MTAF-DTA: multi-type attention fusion network for drug-target affinity prediction

Area of Science:

• Computational chemistry
• Artificial intelligence in drug discovery
• Bioinformatics

Background:

• Drug-target binding affinity (DTA) prediction is crucial for accelerating drug discovery.
• Machine learning models are increasingly used for DTA prediction, offering efficiency over traditional methods.
• Existing DTA prediction methods suffer from drug information loss, unweighted feature contributions, and lack of binding mechanism simulation.

Purpose of the Study:

• To develop an advanced AI-driven method for accurate drug-target binding affinity prediction.
• To overcome limitations of current DTA prediction models, including information loss and inadequate simulation of binding mechanisms.
• To enhance the efficiency and reliability of the drug discovery pipeline.

Main Methods:

• Proposed MTAF-DTA, a novel method for DTA prediction.
• Employed a drug representation module with an attention mechanism to update feature contribution weights across three modalities.
• Introduced a Spiral-Attention Block (SAB) for multi-type attention-based drug-target feature fusion, simulating binding interactions.

Main Results:

• MTAF-DTA demonstrated superior predictive performance on the Davis and KIBA datasets.
• Achieved 1.1% improvement in Concordance Index (CI) and 9.2% improvement in Mean Squared Error (MSE) over state-of-the-art methods in novel target settings.
• Downstream tasks further validated the method's effectiveness in DTA prediction.

Conclusions:

• MTAF-DTA significantly enhances drug-target binding affinity prediction accuracy.
• The method's ability to simulate binding mechanisms and integrate multi-modal features shows great potential for practical drug discovery.
• Results indicate MTAF-DTA's utility in applications like drug discovery and disease treatment.