RFdiffusion
An open-source diffusion model for generating and designing protein structures, including binders, symmetric oligomers, and motif-scaffolded proteins.
What is RFdiffusion?
RFdiffusion is an open-source diffusion model for generating and designing protein structures. It solves the problem of creating novel proteins with desired functions, such as binding to specific targets or scaffolding functional motifs, by using a denoising process that starts from random noise and iteratively refines it into biologically plausible structures.
Computational biologists, protein engineers, and researchers in structural biology who need to design novel proteins for therapeutic, catalytic, or materials applications.
Developers choose RFdiffusion because it outperforms previous protein design methods in silico and experimentally, offers a wide range of design capabilities (binders, symmetric oligomers, motifs), and is open-source with extensive documentation and community support.
Overview
Code for running RFdiffusion
Use Cases
Best For
- Designing de novo protein binders to specific target sites
- Scaffolding functional motifs like enzyme active sites
- Generating symmetric protein oligomers for nanomaterials
- Diversifying existing protein designs with controlled variation
- Creating macrocyclic peptides for therapeutic applications
- Conditioning protein generation on specific folds or secondary structures
Not Ideal For
- Labs without access to high-performance NVIDIA GPUs and CUDA-compatible environments
- Projects requiring fully integrated sequence and structure generation in a single pipeline
- Researchers needing quick, interactive protein design without deep learning expertise or computational resources
Pros & Cons
Pros
Supports multiple protein design tasks like motif scaffolding, binder design, and symmetric oligomer generation, with detailed examples and protocols in the README.
Modular framework that integrates with tools like ProteinMPNN and AlphaFold2, enabling end-to-end experimental validation pipelines as described in the documentation.
Outperforms previous methods such as Constrained Hallucination and RFjoint Inpainting in tasks like motif scaffolding and binder design, based on the cited paper.
Provides extensive guides, Google Colab notebooks, Docker images, and community resources to facilitate setup and usage, as highlighted in the installation section.
Cons
Installation requires managing Conda environments, SE3-Transformer dependencies, and GPU driver compatibility, with notes on customizing for different CUDA versions, which can be error-prone.
Only produces protein backbones; sequence design necessitates external tools like ProteinMPNN, adding extra steps and complexity to the workflow, as admitted in the binder design section.
Runtime scales O(N^2) with residue count, making large targets slow and computationally expensive, and requires high-performance GPUs, as warned in the practical considerations.
Frequently Asked Questions
Related Projects
AlphaFold3AlphaFold 3 inference pipeline.
Evolutionary Scale Modeling (ESM)Evolutionary Scale Modeling (esm): Pretrained language models for proteins
Boltz-1Official repository for the Boltz biomolecular interaction models
OpenFoldTrainable, memory-efficient, and GPU-friendly PyTorch reproduction of AlphaFold 2
Found a gem we're missing?
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.