SGPocket: A New Graph Convolutional Neural Network for Ligand-Protein Binding Site Prediction.

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Drug discovery is a lengthy and expensive process, often exceeding a decade and substantial investment.
• Computational simulations, such as molecular docking, are crucial for efficiently screening large compound libraries.
• Structure-based molecular docking involves extensive surface exploration and energy calculations to identify optimal binding poses.

Purpose of the Study:

• To address the challenge of identifying ligand-protein binding sites without prior ligand information.
• To reduce the computational burden associated with exploring the entire protein surface in molecular docking.
• To develop a method for predicting potential binding sites on protein surfaces.

Main Methods:

• Developed SGPocket (Spherical Graph Pocket), a novel binding site prediction method.
• Utilized deep learning with a spherical graph convolutional operator based on amino acid positioning.
• Employed a clustering approach to identify and extract predicted binding sites.

Main Results:

• SGPocket demonstrated strong performance when compared to existing binding site prediction methods on a custom dataset.
• The method effectively reduces the protein surface area requiring exploration during molecular docking.
• This reduction in surface exploration leads to decreased computational time for docking simulations.

Conclusions:

• SGPocket successfully predicts relevant binding sites on protein surfaces.
• The method significantly streamlines the molecular docking process by focusing simulations on predicted sites.
• This facilitates faster and more cost-effective drug discovery and development.