A drug-target affinity prediction model integrating multimodal feature fusion and structural modeling

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning

Background:

• Predicting drug-target binding affinity (DTA) is vital for computational drug discovery.
• Deep learning advancements offer new opportunities for DTA prediction.

Purpose of the Study:

• To develop a novel DTA prediction framework integrating multimodal feature fusion and structural modeling.
• To enhance the accuracy and interpretability of DTA predictions.

Main Methods:

• Drug representation: ChemBERTa for semantic features and graph neural networks for topology.
• Protein representation: ESM-2 for sequence semantics, geometric vector perceptron, and graph transformer for 3D structure.
• Feature integration: Multi-head attention and gated feature fusion for multimodal data.

Main Results:

• The proposed model significantly outperforms state-of-the-art methods on four benchmark datasets (Davis, KIBA, PDBbind, BindingDB).
• Achieved superior performance in Mean Squared Error (MSE), Concordance Index (CI), and r_m².
• Demonstrated strong generalization and ranking performance, especially on complex datasets like PDBbind and BindingDB.

Conclusions:

• The framework provides a more accurate and interpretable solution for modeling drug-target interactions.
• Offers promising potential for accelerating real-world drug discovery applications.
• The implementation is publicly available for further research and development.