IgLM: Infilling language modeling for antibody sequence design
Peer-reviewed| Поле | Значение |
|---|---|
| Authors | Richard W. Shuai, Jeffrey A. Ruffolo, Jeffrey J. Gray |
| Year | 2023 |
| Venue / status | Peer-reviewed journal; Cell Systems (Q1) |
| DOI | 10.1016/j.cels.2023.10.001 |
| URL | https://doi.org/10.1016/j.cels.2023.10.001 |
| Preprint URL | NR |
| Tools | IgLM |
Задача и claimed novelty
| Поле | Значение |
|---|---|
| Main biological task | Antibody sequence generation, infilling and CDR diversification |
| Claimed novelty | NR |
Архитектура
| Поле | Значение |
|---|---|
| Model architecture | Autoregressive infilling language model conditioned on species and chain type |
| Input representation | NR |
| Encoder / decoder | NR |
| Generative state | Full heavy/light-chain sequence or infilled span/CDR |
| Conditioning | Chain/species tokens; arbitrary infilling span |
| Sampling | NR |
Training
| Поле | Значение |
|---|---|
| Objectives / losses | NR |
| Dataset | NR |
| Preprocessing | NR |
Inference
NR
Baselines и retrospective benchmarks
| Поле | Значение |
|---|---|
| Baselines | NR |
| Retrospective | Naturalness/diversity/infilling benchmarks |
Prospective experimental validation
| Поле | Значение |
|---|---|
| Prospective | NR |
| Tested designs | No prospective wet-lab validation in primary paper |
| Success criterion | Naturalness/diversity/infilling benchmarks |
| Main results | Naturalness/diversity/infilling benchmarks |
Limitations
No antigen conditioning or joint VH/VL pairing; sequence likelihood is not specificity or affinity.
Что статья показывает
- Naturalness/diversity/infilling benchmarks
Чего статья не показывает
- Predicted confidence, likelihood, pLDDT, docking or developability scores are not experimental affinity.
Размеченные утверждения
| Утверждение | Тип | Локатор |
|---|---|---|
| IgLM: Infilling language modeling for antibody sequence design | Interpretation | DOI/URL from landscape matrix |
| Publication status: Peer-reviewed journal; Cell Systems (Q1) | Author-reported | landscape matrix, publication status column |
Review queue
- Bibliographic title must be verified against the PDF.
Source-grounded разбор статьи
IgLM — infilling LM на unpaired OAS. Evidence: perplexity, controllability, in silico developability. Авторы явно оставляют experimental screening на будущее.
Исследовательский вопрос
[PAPER]Можно ли GPT-2 infilling на OAS генерировать controllable antibody sequences и CDR-H3 libraries с лучшими in silico профилями, чем left-to-right LMs?
Архитектура
[PAPER]ILM span permutation; 12.9M GPT-2; species/chain tags. См. /tools/iglm.html.
Утверждения и evidence
[PAPER]Infilling PPL, ANARCI control, OASis/aggregation in silico на 49 therapeutics. Нет SPR/ELISA IgLM clones.
Что evidence не поддерживает
- Experimental affinity or function of IgLM sequences
- Paired-chain co-design
- Antigen-conditioned generation
Ограничения авторов
[PAPER]OAS N-terminal truncation, human-heavy skew, unpaired chains, rat tag failure, in silico-only developability.
Источники
Нерешённые научные вопросы
- Table S2 prompts