Target-conditioned de novo antibody (RFantibody)
Практический сценарий
Есть target structure и epitope hotspots; нужен de novo VHH или scFv на фиксированном framework.
Вход
- target structure
- epitope hotspots
- fixed antibody framework
- CDR lengths
Pipeline diagram
CDR/backbone generation
→CDR sequence design
→Complex prediction filter
→External validation
→Experimental binding
Mermaid diagram
flowchart TD
A[Target structure + epitope] --> B[3D diffusion generator]
B --> C[CDR/backbone proposals]
C --> D[Inverse-folding model]
D --> E[Sequence-structure candidates]
E --> F[Integrated complex predictor]
F --> G[Ranked candidates]
G --> H[External validation]
H --> I[Experimental assays]Стадии pipeline
| Стадия | Архитектурный класс | Что происходит | Генерация или оценка | Generator | Промежуточный объект | Альтернативы |
|---|---|---|---|---|---|---|
| CDR/backbone generation | SE(3)-equivariant diffusion | Генерируются CDR/backbone coordinates | generation | rfantibody | CDR backbone proposals | NR |
| CDR sequence design | Graph-based inverse folding | ProteinMPNN подбирает CDR sequences | generation | proteinmpnn | Sequence–structure candidates | AntiFold |
| Complex prediction filter | Fine-tuned RoseTTAFold2 | Self-consistency/confidence фильтрация | evaluation | NR | Ranked antibody–target complexes | AlphaFold3 |
| External validation | Docking/relaxation optional | Независимая оценка plausibility | external_evaluation | NR | Shortlist | PyRosetta / Rosetta |
| Experimental binding | NR | SPR/display/neutralization | experiment | NR | Confirmed binders | NR |
Hardware и runtime
GPU required. Runtime зависит от trajectories, MPNN samples и RF2. Exact hours NR в матрице.
Evidence level
3–4 prospective binding/function. Target success 0–2% reported. Initial affinities can be modest.
Точки отказа
- Wrong holo conformation/hotspots.
- Framework mismatch.
- Treating RF2/AF3 as affinity.
- Expecting miniprotein-like hit rates.
Критерии решения
- VHH vs scFv format.
- Готовы ли к низкому hit rate и последующей maturation?
- Нужна ли humanization отдельно?
Связанные задачи
De novo antibody design, De novo VHH / nanobody design, Antigen-conditioned antibody design, Генерация CDR backboneСвязанные инструменты
RFantibody, ProteinMPNNАрхитектуры
Frame diffusion, Inverse folding, SE(3)-equivariant networks, Invariant modelsPapers
Atomically accurate de novo design of antibodies with RFantibodySource-grounded notes
Разделение стадий
[PAPER]Стадии должны быть раскрашены как generation (RFdiffusion, MPNN), evaluation (RF2/AF3, Rosetta), experiment (display/SPR). Не называть RF2 генератором. [PAPER]