Configuration

RFDiffusion

RFdiffusion generates protein structures, optionally conditioned on a motif, target or symmetry constraint.

Sources (3)

RosettaCommons/RFdiffusion: README.md; rfdiffusion: Journal full-text XML; rfdiffusion-supp: Publisher supplementary archive · RFdiffusion paper Main text and Extended Data Figure 1 (pretrained representations); README.md: model variants, contigs and licence; Supplementary Methods Sections 1.4 and 4.1, Table 6

1 evaluation · 5 metric rows

How it worksRFdiffusion workflow

Conceptual summary of the documented data flow; optional inputs and configured downstream stages must be reported for a reproducible evaluation.

Sources (3)

Overview

Model type

Diffusion-based protein backbone generator

Sources (3)

Inputs

Unconditional length specification or structural constraints such as a motif, target and contig map.

Sources (3)

Outputs

Generated protein backbone designs for downstream sequence design and assessment.

Sources (3)

Access

Official project documentation and implementation: https://github.com/RosettaCommons/RFdiffusion

Sources (3)

limited source coverage · Automated source review, 2026-09-16. All specifications and missing details

Evaluations and results

1 evaluation · 5 metric rows. Different protocols are not a single leaderboard.

Filter evaluations

Tested model or methodProtocolDatasetEvaluation setupMetricEvidence origin

Applied filters: All linked evaluations

Exact evaluated configurations and original reported results
Tested configuration	Protocol and dataset	Finding	Evidence and details
Configuration: RFDiffusion	Protocol: Genie 3 short monomer generation designability: Unconditional short monomer Designability Dataset subset: 500 generated monomers per run; lengths 50–250 in steps of 50; three runs (Genie 3 short monomer generation split)	0.93 designability reported score · higher Uncertainty: Not reported Coverage: generated per run: 500; repeats: 3; note: Three repeated runs; metric-specific valid subsets not assigned a pooled denominator.	Author-reported evaluation · source checked Methods, coverage and source RFDiffusion on Genie 3 short monomer generation designability: Unconditional short monomer Designability Unconditional short monomer generation: 100 structures at each length 50, 100, 150, 200 and 250 (500 samples per run); 3 repeated runs; Table 3 reports the authors’ average across runs. Designability uses minimum C-alpha scRMSD <2 angstrom across 8 ProteinMPNN sequences refolded with ESMFold. Diversity and novelty use FoldSeek release 10 (2025-01-19); exact definitions and reference sets in Appendix B.1. Method-specific checkpoints and sampling settings remain as Appendix B.2. Aggregation: Mean over three repeated runs Genie 3 primary paper v1, Table 3 · PDF page 20, Appendix B.3, Table 3, data row 1 (RFDiffusion), column Designability
Configuration: RFDiffusion	Protocol: Genie 3 short monomer generation diversity-tm-05: Unconditional short monomer Diversity, TM < 0.5 Dataset subset: 500 generated monomers per run; lengths 50–250 in steps of 50; three runs (Genie 3 short monomer generation split)	0.42 diversity_tm_05 reported score · higher Uncertainty: Not reported Coverage: generated per run: 500; repeats: 3; note: Three repeated runs; metric-specific valid subsets not assigned a pooled denominator.	Author-reported evaluation · source checked Methods, coverage and source RFDiffusion on Genie 3 short monomer generation diversity-tm-05: Unconditional short monomer Diversity, TM < 0.5 Unconditional short monomer generation: 100 structures at each length 50, 100, 150, 200 and 250 (500 samples per run); 3 repeated runs; Table 3 reports the authors’ average across runs. Designability uses minimum C-alpha scRMSD <2 angstrom across 8 ProteinMPNN sequences refolded with ESMFold. Diversity and novelty use FoldSeek release 10 (2025-01-19); exact definitions and reference sets in Appendix B.1. Method-specific checkpoints and sampling settings remain as Appendix B.2. Aggregation: Mean over three repeated runs Genie 3 primary paper v1, Table 3 · PDF page 20, Appendix B.3, Table 3, data row 1 (RFDiffusion), column Diversity, TM < 0.5
Configuration: RFDiffusion	Protocol: Genie 3 short monomer generation diversity-tm-06: Unconditional short monomer Diversity, TM < 0.6 Dataset subset: 500 generated monomers per run; lengths 50–250 in steps of 50; three runs (Genie 3 short monomer generation split)	0.69 diversity_tm_06 reported score · higher Uncertainty: Not reported Coverage: generated per run: 500; repeats: 3; note: Three repeated runs; metric-specific valid subsets not assigned a pooled denominator.	Author-reported evaluation · source checked Methods, coverage and source RFDiffusion on Genie 3 short monomer generation diversity-tm-06: Unconditional short monomer Diversity, TM < 0.6 Unconditional short monomer generation: 100 structures at each length 50, 100, 150, 200 and 250 (500 samples per run); 3 repeated runs; Table 3 reports the authors’ average across runs. Designability uses minimum C-alpha scRMSD <2 angstrom across 8 ProteinMPNN sequences refolded with ESMFold. Diversity and novelty use FoldSeek release 10 (2025-01-19); exact definitions and reference sets in Appendix B.1. Method-specific checkpoints and sampling settings remain as Appendix B.2. Aggregation: Mean over three repeated runs Genie 3 primary paper v1, Table 3 · PDF page 20, Appendix B.3, Table 3, data row 1 (RFDiffusion), column Diversity, TM < 0.6
Configuration: RFDiffusion	Protocol: Genie 3 short monomer generation novelty-afdb: Unconditional short monomer Novelty, AFDB Dataset subset: 500 generated monomers per run; lengths 50–250 in steps of 50; three runs (Genie 3 short monomer generation split)	0.35 novelty_afdb reported score · higher Uncertainty: Not reported Coverage: generated per run: 500; repeats: 3; note: Three repeated runs; metric-specific valid subsets not assigned a pooled denominator.	Author-reported evaluation · source checked Methods, coverage and source RFDiffusion on Genie 3 short monomer generation novelty-afdb: Unconditional short monomer Novelty, AFDB Unconditional short monomer generation: 100 structures at each length 50, 100, 150, 200 and 250 (500 samples per run); 3 repeated runs; Table 3 reports the authors’ average across runs. Designability uses minimum C-alpha scRMSD <2 angstrom across 8 ProteinMPNN sequences refolded with ESMFold. Diversity and novelty use FoldSeek release 10 (2025-01-19); exact definitions and reference sets in Appendix B.1. Method-specific checkpoints and sampling settings remain as Appendix B.2. Aggregation: Mean over three repeated runs Genie 3 primary paper v1, Table 3 · PDF page 20, Appendix B.3, Table 3, data row 1 (RFDiffusion), column Novelty, AFDB
Configuration: RFDiffusion	Protocol: Genie 3 short monomer generation novelty-pdb: Unconditional short monomer Novelty, PDB Dataset subset: 500 generated monomers per run; lengths 50–250 in steps of 50; three runs (Genie 3 short monomer generation split)	0.36 novelty_pdb reported score · higher Uncertainty: Not reported Coverage: generated per run: 500; repeats: 3; note: Three repeated runs; metric-specific valid subsets not assigned a pooled denominator.	Author-reported evaluation · source checked Methods, coverage and source RFDiffusion on Genie 3 short monomer generation novelty-pdb: Unconditional short monomer Novelty, PDB Unconditional short monomer generation: 100 structures at each length 50, 100, 150, 200 and 250 (500 samples per run); 3 repeated runs; Table 3 reports the authors’ average across runs. Designability uses minimum C-alpha scRMSD <2 angstrom across 8 ProteinMPNN sequences refolded with ESMFold. Diversity and novelty use FoldSeek release 10 (2025-01-19); exact definitions and reference sets in Appendix B.1. Method-specific checkpoints and sampling settings remain as Appendix B.2. Aggregation: Mean over three repeated runs Genie 3 primary paper v1, Table 3 · PDF page 20, Appendix B.3, Table 3, data row 1 (RFDiffusion), column Novelty, PDB

Source checking is not independent reproduction. Release 2026-09-23-2b89723c6dd9.

Use this model

How it works, versions and access

Related profile: RFdiffusion. This page retains the exact record and its evaluation context.

This configuration

Author-evaluated generation configuration; checkpoint/version and sampling parameters remain those described in Appendix B.2.

record: RFDiffusion
configuration: Not reported
entity type: Configuration

How it works

RFdiffusion generates protein structures, optionally conditioned on a motif, target or symmetry constraint. Diffusion-based protein structure generation with checkpoint-specific conditioning and denoising configuration. The documented inputs are unconditional length specification or structural constraints such as a motif, target and contig map. The output consists of generated protein backbone designs for downstream sequence design and assessment.

Sources (3)

Versions and reproducibility

Base, active-site, sequence-inpainting and other conditioning-specific checkpoints; preserve the selected weight identity. The RFdiffusion training crop is 384 residues (supplementary Table 6). This crop size is not an inference maximum; contig lengths and conditional task settings remain explicit.

Sources (3)

Strengths, limitations and unresolved questions

Strengths and limitations

Strengths and considerations

Supports motif scaffolding, symmetry, binder design and partial diffusion through explicit conditioning options.
Sources (3)
RosettaCommons/RFdiffusion: README.md; rfdiffusion: Journal full-text XML; rfdiffusion-supp: Publisher supplementary archive · RFdiffusion paper Main text and Extended Data Figure 1 (pretrained representations); README.md: model variants, contigs and licence; Supplementary Methods Sections 1.4 and 4.1, Table 6

Limitations and conditions

Different conditioning modes use different trained weights. Backbone generation alone does not establish a functional experimentally validated protein.
Sources (3)
RosettaCommons/RFdiffusion: README.md; rfdiffusion: Journal full-text XML; rfdiffusion-supp: Publisher supplementary archive · RFdiffusion paper Main text and Extended Data Figure 1 (pretrained representations); README.md: model variants, contigs and licence; Supplementary Methods Sections 1.4 and 4.1, Table 6

Profile review details

Inspected pinned official documentation, relevant implementation files and named primary-paper sections. Claims are limited to those artifacts. Remaining field extraction and identity conflicts are explicit; no new performance claims, model runs or human review are implied.

Stable record: discovery-model-rfdiffusion

Specifications

Inputs, training, access and other details

Explanatory profile: limited source coverage · Automated source review, 2026-09-16. Review applies to the cited claims; unresolved fields are listed below. Numerical results retain their own review status.

Inputs, outputs and configuration
Property	Description and evidence
Model type	Diffusion-based protein backbone generator Sources (3) RosettaCommons/RFdiffusion: README.md; rfdiffusion: Journal full-text XML; rfdiffusion-supp: Publisher supplementary archive · RFdiffusion paper Main text and Extended Data Figure 1 (pretrained representations); README.md: model variants, contigs and licence; Supplementary Methods Sections 1.4 and 4.1, Table 6
Architecture	Diffusion-based protein structure generation with checkpoint-specific conditioning and denoising configuration. Sources (3) RosettaCommons/RFdiffusion: README.md; rfdiffusion: Journal full-text XML; rfdiffusion-supp: Publisher supplementary archive · RFdiffusion paper Main text and Extended Data Figure 1 (pretrained representations); README.md: model variants, contigs and licence; Supplementary Methods Sections 1.4 and 4.1, Table 6
Inputs	Unconditional length specification or structural constraints such as a motif, target and contig map. Sources (3) RosettaCommons/RFdiffusion: README.md; rfdiffusion: Journal full-text XML; rfdiffusion-supp: Publisher supplementary archive · RFdiffusion paper Main text and Extended Data Figure 1 (pretrained representations); README.md: model variants, contigs and licence; Supplementary Methods Sections 1.4 and 4.1, Table 6
Outputs	Generated protein backbone designs for downstream sequence design and assessment. Sources (3) RosettaCommons/RFdiffusion: README.md; rfdiffusion: Journal full-text XML; rfdiffusion-supp: Publisher supplementary archive · RFdiffusion paper Main text and Extended Data Figure 1 (pretrained representations); README.md: model variants, contigs and licence; Supplementary Methods Sections 1.4 and 4.1, Table 6
Parameters	The complete supplementary architecture and training sections specify RoseTTAFold-derived modules and training settings but do not state a total for each released conditional checkpoint. · Not reported in inspected sources Sources (3) RosettaCommons/RFdiffusion: README.md; rfdiffusion: Journal full-text XML; rfdiffusion-supp: Publisher supplementary archive · RFdiffusion paper Main text and Extended Data Figure 1 (pretrained representations); README.md: model variants, contigs and licence; Supplementary Methods Sections 1.4 and 4.1, Table 6
Known versions	Base, active-site, sequence-inpainting and other conditioning-specific checkpoints; preserve the selected weight identity. Sources (3) RosettaCommons/RFdiffusion: README.md; rfdiffusion: Journal full-text XML; rfdiffusion-supp: Publisher supplementary archive · RFdiffusion paper Main text and Extended Data Figure 1 (pretrained representations); README.md: model variants, contigs and licence; Supplementary Methods Sections 1.4 and 4.1, Table 6
Training data	Fine-tunes pretrained RoseTTAFold to denoise protein backbone structures from the PDB; unconditional and task-conditioned variants are distinct configurations. Sources (3) RosettaCommons/RFdiffusion: README.md; rfdiffusion: Journal full-text XML; rfdiffusion-supp: Publisher supplementary archive · RFdiffusion paper Main text and Extended Data Figure 1 (pretrained representations); README.md: model variants, contigs and licence; Supplementary Methods Sections 1.4 and 4.1, Table 6
Training cutoff	The supplementary RoseTTAFold pretraining description specifies a 2 August 2021 PDB cutoff and additional AlphaFold2 models. This is pretraining provenance, not a date for every conditional design fine-tune. Sources (3) RosettaCommons/RFdiffusion: README.md; rfdiffusion: Journal full-text XML; rfdiffusion-supp: Publisher supplementary archive · RFdiffusion paper Main text and Extended Data Figure 1 (pretrained representations); README.md: model variants, contigs and licence; Supplementary Methods Sections 1.4 and 4.1, Table 6
Context limits	The RFdiffusion training crop is 384 residues (supplementary Table 6). This crop size is not an inference maximum; contig lengths and conditional task settings remain explicit. Sources (3) RosettaCommons/RFdiffusion: README.md; rfdiffusion: Journal full-text XML; rfdiffusion-supp: Publisher supplementary archive · RFdiffusion paper Main text and Extended Data Figure 1 (pretrained representations); README.md: model variants, contigs and licence; Supplementary Methods Sections 1.4 and 4.1, Table 6
Weights licence	BSD licence in the inspected LICENSE explicitly covers both source code and linked downloadable model weights. Sources (3) RosettaCommons/RFdiffusion: README.md; rfdiffusion: Journal full-text XML; rfdiffusion-supp: Publisher supplementary archive · RFdiffusion paper Main text and Extended Data Figure 1 (pretrained representations); README.md: model variants, contigs and licence; Supplementary Methods Sections 1.4 and 4.1, Table 6
Access	Official project documentation and implementation: https://github.com/RosettaCommons/RFdiffusion Sources (3) RosettaCommons/RFdiffusion: README.md; rfdiffusion: Journal full-text XML; rfdiffusion-supp: Publisher supplementary archive · RFdiffusion paper Main text and Extended Data Figure 1 (pretrained representations); README.md: model variants, contigs and licence; Supplementary Methods Sections 1.4 and 4.1, Table 6
Code licence	BSD-3-Clause Sources RosettaCommons/RFdiffusion: LICENSE · LICENSE: licence text

Evidence

Source checking verifies the cited claim or transcription. It does not establish independent reproduction.

Evidence table

Inspect claims, sources and review details

Trace each statement to its source and review. A context-only reference supports the record generally; it does not verify an individual field. Source checking does not reproduce an experiment.

One row per statement and cited source. Multiple citations are not independent evaluations. Shared locators are labelled explicitly.

1 evidence row matching the loaded filters

Claims, original sources and review scope · Release 2026-09-23-2b89723c6dd9
Property and statement	Original source and location	Review and provenance
Relationship: family discovery-model-rfdiffusion Individual claims	Genie 3 primary paper v1, Table 3 Original source ↗ Appendix B.2–B.3; Table 3, row RFDiffusion Version: 10.64898/2026.05.01.722168v1; posted 2026-05-05 Retrieved: 2026-09-23T11:22:04.377971+00:00	source checked automated source review · 2026-09-23 Audit details Source-backed evaluated identity only; no independent reproduction. Field: `links:family:discovery-model-rfdiffusion` Claim: genie3-2026-table3-method-rfdiffusion-discovery-model-rfdiffusion-identity-claim Source artifact SHA-256: `43de8eab49ac74ad9141fbe955703a4e5f7879e0a83cbb547de37fc4e1d3e089` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact

All evidence (CSV)All evidence (JSONL)How to read the evidence

Sources and history

View linked audit checks and correction history

Release 2026-09-23-2b89723c6dd9 · Record review: source checked

1 source records and release history

Genie 3 primary paper v1, Table 3 · Original source · 10.64898/2026.05.01.722168v1; posted 2026-05-05

Download this release

Technical metadata and extraction receipts

Stable ID: genie3-2026-table3-method-rfdiffusion

areas: proteins-complexes
source locator: Appendix B.2–B.3; Table 3, row RFDiffusion
missing metadata: checkpoint revision: unreported; parameters: unextracted

Related records

Suggest a correction