Robust deep learning-based protein sequence design using ProteinMPNN
Peer-reviewed| Поле | Значение |
|---|---|
| Authors | Justas Dauparas, Ivan Anishchenko, Nathaniel Bennett, Hua Bai, Robert J. Ragotte, Lukas F. Milles, Basile I. M. Wicky, Alexis Courbet, Robbert J. de Haas, Neville Bethel, Peggy J. Y. Leung, Timothy F. Huddy, Sam Pellock, Doug Tischer, F. Chan, Brian Koepnick, Hannah Nguyen, Alex Kang, Banumathi Sankaran, Asim K. Bera, Neil P. King, David Baker |
| Year | 2022 |
| Venue / status | Peer-reviewed journal; Science (Q1) |
| DOI | 10.1126/science.add2187 |
| URL | https://doi.org/10.1126/science.add2187 |
| Preprint URL | https://doi.org/10.1101/2022.06.03.494563 |
| Tools | ProteinMPNN |
Задача и claimed novelty
| Поле | Значение |
|---|---|
| Main biological task | Fixed-backbone sequence design / inverse folding |
| Claimed novelty | NR |
Архитектура
| Поле | Значение |
|---|---|
| Model architecture | Autoregressive message-passing neural network on backbone geometry |
| Input representation | NR |
| Encoder / decoder | NR |
| Generative state | One or many sequences compatible with input backbone |
| Conditioning | Fixed positions, chain design masks, amino-acid bias, tied positions, multi-chain context |
| Sampling | NR |
Training
| Поле | Значение |
|---|---|
| Objectives / losses | NR |
| Dataset | NR |
| Preprocessing | NR |
Inference
NR
Baselines и retrospective benchmarks
| Поле | Значение |
|---|---|
| Baselines | NR |
| Retrospective | NR |
Prospective experimental validation
| Поле | Значение |
|---|---|
| Prospective | X-ray/cryo-EM/functional rescue and design evidence; general proteins |
| Tested designs | Multiple monomer, oligomer, nanoparticle and binder campaigns; exact aggregate: NR |
| Success criterion | X-ray/cryo-EM/functional rescue and design evidence; general proteins |
| Main results | X-ray/cryo-EM/functional rescue and design evidence; general proteins |
Limitations
Cannot invent a backbone; likelihood is not affinity; fixed backbone can lock in incorrect interface geometry.
Что статья показывает
- X-ray/cryo-EM/functional rescue and design evidence; general proteins
Чего статья не показывает
- Predicted confidence, likelihood, pLDDT, docking or developability scores are not experimental affinity.
Размеченные утверждения
| Утверждение | Тип | Локатор |
|---|---|---|
| Robust deep learning-based protein sequence design using ProteinMPNN | Author-reported | DOI/URL from landscape matrix |
| Publication status: Peer-reviewed journal; Science (Q1) | Author-reported | landscape matrix, publication status column |
Source-grounded разбор статьи
Статья валидирует ProteinMPNN как inverse-folding sequence designer с retrospective recovery и prospective structural/functional rescue. Likelihood и AF2 confidence не равны affinity.
Исследовательский вопрос
[PAPER]Можно ли обучить AR MPNN на backbone geometry так, чтобы sequences fold/assemble/bind лучше, чем Rosetta packing, на monomers, oligomers, nanoparticles и одном peptide-scaffolded binder?
Архитектура и математика
[PAPER]3+3 MPNN, hidden 128, kNN distances N/Cα/C/O/Cβ, encoder edge updates, order-agnostic decode, tied logits. См. /tools/proteinmpnn.html.
Утверждения и evidence
[PAPER]52.4% vs 32.9% recovery vs Rosetta (retrospective). Table 1 ablations. Prospective: soluble yield, SEC oligomer, X-ray 8CYK 2.35 Å, nanoparticle 1.2 Å, Grb2 BLI rescue. Не antibody campaign.
Что evidence не поддерживает
- De novo backbone generation
- Kd матрица для SH3 binder (численный Kd NR в HTML)
- Antibody CDR design
- Что log p = affinity
Ограничения авторов
[PAPER]In silico recovery чувствителен к resolution; design success ultimately experimental; DL lacks physical transparency of Rosetta.
Источники
Нерешённые научные вопросы
- Полный Methods/SI