Multi-view self-attention for interpretable drug-target interaction prediction.

Area of Science:

• Computational chemistry
• Machine learning in drug discovery
• Pharmacology

Background:

• Drug discovery is a critical early stage in pharmaceutical development.
• Machine learning (ML) methods are increasingly used for rational drug discovery, particularly for modeling drug-target interactions.
• Effective numerical representation of molecules and model interpretability are key challenges in ML-based drug discovery.

Purpose of the Study:

• To propose a novel self-attention-based multi-view representation learning approach for modeling drug-target interactions.
• To enhance the prediction performance and interpretability of ML models in drug discovery.
• To provide biologically plausible explanations for predicted drug-target interactions.

Main Methods:

• Developed a self-attention-based multi-view representation learning framework.
• Applied the method to model drug-target interactions.
• Utilized three benchmark kinase datasets for evaluation.
• Compared the proposed approach against established baseline models.

Main Results:

• The proposed method achieved competitive prediction performance on benchmark datasets.
• The approach demonstrated the ability to provide biologically plausible interpretations of drug-target interactions.
• Experimental results validate the effectiveness of the self-attention mechanism in this context.

Conclusions:

• The self-attention-based multi-view representation learning approach is effective for modeling drug-target interactions.
• This method offers a promising direction for rational drug discovery by improving both prediction accuracy and interpretability.
• The findings have potential pharmacological applications by providing insights into drug-target relationships.