DEELIG: A Deep Learning Approach to Predict Protein-Ligand Binding Affinity.

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Drug Discovery

Background:

• Protein-ligand binding affinity is crucial for understanding molecular interactions and drug design.
• Traditional methods like docking and simulations demand significant computational resources.
• Deep learning offers a powerful approach to analyze complex biological data and identify intrinsic patterns.

Purpose of the Study:

• To develop a deep learning model for predicting protein-ligand binding affinity.
• To enable predictions for entire protein superfamilies and diverse ligands without requiring docked poses.
• To assess the model's performance against existing methods and its applicability to specific complexes, such as COVID-19 protease inhibitors.

Main Methods:

• Incorporation of convolutional neural networks (CNNs) to identify spatial relationships in data.
• Training and validation using a rigorous feature extraction methodology on an in-house protein-ligand dataset.
• Testing the model, named DEELIG, on high-resolution protein crystal structures and nonpeptide ligands.

Main Results:

• The DEELIG model demonstrates superior performance compared to widely used existing methods.
• The model accurately predicts binding affinity for proteins across superfamilies and various ligands.
• DEELIG is suitable for predictions using high-resolution protein crystal structures (≤2.5 Å) and individual nonpeptide ligands.

Conclusions:

• The developed CNN-based approach (DEELIG) provides an efficient and accurate method for predicting protein-ligand binding affinity.
• DEELIG's ability to predict affinity without docked poses or complexes streamlines the drug discovery process.
• DEELIG predictions can enhance biological databases like PDBbind by filling in missing binding affinity data.