RFdiffusion

RFdiffusion RoseTTAFold diffusion

Peer-reviewed Prospective flag

Summary card

Поле	Значение
Год / version	2023 / NR
Last verified	2026-08-13
Что генерирует	Protein backbone / complex geometry; sequence is downstream
Форматы	Monomer, oligomer, protein complex; not antibody by default
De novo или optimization	De novo backbone or partial structural redesign
DOI	10.1038/s41586-023-06415-8
Paper	https://doi.org/10.1038/s41586-023-06415-8
GitHub / repo	https://github.com/RosettaCommons/RFdiffusion
Service / docs	NR
Лицензия	BSD-style governed license; verify bundled components
Доступность / commercial-use	Code + weights

Задача и generated object

Used in these tasks: De novo генерация backbone, Motif scaffolding, Target-conditioned binder generation

Part of these workflows: De novo backbone → inverse folding

Поле	Значение
Required input	None, motif, target structure, hotspot residues, symmetry specification
Conditioning	Contigs, motifs, hotspots, target, symmetry, partial diffusion
Fixed	NR
Editable	NR
Generated	Protein backbone / complex geometry; sequence is downstream
Single mutations	Not primary
Combinatorial mutations	Via downstream MPNN
Multi-objective	Guidance potentials, not general Pareto optimization

Architecture and operating principle

Author-reported (matrix)

Поле	Значение
Base architecture	NR
Model type	3 Backbone/structure + 5 target-conditioned binder generator
Representation	NR
Generation unit	Protein backbone / complex geometry; sequence is downstream
Conditioning	Contigs, motifs, hotspots, target, symmetry, partial diffusion
Output	Protein backbone / complex geometry; sequence is downstream
Sampling	NR
Internal scorer	No universal internal experimental ranker
Structure generation	RoseTTAFold-derived denoiser is generative, not a downstream predictor

Input representation

Поле	Значение
Node features	NR
Edge / pair features	NR

Encoder и decoder

Поле	Значение
Encoder	NR
Decoder	NR

Latent or generative state

Conditioning mechanism

Contigs, motifs, hotspots, target, symmetry, partial diffusion

Generation mechanism

SE(3) diffusion fine-tuned from RoseTTAFold

Sampling: NR

Training objectives

Losses: NR

Mathematical details: NR

Integrated evaluation

No universal internal experimental ranker

Architectural limitations

Backbone-only.
sequence and validation are separate.
input hotspot choice strongly affects outcome.
vanilla model fails on antibody loop geometry.

Uses these architecture components: Frame diffusion, Coordinate diffusion, SE(3)-equivariant networks

Generator versus evaluation

Component	Role	Architecture class	Integrated or external	Required or optional	Influences generation or post hoc	Runtime included or excluded
RFdiffusion backbone model	primary_generator	SE(3) diffusion fine-tuned from RoseTTAFold	integrated	required	generation	included
No universal internal experimental ranker	integrated_scorer	NR	integrated	optional	post_hoc	unknown
RoseTTAFold-derived denoiser is generative, not a downstream predictor	integrated_structure_predictor	NR	integrated	optional	post_hoc	included
ProteinMPNN + AF2/RF self-consistency; interface/functional tests	mandatory_external_dependency	NR	external	required	post_hoc	excluded
Experimental binding/function assays	experimental_validation	NR	external	required	post_hoc	excluded

Training data и biases

Поле	Значение
Training data	NR
Preprocessing	NR
Negative examples	NR
Fine-tuning	NR
Label requirements	NR
Training hardware	NR
Training time	NR

Inference hardware и runtime

Поле	Значение
Candidate count	NR
Batch size	NR
Generation steps	NR
Recycles	NR
Size	NR
Reported runtime	NR
Runtime boundary	NR
CPU	NR
GPU	NR
VRAM	NR
RAM	NR
Disk	NR
Databases	NR

Experimental evidence

Поле	Значение
Evidence class	3–4 prospective binding/functional/structural validation
Retrospective	NR
Prospective binding	Strong general protein evidence; not antibody-specific
Prospective function	Strong general protein evidence; not antibody-specific
Developability	NR
Tested design count	Hundreds of designs across motif, binder and assembly campaigns; task-specific counts in paper
Success criterion	Strong general protein evidence; not antibody-specific
Independent validation	NR
Summary	Strong general protein evidence; not antibody-specific

Limitations and failure modes

Backbone-only.
sequence and validation are separate.
input hotspot choice strongly affects outcome.
vanilla model fails on antibody loop geometry.

Deployment: Limited internal pilot / workhorse backbone stage

Reproducibility, license, commercial use

Поле	Значение
Repo	https://github.com/RosettaCommons/RFdiffusion
License	BSD-style governed license; verify bundled components
Commercial-use status	Code + weights

Alternative tools: NR

Complementary: NR

Related: NR

Not directly comparable with

AntiFold, Efficient evolution with ESM-1v, ESM-IF1, EvoDiff, LigandMPNN, ProGen / ProGen2, ProteinMPNN, Structural evolution (ESM-IF1 complex)

Сравнение допустимо только внутри совместимых категорий. Runtime и quality AntiFold vs RFantibody нельзя читать как одну шкалу: разные starting points и generated objects.

Depends on these external modules

Primary paper и полный список источников

De novo design of protein structure and function with RFdiffusion

Source matrix: landscape xlsx, last verified 2026-08-13.

Детальный source-grounded разбор

1. Что именно делает инструмент

[MATRIX]Вход: None, motif, target structure, hotspot residues, symmetry specification

[MATRIX]Generated object: Protein backbone / complex geometry; sequence is downstream

[MATRIX]Фиксируется: NR. Изменяется: NR.

[MATRIX]De novo или parent-based: De novo backbone or partial structural redesign.

2. Архитектура в одном абзаце

[MATRIX]SE(3) diffusion fine-tuned from RoseTTAFold

4. Математическая модель

NRТочные уравнения, losses и sampling solver будут внесены после чтения paper/code. Не подставляйте родовой UCB/transformer/diffusion, если он не подтверждён для этого инструмента.

Поля NR: conditional probability, update equations, sampling algorithm

5. Физическое устройство состояния

[MATRIX]Seed generated state / output: Protein backbone / complex geometry; sequence is downstream. Structure generation: RoseTTAFold-derived denoiser is generative, not a downstream predictor.

6. Физико-химическая интерпретация

[INTERPRETATION]Координаты, frames, PLM embeddings или residue tokens не являются расчётом ΔG, электростатики или растворителя, пока это не сделано внешней energy function. Likelihood / pLDDT / docking score ≠ экспериментальный binding.

7. Training

[MATRIX]Данные: NR. Fine-tuning: NR. Negative examples: NR.

8. Inference

[MATRIX]Sampling: NR. Generation steps: NR. Runtime boundary: NR.

10. Почему, по мнению авторов, метод эффективен

[MATRIX]Author-reported summary (матрица): Strong general protein evidence; not antibody-specific

NRCausal support не классифицирован, пока не прочитаны ablation и experimental sections.

12. Ablation analysis

NRAblation в этом выпуске не разобрана.

13. What the model does not represent

Явный растворитель и ионная сила — обычно отсутствуют в генераторе.
Протонирование, гликозилирование, кофакторы — обычно отсутствуют, если не оговорено.
Конформационная энтропия и долгая динамика — не моделируются одним forward pass.
Экспрессия, агрегация, иммуногенность — не равны likelihood генератора.
Истинная binding free energy — не равна internal score.

14. Evidence and limitations

[MATRIX]Evidence class (матрица): 3–4 prospective binding/functional/structural validation. Prospective binding: Strong general protein evidence; not antibody-specific. Function: Strong general protein evidence; not antibody-specific. Developability: NR.

Backbone-only.
sequence and validation are separate.
input hotspot choice strongly affects outcome.
vanilla model fails on antibody loop geometry.

15. Reproducibility

NRРепозиторий: https://github.com/RosettaCommons/RFdiffusion. Лицензия: BSD-style governed license; verify bundled components. Commit: NR, пока страница не имеет отдельного analysis YAML.

Модульная схема

Модуль	Класс модели	Вход	Выход	Обучается или frozen	Generator или evaluator
RFdiffusion backbone model	SE(3) diffusion fine-tuned from RoseTTAFold	NR	NR	NR	primary_generator
No universal internal experimental ranker	NR	NR	NR	NR	integrated_scorer
RoseTTAFold-derived denoiser is generative, not a downstream predictor	NR	NR	NR	NR	integrated_structure_predictor
ProteinMPNN + AF2/RF self-consistency; interface/functional tests	NR	NR	NR	NR	mandatory_external_dependency
Experimental binding/function assays	NR	NR	NR	NR	experimental_validation

Generator versus evaluator

Компонент	Роль	Integrated или external	Required или optional	Влияет на generation или только ранжирует после
RFdiffusion backbone model	primary_generator	integrated	required	generation
No universal internal experimental ranker	integrated_scorer	integrated	optional	post_hoc
RoseTTAFold-derived denoiser is generative, not a downstream predictor	integrated_structure_predictor	integrated	optional	post_hoc
ProteinMPNN + AF2/RF self-consistency; interface/functional tests	mandatory_external_dependency	external	required	post_hoc
Experimental binding/function assays	experimental_validation	external	required	post_hoc

Claim–evidence matrix

Утверждение	Предполагаемый механизм	Тип evidence	Датасет / эксперимент	Comparator	Метрика / результат	Источник	Оговорка
Strong general protein evidence; not antibody-specific	SE(3) diffusion fine-tuned from RoseTTAFold	5. prospective functional validation	Hundreds of designs across motif, binder and assembly campaigns; task-specific counts in paper	NR	Strong general protein evidence; not antibody-specific	[MATRIX] landscape matrix 2026-08-13	Не читать как полный разбор primary paper.