Structure-based de novo antibody and nanobody design on Google Colab - no Docker required.
A Google Colab notebook for running RFantibody without Docker. Design antibodies or nanobodies against any protein target.
| Step | Model | Purpose |
|---|---|---|
| 1 | RFdiffusion (Ab) | Generate antibody backbone |
| 2 | ProteinMPNN | Design amino acid sequence |
| 3 | RF2 (Ab) | Validate structure prediction |
- Antibody mode: Design with Heavy + Light chains
- Nanobody mode: Design single-domain antibodies (VHH)
- GPU accelerated: Runs on Colab's free T4 GPU
- No installation: Everything runs in the cloud
- Google account (for Colab)
- Target PDB file (your antigen)
- Hotspot residues (epitope on target)
- Click the "Open in Colab" badge above
- Enable GPU: Runtime → Change runtime type → T4 GPU
- Run setup cells (clone, weights, dependencies)
- Upload your target PDB
- Set parameters (hotspots, loop lengths, number of designs)
- Run the pipeline
- Download results
- Your antigen structure (from RCSB PDB or AlphaFold)
- Chain will be automatically renamed to T
- Epitope residues where you want the antibody to bind
- Format:
[T195,T197,T256] - Find these using ChimeraX:
select zone /antibody 4 /target
- Antibody:
[L1:8-13,L2:7,L3:9-11,H1:7,H2:6,H3:5-13] - Nanobody:
[H1:7,H2:6,H3:5-13]
- Validated PDB files with designed antibody/nanobody
- Each design includes:
- Backbone structure
- Amino acid sequence
- Target complex
Designs pass RF2 validation if:
- pAE < 10 (predicted aligned error)
- RMSD < 2 Å (structure deviation)
On T4 GPU:
- Setup: ~5 min
- Per design: ~1-2 min
- Total (10 designs): ~20-25 min
Based on RFantibody by:
- RosettaCommons / Baker Lab
- Institute for Protein Design, University of Washington
If you use this notebook, please cite the original RFantibody paper:
[RFantibody citation - check original repo]
This Colab adaptation follows the license of the original RFantibody repository.