NIM for RFdiffusion
Type: Microservice Tags: NVIDIA, NIM, BioNeMo, RFdiffusion, protein design, protein structure generation, diffusion model, drug discovery Related: NVIDIA-BioNeMo, NVIDIA-NIM, NIM-for-ProteinMPNN, NIM-for-AlphaFold2, NIM-for-OpenFold3, NIM-for-Boltz2, NVIDIA-Warp, TensorRT, NVIDIA-Clara, NVIDIA-AI-Enterprise, NGC Sources: https://docs.nvidia.com/nim/bionemo/rfdiffusion/latest/overview.html, https://docs.nvidia.com/nim/bionemo/rfdiffusion/latest/index.html, https://docs.nvidia.com/nim/bionemo/rfdiffusion/latest/prerequisites.html Last Updated: 2026-04-29
Summary
NIM for RFdiffusion is NVIDIA’s BioNeMo NIM for generating novel protein structures and complexes. Current NVIDIA docs describe RFdiffusion as a diffusion-based model that refines protein structures from constraints or partial structures and outputs generated 3D protein structures in PDB format.
Detail
Purpose
RFdiffusion supports de novo protein design, binder design, motif scaffolding, and other workflows where researchers need to generate plausible protein structures before designing sequences or validating downstream properties.
Current scope
- Accepts constraints or specifications in formats that include partial protein structures in PDB format.
- Outputs generated 3D protein structures in PDB format.
- Can be used as a first step to generate a binder or scaffold structure.
- Pairs naturally with NIM-for-ProteinMPNN, which can design amino-acid sequences for generated structures.
- Current docs expose endpoints for protein structure generation and readiness checks.
- Current prerequisites describe a single-GPU NIM with minimum GPU memory guidance and TensorRT/Warp optimization in current releases.
NVIDIA context
RFdiffusion is the protein structure generation NIM in the BioNeMo graph. It sits upstream of sequence design with NIM-for-ProteinMPNN and validation/complex prediction with NIM-for-AlphaFold2, NIM-for-OpenFold3, or NIM-for-Boltz2.
Connections
- NIM-for-ProteinMPNN - companion sequence-design NIM for structures generated by RFdiffusion.
- NIM-for-AlphaFold2, NIM-for-OpenFold3, and NIM-for-Boltz2 - structure prediction and validation NIMs adjacent to generated proteins.
- NVIDIA-Warp and TensorRT - current docs call out Warp/TensorRT optimization in RFdiffusion releases.
- NVIDIA-BioNeMo and NVIDIA-Clara - life-sciences and drug-discovery platform context.
- NVIDIA-AI-Enterprise and NGC - enterprise deployment and distribution context.
Source Excerpts
- NVIDIA docs describe RFdiffusion as generating novel protein structures and complexes through a diffusion-based approach.
- The current docs show RFdiffusion feeding ProteinMPNN to design sequences for generated structures.