Accurate and Generalizable Protein-Ligand Binding Affinity Prediction With Geometric Deep Learning.

Area of Science:

• Computational Biology
• Drug Discovery
• Structural Bioinformatics

Background:

• Protein-ligand interactions are vital for biological processes.
• Accurate prediction of binding affinity is crucial for drug design.
• Current methods struggle with novel protein targets.

Purpose of the Study:

• To develop a robust computational method for protein-ligand binding affinity prediction.
• To address the performance decline of existing algorithms on unseen proteins.
• To provide atom-level insights into binding predictions.

Main Methods:

• Developed IPBind, a geometric deep learning approach.
• Leveraged interatomic potential comparing bound and unbound states.
• Validated on established binding affinity prediction benchmarks.

Main Results:

• IPBind demonstrates effectiveness and universality across benchmarks.
• The method shows robust performance even with novel protein targets.
• Achieved atom-level insights into prediction mechanisms.

Conclusions:

• Machine learning-based interatomic potential is advantageous for binding affinity prediction.
• IPBind offers a promising tool for drug discovery and development.
• The study highlights the potential of geometric deep learning in structural bioinformatics.