Predicting drug-target binding affinity through molecule representation block based on multi-head attention and skip...

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Existing drug-target binding affinity prediction models require improvements in accuracy, robustness, and generalization.
• Most deep learning models lack interpretability and practical application examples.

Purpose of the Study:

• To present a novel, interpretable deep learning model, Molecule Representation Block-based Drug-Target binding Affinity prediction (MRBDTA), for enhanced drug-target binding affinity prediction.
• To improve prediction accuracy, robustness, and generalization ability compared to existing state-of-the-art models.

Main Methods:

• Developed MRBDTA incorporating embedding, positional encoding, a novel molecule representation block (Trans block with skip connections), and an interaction learning module.
• Utilized a multi-head attention mechanism for interpretability analysis to identify key interaction sites.
• Evaluated MRBDTA on two benchmark datasets and performed case studies on SARS-CoV-2 related proteins and approved drugs.

Main Results:

• MRBDTA achieved superior performance compared to 11 state-of-the-art models on benchmark datasets.
• Ablation studies confirmed the effectiveness of the Trans block and skip connections in improving prediction accuracy and reliability.
• Interpretability analysis demonstrated MRBDTA's ability to identify relevant protein-drug interaction sites.
• Case studies showed reliable performance in predicting binding affinities for SARS-CoV-2 targets.

Conclusions:

• MRBDTA offers significant improvements in drug-target binding affinity prediction accuracy and interpretability.
• The model's ability to capture interaction sites and its reliable performance in case studies suggest its utility in drug design, particularly for viral targets like SARS-CoV-2.