HFGuidedDesign

De Novo Design of Cyclic Peptide Binders via Structure-Guided Discrete Diffusion

Model Framework

Installation

Prerequisites

Before installing HFGuidedDesign, please make sure to install an external structure prediction model.

We recommend installing HighFold:

HighFold GitHub: https://github.com/hongliangduan/HighFold

HFGuidedDesign is designed as a flexible framework. In addition to HighFold, other structure prediction models (e.g., Boltz-2, AlphaFold3, etc.) can also be integrated as external structure evaluators to guide the diffusion process.

Installation HFGuidedDesign

conda create -n HFGuidedDesign python=3.9 -y
conda activate HFGuidedDesign
pip install -r requirements.txt

Pretrained Weights

You can download the weights from:

[https://zenodo.org/records/18768564]

After downloading, place the checkpoint files into the following directory:/checkpoints

Training Discrete Diffusion model

python /models/discrete_diffusion_peptides.py
python /models/discrete_diffusion_complexes.py

Using HFGuidedDesign model

python /models/hfguideddesign.py

Copyright and License

This project is governed by the terms of the MIT License. Prior to utilization, kindly review the LICENSE document for comprehensive details and compliance instructions.

Version History

• v1.0.0