Estimating protein-ligand interactions with geometric deep learning and mixture density models.

Area of Science:

• Computational chemistry
• Structural biology
• Artificial intelligence in drug discovery

Background:

• Understanding ligand-target interactions is vital for drug design.
• Existing computational methods for predicting binding conformations have limitations.
• Large structural datasets necessitate advanced statistical frameworks.

Purpose of the Study:

• To develop a novel computational method for predicting ligand-protein binding conformations.
• To leverage geometric deep learning for improved drug design workflows.
• To create a statistical potential tailored for specific ligand-target pairs.

Main Methods:

• Developed a geometric deep learning framework using graph neural networks.
• Created graphical representations of proteins to capture binding region properties.
• Trained a statistical potential based on distance likelihood for each ligand-target pair.
• Coupled the potential with global optimization algorithms like differential evolution.

Main Results:

• The method accurately predicts experimental binding conformations of ligands.
• The learned statistical potential performs comparably to or better than established scoring functions.
• Demonstrated effectiveness in docking and screening tasks.
• Showcased the utility of AI in structure-based drug design.

Conclusions:

• The proposed deep learning method offers a powerful new tool for structure-based drug design.
• This approach enhances the prediction of ligand-protein binding, aiding molecular optimization.
• Artificial intelligence can significantly improve computational drug discovery workflows.