Drug-target binding affinity prediction method based on a deep graph neural network.

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning

Background:

• Drug development is lengthy and expensive.
• Computer-aided drug design (CADD) offers a solution by reducing costs and development time.
• Drug-Target binding Affinity (DTA) prediction is crucial for screening potential drug candidates.

Purpose of the Study:

• To develop a novel Graph Neural Network (GNN) model for accurate DTA prediction.
• To incorporate molecular structural information, often overlooked in sequence-based methods.
• To evaluate the model's performance on established benchmark datasets.

Main Methods:

• Utilized Graph Neural Networks (GNNs) to process graph representations of drugs and proteins.
• Trained and evaluated the GNN model on the Davis and KIBA benchmark datasets.
• Focused on extracting and utilizing structural information from molecular graph data.

Main Results:

• Achieved competitive performance on both Davis and KIBA datasets.
• Obtained a Mean Squared Error (MSE) of 0.227 and a Concordance Index (CI) of 0.895 on the Davis dataset.
• Achieved an MSE of 0.127 and a CI of 0.903 on the KIBA dataset.

Conclusions:

• The GNN-based approach effectively predicts DTA by utilizing structural information.
• This method shows promise for accelerating the drug discovery pipeline.
• The model demonstrates superior performance compared to existing sequence-based methods.