SurfDock is a surface-informed diffusion generative model for reliable and accurate protein-ligand complex...

Area of Science:

• Computational Biology
• Structural Biology
• Drug Discovery

Background:

• Accurate prediction of protein-ligand interactions is essential for understanding cellular mechanisms.
• Existing computational methods face limitations in accuracy and generalizability.

Purpose of the Study:

• To introduce SurfDock, a novel deep learning approach for predicting protein-ligand binding poses.
• To evaluate SurfDock's performance against established methods and assess its applicability in real-world scenarios.

Main Methods:

• SurfDock integrates protein sequence, 3D structural graphs, and surface features within an equivariant architecture.
• A generative diffusion model on a non-Euclidean manifold optimizes molecular poses (translations, rotations, torsions).

Main Results:

• SurfDock demonstrates superior docking success rates and physical plausibility compared to existing methods.
• The model shows excellent generalizability to novel proteins and predicted apo structures.
• Achieved state-of-the-art performance in virtual screening tasks, identifying seven novel hit molecules for aldehyde dehydrogenase 1B1.

Conclusions:

• SurfDock offers enhanced accuracy and physical realism in predicting protein-ligand interactions.
• The method has significant potential as a transformative tool in structural biology and drug discovery.
• SurfDock can elucidate molecular mechanisms and accelerate the identification of potential therapeutic compounds.