Simulating 500 million years of evolution with a language model
Peer-reviewed| Поле | Значение |
|---|---|
| Authors | NR |
| Year | 2025 |
| Venue / status | Peer-reviewed journal; Science (Q1) |
| DOI | 10.1126/science.ads0018 |
| URL | https://doi.org/10.1126/science.ads0018 |
| Preprint URL | NR |
| Tools | ESM3 |
Задача и claimed novelty
| Поле | Значение |
|---|---|
| Main biological task | Multimodal protein generation from sequence/structure/function prompts |
| Claimed novelty | NR |
Архитектура
| Поле | Значение |
|---|---|
| Model architecture | Masked multimodal Transformer; iterative unmasking of sequence, structure and function tracks |
| Input representation | NR |
| Encoder / decoder | NR |
| Generative state | Sequence, structure tokens, function annotations; can co-generate modalities |
| Conditioning | Residue masks, partial sequence/structure, functional keywords/motifs |
| 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 | Functional fluorescent protein demonstrated; not antibody evidence |
| Tested designs | esmGFP and additional design demonstrations; exact total by modality: NR |
| Success criterion | Functional fluorescent protein demonstrated; not antibody evidence |
| Main results | Functional fluorescent protein demonstrated; not antibody evidence |
Limitations
Commercial-use terms differ by asset; multimodal confidence is not functional evidence; complex/antibody proof is limited.
Что статья показывает
- Functional fluorescent protein demonstrated; not antibody evidence
Чего статья не показывает
- Predicted confidence, likelihood, pLDDT, docking or developability scores are not experimental affinity.
Размеченные утверждения
| Утверждение | Тип | Локатор |
|---|---|---|
| Simulating 500 million years of evolution with a language model | 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
Исследовательский вопрос
[MATRIX]Multimodal protein generation from sequence/structure/function prompts
Утверждения и поддержка evidence
[MATRIX]Claimed novelty: NR
[INTERPRETATION]Статья не «доказывает» общее преимущество метода вне своего экспериментального дизайна.
Что представленный evidence не поддерживает
- Predicted confidence, likelihood, pLDDT, docking or developability scores are not experimental affinity.
Ограничения, признанные авторами
[MATRIX]Commercial-use terms differ by asset; multimodal confidence is not functional evidence; complex/antibody proof is limited.
Дополнительные ограничения интерпретации
NRПока Methods/SI не прочитаны, нельзя утверждать split leakage, точные гиперпараметры и полноту ablation.