating units for refined drug features.
• Utilized variant convolutional networks for local encoding of target protein binding sites.

Main Results:

• GNNBlockDTI demonstrated competitive performance against state-of-the-art models on benchmark datasets.
• Experimental results confirmed the model's high efficacy in DTI prediction.
• A case study validated the practical utility of GNNBlockDTI for drug candidate ranking.

Conclusions:

• GNNBlockDTI offers a robust framework for DTI prediction.
• The model effectively integrates local and global structural information for better predictions.
• GNNBlockDTI shows significant potential for accelerating drug discovery pipelines.