MMD-DTA: A Multi-Modal Deep Learning Framework for Drug-Target Binding Affinity and Binding Region Prediction.

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• Drug-target affinity (DTA) prediction is vital for drug development.
• Computer-assisted DTA prediction is a growing field.
• Identifying binding regions offers deeper interaction insights.

Purpose of the Study:

• To develop a multi-modal deep learning framework (MMD-DTA) for predicting DTA and binding regions.
• To enable simultaneous prediction of affinity and interaction sites.
• To improve the interpretability of drug-target interactions.

Main Methods:

• Utilized graph neural networks for drug representation.
• Employed target structural feature extraction networks.
• Integrated feature interaction and fusion modules for prediction.
• Applied unsupervised learning for binding region identification.

Main Results:

• MMD-DTA demonstrated superior performance over existing models in DTA prediction.
• External validation confirmed MMD-DTA's enhanced generalization capabilities.
• The model successfully predicted binding regions, offering interpretable insights.
• MMD-DTA effectively generalized to virtual screening tasks.

Conclusions:

• MMD-DTA provides an effective multi-modal deep learning approach for DTA and binding region prediction.
• Integrating sequence and structural information improves model generalization.
• The model's interpretability aids in understanding drug-target interactions and functional regions.