IgLM
IgLM
Peer-reviewedSummary card
| Поле | Значение |
|---|---|
| Год / version | 2023 / NR |
| Last verified | 2026-08-27 |
| Что генерирует | Full heavy/light-chain sequence or infilled span/CDR |
| Форматы | Heavy or light variable-domain sequence; chains generated independently |
| De novo или optimization | De novo sequence or local CDR redesign |
| DOI | 10.1016/j.cels.2023.10.001 |
| Paper | https://doi.org/10.1016/j.cels.2023.10.001 |
| GitHub / repo | https://github.com/Graylab/IgLM |
| Service / docs | NR |
| Лицензия | JHU Academic Software License; commercial license required |
| Доступность / commercial-use | Code + weights; non-commercial by default |
Задача и generated object
Used in these tasks: Генерация CDR sequence, Structure-aware дизайн antibody sequence
Part of these workflows: Antibody sequence infilling and CDR diversification
| Поле | Значение |
|---|---|
| Required input | Optional sequence context, span mask, species and chain type |
| Conditioning | Chain/species tokens; arbitrary infilling span |
| Fixed | NR |
| Editable | NR |
| Generated | Full heavy/light-chain sequence or infilled span/CDR |
| Single mutations | Yes |
| Combinatorial mutations | Yes |
| Multi-objective | No |
Architecture and operating principle
Author-reported (Cell Systems)| Поле | Значение |
|---|---|
| Base architecture | Autoregressive infilling Transformer |
| Model type | Antibody sequence LM |
| Representation | Tokens |
| Generation unit | Amino-acid token |
| Conditioning | Species, chain type, span context |
| Output | Sequence |
| Sampling | Autoregressive |
| Internal scorer | Likelihood |
| Structure generation | Нет |
Input representation
Sequence tokens only. No antigen, no coordinates, no paired VH/VL tokenisation in the base generator.
| Поле | Значение |
|---|---|
| Node features | NR |
| Edge / pair features | NR |
Encoder и decoder
Transformer LM (decoder-style infilling)
| Поле | Значение |
|---|---|
| Encoder | Transformer LM (decoder-style infilling) |
| Decoder | Autoregressive token generation into the masked span |
Latent or generative state
Discrete amino-acid tokens
Conditioning mechanism
Species token, chain type, bidirectional context around the span
Generation mechanism
Infilling uses both left and right context of a span, unlike left-to-right ProGen-style models.
Sampling: Temperature categorical sampling
Training objectives
Language-modelling / infilling CE on unpaired antibody sequences
Losses: Cross-entropy
Mathematical details: NR
Integrated evaluation
Likelihood/naturalness only. Pairing, structure, developability and binding are external.
Architectural limitations
| Архитектурная особенность | Возможное ограничение |
|---|---|
| Sequence-only generation | Нет прямой геометрической согласованности. |
| Independent chain generation | Не обеспечивает joint VH/VL pairing. |
| No antigen conditioning | Не target-specific generator. |
Uses these architecture components: Protein language models, Autoregressive generation, Masked-token infilling
Generator versus evaluation
| Component | Role | Architecture class | Integrated or external | Required or optional | Influences generation or post hoc | Runtime included or excluded |
|---|---|---|---|---|---|---|
| IgLM | primary_generator | Autoregressive infilling language model conditioned on species and chain type | integrated | required | generation | included |
| Model likelihood | integrated_scorer | NR | integrated | optional | post_hoc | unknown |
| Pairing, structure/complex, developability and experimental validation | 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 | 2 retrospective antibody evaluation |
| Retrospective | No prospective wet-lab validation in primary paper |
| Prospective binding | NR |
| Prospective function | NR |
| Developability | NR |
| Tested design count | No prospective wet-lab validation in primary paper |
| Success criterion | Naturalness/diversity/infilling benchmarks |
| Independent validation | NR |
| Summary | Naturalness/diversity/infilling benchmarks |
Limitations and failure modes
- No antigen conditioning or joint VH/VL pairing.
- sequence likelihood is not specificity or affinity.
Deployment: Retrospective only pending license
Reproducibility, license, commercial use
| Поле | Значение |
|---|---|
| Repo | https://github.com/Graylab/IgLM |
| License | JHU Academic Software License; commercial license required |
| Commercial-use status | Code + weights; non-commercial by default |
Related tools
Alternative tools: PALM-H3, ProGen / ProGen2
Complementary: AntiFold, Humatch
Related: PALM-H3, ProGen / ProGen2, AntiFold
Not directly comparable with
AlphaProteo, BindCraft, BoltzGen, FrameDiff, RFdiffusion, RFdiffusion All-Atom (RFdiffusionAA)Depends on these external modules
NR
Related glossary terms
Protein language model (PLM), Infilling, CDR, Likelihood / log-probabilityPrimary paper и полный список источников
IgLM: Infilling language modeling for antibody sequence designSource matrix: landscape xlsx, last verified 2026-08-13.
Review queue
- Bibliographic title must be verified against the PDF.
Детальный source-grounded разбор
Недоступные или неполные источники
| Источник | Почему недоступен | Что использовано вместо |
|---|---|---|
| Cell Systems STAR supplemental PDF (Table S1 species counts, Table S2 prompt motifs) | S-tables не извлечены построчно | Основной HTML: 809M → 588M cluster → 558M train / 1M val / 5% test; N-terminal truncation >40% OAS |
IgLM — sequence infilling generator для unpaired VH/VL. GPT-2 decoder обучается на перестановке span (ILM): маска 10–20 residues уходит в конец после [SEP]. Это не inverse folding, не structure diffusion и не affinity predictor. In silico developability / OASis / AF pLDDT $\neq$ experimental binding.
Происхождение источников
| Метка | Источник | Ссылка | Примечание |
|---|---|---|---|
| [PAPER] | Shuai, Ruffolo, Gray. IgLM: Infilling language modeling for antibody sequence design. Cell Systems 14, 979–989 (2023). | https://doi.org/10.1016/j.cels.2023.10.001 | NR |
| [PAPER] | bioRxiv 10.1101/2021.12.13.472419 (preprint of the same work) | https://doi.org/10.1101/2021.12.13.472419 | NR |
| [CODE] | Graylab/IgLM | https://github.com/Graylab/IgLM | commit de819b56238b5cb9c23d7a6fec4c55d7fc0395e4 |
| Поле | Значение |
|---|---|
| Репозиторий | https://github.com/Graylab/IgLM |
| Commit | de819b56238b5cb9c23d7a6fec4c55d7fc0395e4 |
| Ветка / tag | main |
| Лицензия | JHU Academic Software License (non-commercial); commercial via JHTV |
Проверенные файлы реализации:
iglm/model/IgLM.pyiglm/model/tokens.py (imported; not fully re-read)
1. Что именно делает инструмент
[CODE]Вход (infill): родительская unpaired sequence, chain token ([HEAVY]/[LIGHT]), species token, полуинтервал [start, end) для span. Вход (generate): только conditioning tags ± optional prompt. IgLM.infill / generate
[PAPER]Выход: синтетические AA sequences (полная цепь или infilled span). Нет координат, нет paired VH–VL в одном forward (цепи независимы).
[PAPER]Parent-based CDR diversification или species-conditioned full-chain generation. Не de novo docked antibody.
2. Архитектура в одном абзаце
[PAPER]Decoder-only GPT-2 (HuggingFace GPT2LMHeadModel). IgLM: 4 layers, $d=512$, 8 heads, FFN 2048, 12,889,600 params. IgLM-S: 3 layers, $d=192$, 6 heads, FFN 768, 1,439,616. Токенизатор BertTokenizerFast по vocab.txt (AA + specials). Infilling Donahue-style: span $S$ заменяется [MASK], исходная последовательность + [SEP] + $S$ + [ANS]. Table 1; STAR Methods; CODE from_pretrained
4. Математическая модель
[ANS]. $A\setminus S$ содержит фланги, [MASK], [SEP].| Символ | Тип | Определение |
|---|---|---|
| $A$ | token sequence | антитело VH или VL |
| $S$ | span | маскированный участок 10–20 residues при train; при infill — произвольный пользовательский диапазон |
| $c_c,c_s$ | discrete tags | chain and species |
temperature=1, top_p=1, max_length=150.[SEP] (Eq. 4).5. Физическое устройство состояния
[INTERPRETATION]Состояние — дискретная unpaired antibody sequence. Нет атомов, frames, antigen. Species/chain tags — датасетные метки OAS, не 3D. Перестановка span — NLP-трюк, не биологический splicing (V(D)J уже произошёл в train sequences).
7. Training
[PAPER]OAS unpaired: после фильтра ошибок 809M unique → LinClust 95% → 588M. Random 5% test; 558M train, 1M val. Skew: human heavy ~70%. Split не germline-holdout: авторы сознательно оставляют все conserved regions в train. STAR: Antibody sequence dataset
[PAPER]Batch 512, 2 grad accum, DeepSpeed, ~3 days × 4×A100. Objective — стандартный AR NLL на перестроенной последовательности (включая infilled span).
8. Inference
[PAPER]generate: sample until [ANS] (eos = CLS token в коде), bad_words_ids, duplicate filter. Full-length sequences из ILM содержат [MASK]/[SEP]/[ANS] и пересобираются. Prompt из 3 N-terminal residues (Table S2, NR exact motifs here) из-за >40% OAS truncations. Full-length antibody generation; CODE _generate
[PAPER]Infill библиотек в статье: 49 Thera-SAbDab mAbs, CDR H3 по Chothia удалён, 1000 seq × {T∈0.8,1.0,1.2} × {P∈0.5,0.75,1.0} = 9000/parent. AF-Multimer pLDDT — внешний predictor.
10. Почему, по мнению авторов, метод эффективен
[PAPER]Mechanism. Bidirectional context для CDR (фланги видны до генерации span) vs left-to-right ProGen2-OAS; species/chain tags vs prompting.
[PAPER]Author claim. Generates full-length and infilled libraries resembling natural antibodies with tunable diversity and improved in silico developability/humanness vs random; could then be experimentally screened. Discussion: 'which could then be experimentally screened'
[PAPER]Evidence. Infilling PPL vs ProGen2 (Fig. 1D); ANARCI species/chain controllability; germline identity vs T; in silico aggregation/solubility/OASis on 49 therapeutics. Нет prospective binding/function assay IgLM-designed clones в primary paper.
[INTERPRETATION]Causal support. Bidirectional vs preceding-only PPL — ablation-like. Developability — in silico scores, не wet-lab. Causal claim 'reduces non-functional fraction vs random libraries' не подтверждён экспериментом в этой статье.
12. Ablation analysis
[PAPER]IgLM vs IgLM-S (capacity). Bidirectional vs preceding-only infill PPL (Fig. 1D). Temperature / nucleus vs diversity, length, OASis (Figs. 2–3). Нет ablation 'remove species token' с таблицей recovery в извлечённом HTML (NR beyond qualitative rat failure).
13. What the model does not represent
- Структура, antigen, epitope, dock
- Paired chain covariation (unpaired OAS)
- Растворитель, PTM, glycosylation
- Истинная аффинность; AF pLDDT и OASis ≠ Kd / immunogenicity assay
- Developability в CHO как экспериментальный readout
14. Evidence and limitations
[PAPER]Авторы прямо откладывают experimental screening. N-terminal truncations, human-heavy skew, rat generation failure, OAS sequencing error residuals. max_length=150 в коде может обрезать длинные CDR3. Non-commercial license.
15. Reproducibility
[CODE]Checkpoints trained_models/IgLM и IgLM-S. Dataset OAS95: data.graylab.jhu.edu/OAS95.tar.gz (не скачивался). Commit de819b56 — licensing text update (2026-07-20).
Физико-химический смысл
| Признак или операция модели | Физико-химическая интерпретация | Прямо / косвенно / внешне | Важное ограничение |
|---|---|---|---|
| AA tokens | первичная структура VH/VL | прямо | нет 3D / antigen |
| species / chain tags | репертуарный prior OAS | прямо как special tokens | OAS label ≠ ANARCI class (camel↔alpaca) |
| span permutation ILM | двусторонний sequence context для CDR | train-time rearrangement | не физический механизм созревания |
| OASis / aggregation predictors in paper | in silico developability / humanness | external evaluators | не часть generator; не assay |
Модульная схема
| Модуль | Класс модели | Вход | Выход | Обучается или frozen | Generator или evaluator |
|---|---|---|---|---|---|
| GPT2LMHeadModel | decoder-only transformer | tags + (masked) tokens | next-token logits | trained (released frozen) | generator |
| log_likelihood() | same GPT-2 | sequence ± infill_range | mean NLL (negative returned as likelihood) | frozen | evaluator |
| AlphaFold-Multimer / OASis / aggregation | external | generated sequences | pLDDT / humanness / in silico scores | n/a | external evaluation |
Generator versus evaluator
| Компонент | Роль | Integrated или external | Required или optional | Влияет на generation или только ранжирует после |
|---|---|---|---|---|
| GPT-2 infill/generate | primary_generator | integrated | required | generation |
| log_likelihood | integrated_scorer | integrated | optional | post_hoc |
| AF-Multimer pLDDT | structure predictor | external | paper only | post_hoc; не sample() |
| wet-lab affinity | experimental_validation | external | not in primary paper | absent |
Claim–evidence matrix
| Утверждение | Предполагаемый механизм | Тип evidence | Датасет / эксперимент | Comparator | Метрика / результат | Источник | Оговорка |
|---|---|---|---|---|---|---|---|
| Bidirectional infill лучше left-to-right на CDR PPL | ILM span-at-end | 2. retrospective benchmark | 30M held-out OAS infill PPL | IgLM [pre], ProGen2-base, ProGen2-OAS | infilling perplexity by CDR (Fig. 1D); CDR3 hardest | [PAPER] Fig. 1D | PPL ≠ function |
| Species/chain tags контролируют генерацию | conditioning tokens | 2. retrospective benchmark | 220k full-length samples | ANARCI classification vs tag | agreement high for most species; rat failure; camel heavy-only | [PAPER] Fig. 2C | ANARCI taxonomy ≠ OAS labels |
| Infilled CDR-H3 libraries improve in silico developability/humanness | natural OAS prior | 6. developability / manufacturability | 49 Thera-SAbDab; 9000 variants/parent | parent sequences; sampling T/P | in silico aggregation/solubility/OASis (figures) | [PAPER] Fig. 3, S10 | Класс evidence 6 in silico, не experimental developability и не binding |
| IgLM designed antibodies bind antigen | — | 1. только architectural rationale | n/a | n/a | не показано в primary paper | [PAPER] Discussion: libraries 'could then be experimentally screened' | Не заполнять как prospective binding |
Различия между статьёй и реализацией
| Версия статьи | Commit репозитория | Checkpoint / model | Файл | Различие | Практическое следствие |
|---|---|---|---|---|---|
| Cell Systems 2023 | de819b56238b5cb9c23d7a6fec4c55d7fc0395e4 | IgLM 12.9M / IgLM-S 1.4M | iglm/model/IgLM.py | Paper max context не равен коду max_length=150 в generate(). eos_token_id = tokenizer.cls_token_id. log_likelihood returns -nll (higher better). | Очень длинные CDR3 могут обрезаться; не путать знак likelihood. |
Схемы information flow
flowchart TD Tags["chain + species tokens"] --> ILM["GPT-2 decoder AR"] Seq["Unpaired VH or VL"] --> Mask["Replace span with MASK, append span after SEP"] Mask --> ILM ILM --> Out["Infilled or full-length sequence"] Out -.->|"paper only"| AF["AF-Multimer / OASis"]
Источники
- [1] Shuai R.W., Ruffolo J.A., Gray J.J. IgLM: Infilling language modeling for antibody sequence design. Cell Systems 14, 979–989 (2023). · DOI · источник
- [2] bioRxiv 10.1101/2021.12.13.472419. · DOI · источник
- [3] Graylab/IgLM commit de819b56. · источник
Нерешённые научные вопросы
- Table S2 exact 3-mer prompts
- tokens.py BAD_WORD_IDS полный список
- Independent experimental validation of IgLM libraries — NR this pass