MHAN-DTA: A Multiscale Hybrid Attention Network for Drug-Target Affinity Prediction.

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Drug-target affinity prediction is crucial for efficient drug discovery.
• Current deep learning methods inadequately represent drug-target interactions, limiting predictive accuracy.

Purpose of the Study:

• To develop an advanced deep learning model, MHAN-DTA, for more effective drug-target affinity prediction.
• To improve the mining of multiscale features within drug-target binding sites.

Main Methods:

• Proposed a Multiscale Hybrid Attention Network (MHAN-DTA) incorporating a pocket-oriented feature aggregation module with self-attention.
• Implemented a hierarchical strategy for target proteins and cross-modal/entity interaction modules.
• Utilized four benchmark datasets for comprehensive model evaluation.

Main Results:

• MHAN-DTA demonstrated superior and robust performance across all tested datasets.
• The model effectively addresses insufficient feature mining in existing drug-target affinity prediction approaches.
• The proposed architecture enhances the global perception and feature extraction capabilities.

Conclusions:

• MHAN-DTA offers a significant advancement in drug-target affinity prediction accuracy and reliability.
• The model's architecture provides a powerful framework for mining complex interactions in drug discovery.
• Publicly available code facilitates further research and application in the field.