DiffPepDock: Efficient protein-peptide docking and binder screening via SE(3)-equivariant diffusion

Area of Science:

• Computational biology
• Structural biology
• Drug discovery

Background:

• Accurate protein-peptide interaction modeling is vital for understanding biological processes and developing new therapeutics.
• Existing computational methods often struggle with accuracy, generalizability, efficiency, and integrating prior binding information.

Purpose of the Study:

• To introduce DiffPepDock, an efficient and accurate protein-peptide docking tool.
• To enhance protein-peptide interaction modeling by leveraging SE(3)-equivariant diffusion models.
• To provide a practical tool for drug discovery and peptide therapeutic development.

Main Methods:

• Developed DiffPepDock, a tool utilizing SE(3)-equivariant diffusion models for protein-peptide docking.
• Pretrained the model on synthetic protein-fragment complexes and fine-tuned it on experimental protein-peptide structures.
• Incorporated user-specified binding priors, such as motifs or reference ligands, to guide pocket selection and improve accuracy.

Main Results:

• DiffPepDock demonstrates accuracy comparable to state-of-the-art methods like AlphaFold3, with significantly reduced inference time.
• Case studies show successful reconstruction of native binding structures, even in challenging scenarios where AlphaFold3 faced limitations.
• Achieved competitive performance in in silico screening for identifying peptide binders on predicted protein targets.

Conclusions:

• DiffPepDock offers a practical and reliable solution for protein-peptide docking.
• The tool complements existing biomolecular structure prediction methods and aids peptide therapeutic modeling.
• Public availability and interactive demonstrations facilitate broader adoption and application in research.