Skip to main content
model-family passport · Review date not recorded

RFdiffusion

Diffusion models that hallucinate brand-new proteins.

5/7Evidence fields documented
60-SECOND EVALUATION VIEW

What should a scientist know before using RFdiffusion?

SupportedEvidence supports the stated context with explicit boundaries
Best suited forGeneration · Design
Evidence supportsExperimental binder validation: Peer-reviewed + experimental
Evidence does not establishUniversal superiority, therapeutic success, clinical utility or regulatory acceptance.
Major limitationPerformance depends on the evaluation dataset and operating conditions.
Current registry recordRFdiffusion22 recorded releases · Review date not recorded. A newer version is not assumed to be universally better.

What it is

RFdiffusion applies denoising diffusion to protein backbones, generating novel folds, binders, symmetric assemblies and scaffolds around functional motifs. Paired with ProteinMPNN for sequence design, it is the workhorse of modern de novo protein design and lab-validated binders.

Evidence trail

BioAtlas keeps the path from source to decision visible. A connection records provenance; it does not imply that evidence is sufficient for every context.

Sources4 connectedPrimary resources and normalized claims
Claims1 normalizedDe novo protein or binder design
EntityRFdiffusionmodel-family · RFdiffusion2
ReviewReview date not recordedReview date not claimed
ConclusionContext requiredAdd to an evaluation before operational use

Model passport

Entity typemodel-family
OrganizationInstitute for Protein Design, UW
Model family introduced2025
AccessOpen source
Commercial useAllowed / verify checkpoint terms
DeploymentSelf-hosted
ComputeGPU recommended
Domainsdesign
Biology → representation → computation → evidence

How RFdiffusion represents biology

model-familydesign

Category is navigation. These fields describe the model-specific computational transformation and deliberately override broad category defaults.

1 · Biological inputs
Motif, target, symmetry or design constraints
2 · Input representation
Protein backbone / residue framesOptional atom-level context
3 · Internal representation
SE(3)-equivariant geometric state
4 · Architecture
RoseTTAFold-derived diffusion model
5 · Learning objective
Denoising diffusion
6 · Output representation
Backbone coordinatesAll-atom designs in newer releases

Biological scale

proteincomplex

Modalities & tasks

ProteinMoleculeGenerationDesign

Registry, claims and frontier intelligence

Versioned registry

RFdiffusion2

2 version records · latest curated year 2025. Model-family identity remains separate from capability and access changes.

Explore version lineage →

Connected research frontiers

These records describe active research directions, not guaranteed capabilities of this model. Evidence stages and unresolved questions are preserved separately.

Generative biomolecular design

Prospective de novo binder generation

Google DeepMind · 2024-09-05
Prospective demonstration

Can AI generate high-affinity protein binders for diverse targets with fewer rounds of experimental optimization?

Evidence boundary and unresolved questions

Performance varies by target; the system did not succeed on every attempted target and is not publicly released for unrestricted reproduction.

  • Which target properties predict designability?
  • How transferable are success rates to membrane and flexible targets?
  • Can developability, immunogenicity and function be optimized jointly?
protein binders · prospective validation · generative design · wet labOpen frontier record →

Inputs and outputs

Inputs

Motif, target, symmetry or design constraints

Outputs

Novel protein backbonesDesigned scaffolds / binders / enzymes

Scientific and technical profile

Scientific principles

Denoising diffusionDe novo generative designMotif scaffolding

Technology

RoseTTAFold backboneSE(3) diffusionGuided/conditioned generation
Ideas before algorithms

Scientific lineage

Explore all foundations

These are transparent concept matches—not claims that one scientist alone caused this model. Each connection is based on the model’s recorded domain, scientific principles, technical terms or an explicit lineage link.

Computational intelligence

Denoising diffusion generative models

Jascha Sohl-Dickstein, Jonathan Ho and collaborators

Modern protein-backbone, molecular-pose and biomolecular-complex generators use diffusion to sample valid three-dimensional structures and designs.

Explicit model lineage
Structural biology

Anfinsen’s dogma—the thermodynamic hypothesis

Christian B. Anfinsen

Protein structure prediction, inverse folding and generative protein design all assume that sequence strongly constrains structure and function.

Explicit model lineage

Evaluation evidence

Dataset or evaluationExperimental binder validation
Task or metricDe novo protein design
Evidence statusPeer-reviewed + experimental
Open source ↗

Task-specific evidence only; not comparable as a universal leaderboard score.

De novo protein or binder design

Experimental binder validation

RFdiffusion2 · Prospective designed proteins and binders
experimental

The primary publication includes experimental validation of generated designs.

Claim caveats
  • Experimental success rates depend on design objective and filtering pipeline.
  • A generated backbone still requires sequence design and laboratory confirmation.

Known limitations

  • Performance depends on the evaluation dataset and operating conditions.
  • Task-specific benchmark results should not be compared across unlike domains.
  • Outputs require task-specific scientific and experimental validation.

Milestones

2025

RFdiffusion All-Atom extends to small-molecule binders.

Evidence

From the 2024 Nobel-winning Baker lab.