Protein-ligand binding affinity prediction with edge awareness and supervised attention

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Accurate prediction of protein-ligand binding affinity is essential for structure-based drug design.
• Current deep learning methods face challenges with edge information and capturing true binding interactions in limited datasets.

Purpose of the Study:

• To develop a novel deep learning method, SEGSA-DTA, for enhanced drug-target affinity prediction.
• To address limitations in existing methods by utilizing comprehensive structural information and improved attention mechanisms.

Main Methods:

• Proposed SEGSA-DTA, a SuperEdge Graph convolution-based and Supervised Attention-based Drug-Target Affinity prediction method.
• Employed super edge graph convolution to leverage both node and edge information.
• Utilized a multi-supervised attention module to learn attention distributions aligned with real interactions.

Main Results:

• SEGSA-DTA significantly outperforms existing state-of-the-art methods across multiple datasets.
• The method was successfully applied to repurpose FDA-approved drugs for potential COVID-19 treatments.
• SHapley Additive exPlanations (SHAP) confirmed the interpretability of SEGSA-DTA.

Conclusions:

• SEGSA-DTA offers a robust and interpretable solution for predicting drug-target affinity.
• The approach provides a new quantitative analytical framework for structure-based lead optimization.
• This method advances drug discovery by improving affinity prediction and enabling drug repurposing.