AntiFold: improved structure-based antibody design using inverse folding
Peer-reviewed| Поле | Значение |
|---|---|
| Authors | NR |
| Year | 2025 |
| Venue / status | Peer-reviewed journal; Bioinformatics Advances; Q1 status NR |
| DOI | 10.1093/bioadv/vbae202 |
| URL | https://doi.org/10.1093/bioadv/vbae202 |
| Preprint URL | NR |
| Tools | AntiFold |
Задача и claimed novelty
| Поле | Значение |
|---|---|
| Main biological task | Antibody-specific fixed-backbone inverse folding and masked CDR/framework redesign |
| Claimed novelty | NR |
Архитектура
| Поле | Значение |
|---|---|
| Model architecture | ESM-IF1 fine-tuned on antibody structures |
| Input representation | NR |
| Encoder / decoder | NR |
| Generative state | Fv/VHH sequences, per-residue probabilities, sampled masked variants |
| Conditioning | CDR/framework masks, fixed residues, optional antigen |
| Sampling | NR |
Training
| Поле | Значение |
|---|---|
| Objectives / losses | NR |
| Dataset | NR |
| Preprocessing | NR |
Inference
NR
Baselines и retrospective benchmarks
| Поле | Значение |
|---|---|
| Baselines | NR |
| Retrospective | Improved CDR recovery and retrospective mutation ranking; not demonstrated affinity improvement |
Prospective experimental validation
| Поле | Значение |
|---|---|
| Prospective | NR |
| Tested designs | No prospective wet-lab validation in primary paper |
| Success criterion | Improved CDR recovery and retrospective mutation ranking; not demonstrated affinity improvement |
| Main results | Improved CDR recovery and retrospective mutation ranking; not demonstrated affinity improvement |
Limitations
Depends on input structure; no prospective therapeutic evidence; likelihood must not be read as affinity.
Что статья показывает
- Improved CDR recovery and retrospective mutation ranking; not demonstrated affinity improvement
Чего статья не показывает
- Predicted confidence, likelihood, pLDDT, docking or developability scores are not experimental affinity.
Размеченные утверждения
| Утверждение | Тип | Локатор |
|---|---|---|
| AntiFold: improved structure-based antibody design using inverse folding | Author-reported | DOI/URL from landscape matrix |
| Publication status: Peer-reviewed journal; Bioinformatics Advances; Q1 status NR | Author-reported | landscape matrix, publication status column |
Source-grounded разбор статьи
Недоступные или неполные источники
| Источник | Почему недоступен | Что использовано вместо |
|---|---|---|
| OUP HTML 10.1093/bioadv/vbae202 | Cloudflare | arXiv 2405.03370 |
Вопрос: улучшает ли antibody fine-tune ESM-IF1 recovery CDR и zero-shot ранжирование affinity-related DMS относительно ProteinMPNN/AbMPNN/ESM-IF1.
Полная библиография
[PAPER]Høie M.H., Hummer A.M., Olsen T.H., Aguilar-Sanjuan B., Nielsen M., Deane C.M. AntiFold: Improved antibody structure-based design using inverse folding. Bioinformatics Advances (2025). DOI 10.1093/bioadv/vbae202. Preprint arXiv:2405.03370. Tool AntiFold.
Исследовательский вопрос
[PAPER]Fixed-backbone antibody sequence design и можно ли IF likelihood использовать как zero-shot proxy для binding-related mutation effects.
Модель
[SI]ESM-IF1 GVP+Transformer, H/L concat with 10-gap, CDR-weighted masking, OAS ABB2 structures. $p(Y\mid X)$ AR. Не structure generator.
Training, split, inference
[PAPER]CE on masked residues. Split 90% concatenated CDR ID with AbMPNN set. Inference temperature 0.20; AAR; ABB2 RMSD; Spearman on Warszawski DMS; rank on Hie 2023 variants.
Baselines и ablations
[SI]ProteinMPNN, ESM-IF1, AbMPNN, ESM-2. Ablations SI S1–S4 on validation CDRH3 AAR (masking, LR decay, OAS, from-scratch).
Prospective evidence
[PAPER]Нет wet-lab designed antibodies в primary paper. Retrospective only.
Утверждения и поддержка
[PAPER]SOTA-like CDR AAR на их test; лучший Spearman на одном Ab DMS; лучшее отделение improved Hie variants. Ablation поддерживает ценность pretrain и OAS для AAR, не для Kd.
Что не поддержано
- Prospective affinity improvement
- Likelihood = experimental affinity
- Developability
- De novo CDR backbone
Ограничения авторов
[PAPER]Performance driven primarily by de-selection of LoF; structure-mediated properties. IgMPNN не сравнивали (weights unavailable).
Дополнительная интерпретация
[INTERPRETATION]ABB2 refold RMSD смешивает IF quality и predictor error (native ABB2 RMSD 0.63). Independent Li 2025 / Clifford 2025 не читались.
SI и repo
[SI]SI: masking, LR decay, AAR/perplexity/temperature equations, DMS protocol. Repo oxpig/AntiFold commit 789d4678; antifold/main.py.
Источники
Нерешённые научные вопросы
- Journal copy-editing diffs vs arXiv