HuDiff

HuDiff HuDiff-Ab HuDiff-Nb

Preprint Prospective flag Direct antibody flag

Summary card

Поле	Значение
Год / version	2024/2025 / NR
Last verified	2026-08-13
Что генерирует	Humanized variable-domain sequence around supplied CDRs
Форматы	Antibody variable domains and VHH/nanobody
De novo или optimization	Framework generation/humanization
DOI	10.1101/2024.10.22.619416
Paper	https://doi.org/10.1101/2024.10.22.619416
GitHub / repo	https://github.com/TencentAI4S/HuDiff
Service / docs	NR
Лицензия	Custom/NR in repository metadata; commercial review required
Доступность / commercial-use	Code + models

Задача и generated object

Used in these tasks: Humanization, Framework redesign, De novo VHH / nanobody design

Part of these workflows: Framework humanization

Поле	Значение
Required input	CDR sequences from murine antibody or alpaca nanobody
Conditioning	CDRs fixed; human/nanobody priors
Fixed	NR
Editable	NR
Generated	Humanized variable-domain sequence around supplied CDRs
Single mutations	Implicit
Combinatorial mutations	Yes
Multi-objective	Humanness/nativeness; explicit Pareto support NR

Architecture and operating principle

Author-reported (matrix)

Поле	Значение
Base architecture	NR
Model type	7 Humanization/framework redesign generator
Representation	NR
Generation unit	Humanized variable-domain sequence around supplied CDRs
Conditioning	CDRs fixed; human/nanobody priors
Output	Humanized variable-domain sequence around supplied CDRs
Sampling	NR
Internal scorer	Humanness/nativeness metrics
Structure generation	No integrated structure/complex predictor reported

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

CDRs fixed; human/nanobody priors

Generation mechanism

Adaptive autoregressive diffusion; separate antibody and nanobody models

Sampling: NR

Training objectives

Losses: NR

Mathematical details: NR

Integrated evaluation

Humanness/nativeness metrics

Architectural limitations

Preprint.
only three parent series prospectively tested.
retained affinity does not establish reduced immunogenicity/manufacturability.

Uses these architecture components: Sequence diffusion, Autoregressive generation

Generator versus evaluation

Component	Role	Architecture class	Integrated or external	Required or optional	Influences generation or post hoc	Runtime included or excluded
HuDiff-Ab / HuDiff-Nb diffusion	primary_generator	Adaptive autoregressive diffusion; separate antibody and nanobody models	integrated	required	generation	included
Humanness/nativeness metrics	integrated_scorer	NR	integrated	optional	post_hoc	unknown
Binding retention, structure and developability 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 prospective binding validation
Retrospective	NR
Prospective binding	Best antibody variant 0.15 nM vs parent 0.12 nM by BLI; nanobody binding retained in reported study
Prospective function	NR
Developability	NR
Tested design count	One murine antibody and two alpaca nanobody parents prospectively humanized
Success criterion	Best antibody variant 0.15 nM vs parent 0.12 nM by BLI; nanobody binding retained in reported study
Independent validation	NR
Summary	Best antibody variant 0.15 nM vs parent 0.12 nM by BLI; nanobody binding retained in reported study

Limitations and failure modes

Preprint.
only three parent series prospectively tested.
retained affinity does not establish reduced immunogenicity/manufacturability.

Deployment: Limited pilot after reproduction

Reproducibility, license, commercial use

Поле	Значение
Repo	https://github.com/TencentAI4S/HuDiff
License	Custom/NR in repository metadata; commercial review required
Commercial-use status	Code + models

Alternative tools: NR

Complementary: NR

Related: NR

Not directly comparable with

AlphaProteo, BindCraft, BoltzGen, FrameDiff, RFdiffusion, RFdiffusion All-Atom (RFdiffusionAA)

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

Humanization of antibodies and nanobodies by diffusion models

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

Review queue

Bibliographic title must be verified against the PDF.

Детальный source-grounded разбор

1. Что именно делает инструмент

[MATRIX]Вход: CDR sequences from murine antibody or alpaca nanobody

[MATRIX]Generated object: Humanized variable-domain sequence around supplied CDRs

[MATRIX]Фиксируется: NR. Изменяется: NR.

[MATRIX]De novo или parent-based: Framework generation/humanization.

2. Архитектура в одном абзаце

[MATRIX]Adaptive autoregressive diffusion; separate antibody and nanobody models

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: Humanized variable-domain sequence around supplied CDRs. Structure generation: No integrated structure/complex predictor reported.

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): Best antibody variant 0.15 nM vs parent 0.12 nM by BLI; nanobody binding retained in reported study

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 prospective binding validation. Prospective binding: Best antibody variant 0.15 nM vs parent 0.12 nM by BLI; nanobody binding retained in reported study. Function: NR. Developability: NR.

Preprint.
only three parent series prospectively tested.
retained affinity does not establish reduced immunogenicity/manufacturability.

15. Reproducibility

NRРепозиторий: https://github.com/TencentAI4S/HuDiff. Лицензия: Custom/NR in repository metadata; commercial review required. Commit: NR, пока страница не имеет отдельного analysis YAML.

Модульная схема

Модуль	Класс модели	Вход	Выход	Обучается или frozen	Generator или evaluator
HuDiff-Ab / HuDiff-Nb diffusion	Adaptive autoregressive diffusion; separate antibody and nanobody models	NR	NR	NR	primary_generator
Humanness/nativeness metrics	NR	NR	NR	NR	integrated_scorer
Binding retention, structure and developability 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
HuDiff-Ab / HuDiff-Nb diffusion	primary_generator	integrated	required	generation
Humanness/nativeness metrics	integrated_scorer	integrated	optional	post_hoc
Binding retention, structure and developability 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
Best antibody variant 0.15 nM vs parent 0.12 nM by BLI; nanobody binding retained in reported study	Adaptive autoregressive diffusion; separate antibody and nanobody models	5. prospective functional validation	One murine antibody and two alpaca nanobody parents prospectively humanized	NR	Best antibody variant 0.15 nM vs parent 0.12 nM by BLI; nanobody binding retained in reported study	[MATRIX] landscape matrix 2026-08-13	Не читать как полный разбор primary paper.