Deep drug-target binding affinity prediction with multiple attention blocks.

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Drug-target interaction (DTI) prediction is crucial for drug discovery.
• Existing computational models often overlook atom correlations and interaction complexities.
• Directly predicting binding affinity is a common regression approach.

Purpose of the Study:

• To propose an end-to-end model for predicting drug-target pair binding affinity.
• To address limitations in existing models regarding atom correlations and interaction modeling.
• To improve the accuracy of DTI prediction.

Main Methods:

• Developed a novel end-to-end model incorporating multiple attention blocks.
• Utilized a relation-aware self-attention block to encode atom correlations.
• Employed a multi-head attention block to model drug-target interactions.
• Investigated the impact of max relative position length (k=3, 5).

Main Results:

• The proposed model significantly outperforms existing methods on benchmark DTI datasets.
• The relation-aware self-attention block effectively captures essential correlation information.
• The multi-head attention block successfully extracts vital interaction information.
• Identified optimal parameters for max relative position length.

Conclusions:

• The novel attention-based model enhances DTI prediction accuracy.
• The model's ability to encode correlations and interactions is key to its superior performance.
• The approach shows promise for predicting binding affinities for COVID-19 related targets and FDA-approved drugs.