GraphscoreDTA: optimized graph neural network for protein-ligand binding affinity prediction.

Area of Science:

• Computational chemistry
• Drug discovery and development
• Bioinformatics

Background:

• Accurate prediction of protein-ligand binding affinity is crucial for accelerating drug discovery.
• Current deep learning models face challenges in capturing complex molecular interactions and identifying critical binding sites.
• Existing methods struggle to effectively represent the mutual information between proteins and ligands.

Purpose of the Study:

• To develop an advanced computational model for predicting protein-ligand binding affinity.
• To address the limitations of existing methods in capturing molecular mutual information and highlighting important features.
• To improve the accuracy and reliability of binding affinity predictions for drug development.

Main Methods:

• Development of GraphscoreDTA, a novel graph neural network strategy incorporating Vina distance optimization terms.
• Integration of graph neural networks, a bi-transport information mechanism, and physics-based distance terms.
• Novel approach to effectively capture protein-ligand mutual information and identify key atoms/residues.

Main Results:

• GraphscoreDTA significantly outperforms existing methods on multiple test datasets.
• The model successfully captures mutual information between protein-ligand pairs.
• Key atoms in ligands and residues in proteins critical for binding are effectively highlighted.
• Demonstrated reliability in drug-target selectivity tests on cyclin-dependent kinase and homologous protein families.

Conclusions:

• GraphscoreDTA represents a significant advancement in predicting protein-ligand binding affinity.
• The model's ability to capture intricate molecular interactions and highlight key features makes it a valuable tool for drug discovery.
• GraphscoreDTA offers a reliable and accurate approach for computational drug development.