OneScience-Sugon/RFdiffusion

RFdiffusion

Model Overview

RFdiffusion is a diffusion-based method for protein backbone generation and design. It can be used for unconditional backbone generation, motif scaffolding, PPI/binder design, and symmetric oligomer sampling.

Model Description

RFdiffusion is a generative protein design model based on the RoseTTAFold three-track network and an SE(3)-equivariant denoising diffusion process. It can progressively generate protein backbones from random structures while satisfying specified topology or functional constraints.

The current Hugging Face package is designed for download-and-use workflows, local quick validation, and OneCode automated runtime scenarios. Code, configurations, example inputs, and weights are all included in the current directory.

Use Cases

Use case	Description
Unconditional backbone generation	Takes contig constraints as input and outputs designed backbone PDB files.
Motif scaffolding	Takes a PDB file containing the motif and contig constraints as input, and outputs scaffold design results.
PPI/binder design	Takes the target structure, hotspot, and contig parameters as input, and outputs candidate binder designs.
Symmetric oligomer sampling	Uses symmetry configuration to generate symmetric structure designs.
Hugging Face full-package validation	Uses the package layout config/ modules/ scripts/ examples/ weight/ directly for preflight checks and inference.

Usage

1. Using OneCode

You can try intelligent one-click AI4S programming through the OneCode online environment:

Try intelligent one-click AI4S programming

2. Manual Installation and Usage

Hardware Requirements

• Running on a GPU or DCU is recommended.
• CPU can be used for connectivity checks, but it is relatively slow.
• DCU users need to install DTK in advance. DTK 25.04.2 or later is recommended, or the OneScience-recommended version that matches the current cluster.

Software Requirements

For more information about adaptation details, contact liubiao@sugon.com.

Environment Checks

• NVIDIA GPU:

nvidia-smi

• Hygon DCU:

hy-smi

Quick Start

1. Install the Runtime Environment

conda create -n onescience311 python=3.11 -y
conda activate onescience311
pip install onescience[bio] -i http://mirrors.onescience.ai:3141/pypi/simple/  --trusted-host mirrors.onescience.ai

If the following code cannot find required libraries at runtime, activate CUDA as shown below.

source ${ROCM_PATH}/cuda/env.sh
export LD_LIBRARY_PATH="$CONDA_PREFIX/lib:$LD_LIBRARY_PATH"
export LD_LIBRARY_PATH="$CONDA_PREFIX/lib/python3.11/site-packages/fastpt/torch/lib:$LD_LIBRARY_PATH"

2. Download the Model Package and Install the Environment

hf download --model OneScience-Sugon/RFdiffusion --local-dir ./RFdiffusion

Training Weights

Training weights are already included in the weights folder and can be used directly after downloading the model package.

3. Run Preflight Checks

Check files and real weights:

python scripts/preflight.py --strict-weights

Check local imports after installing dependencies:

python scripts/preflight.py --strict-weights --strict-imports

Validate only the entry point and Hydra configuration without running sampling:

RF_DIFFUSION_SMOKE_TEST=1 python scripts/run_inference.py

4. Run Inference

If execution fails because a .cache file is missing, you can create it manually.

Example of unconditional backbone sampling:

python scripts/run_inference.py \
  'contigmap.contigs=[80-80]' \
  diffuser.T=15 \
  inference.final_step=15 \
  inference.num_designs=1 \
  inference.write_trajectory=False \
  inference.output_prefix=outputs/smoke/design

Example of motif scaffolding:

python scripts/run_inference.py \
  inference.input_pdb=examples/input_pdbs/1YCR.pdb \
  'contigmap.contigs=[10-40/A163-181/10-40]' \
  inference.output_prefix=outputs/motif/design

Example of symmetric sampling:

python scripts/run_inference.py --config-name symmetry \
  diffuser.T=15 \
  inference.final_step=15 \
  inference.output_prefix=outputs/symmetry/c2

5. Common Environment Variables

export RF_DIFFUSION_MODEL_DIR=weight
export RF_DIFFUSION_INPUT_PDB=examples/input_pdbs/1qys.pdb
export RF_DIFFUSION_OUTPUT_PREFIX=outputs/design
export RF_DIFFUSION_SCHEDULE_DIR=.cache/schedules

Official OneScience Information

Platform	OneScience main repository	Skills repository
Gitee	https://gitee.com/onescience-ai/onescience	https://gitee.com/onescience-ai/oneskills
GitHub	https://github.com/onescience-ai/OneScience	https://github.com/onescience-ai/oneskills

Citation and License

RFdiffusion is released under the BSD open-source license (see the LICENSE file) and can be used free of charge for both non-profit and commercial purposes.