DW Darren Wang
Back to projects

Published research / Published

Protein-Peptide Binding Research

AI-assisted protocol for protein-peptide binding pose prediction using AlphaFold 2, ZDOCK, and ANI-2x refinement.

DOI

Research artifact

AI-Assisted Docking Protocol

  1. 01Predict

    Generate initial receptor and complex structures using AlphaFold 2.

  2. 02Dock

    Perform rigid protein-peptide docking using ZDOCK to generate candidate poses.

  3. 03Refine

    Relax poses via AMBER restrained minimization and ANI-2x/CG-BS optimization.

  4. 04Rank

    Select the top docking poses by evaluating their refined ANI-2x potential energies.

Problem

Protein-peptide docking is harder when there is no experimental structure to start from.

Approach

The study combined generated structures, docking candidates, and machine-learning potential refinement to rerank predicted poses.

Result

On 62 challenging systems, the method recovered the correct structure for 34 percent of top-ranked poses and 45 percent within the top three.