BoltzDesign1 is a molecular design tool powered by the Boltz model for designing protein-protein interactions and biomolecular complexes.
📄 Paper: BoltzDesign1: AI-Powered Molecular Design
🚀 Colab: https://colab.research.google.com/github/yehlincho/BoltzDesign1/blob/main/Boltzdesign1.ipynb
-
Clone the repository
git clone https://github.com/yehlincho/BoltzDesign1.git cd BoltzDesign1 -
Run the automated setup
chmod +x setup.sh ./setup.sh
⚠️ Note: AlphaFold3 setup not included. Install separately following official instructions
The setup script will automatically:
- ✅ Create conda environment with Python 3.10
- ✅ Install all required dependencies
- ✅ Set up Jupyter kernel for notebooks
- ✅ Download Boltz model weights
- ✅ Configure LigandMPNN and ProteinMPNN
- ✅ Optionally install PyRosetta
- ❌ Need to install AF3 separately
Run the complete pipeline from BoltzDesign to LigandMPNN/ProteinMPNN redesign and AlphaFold3 cross-validation.
Examle for small molecule: python boltzdesign.py --target_name 7v11 --target_type small_molecule --target_mols OQO --gpu_id 0 --design_samples 2 --suffix 1
Example for DNA/RNA PDB design: python boltzdesign.py --target_name 5zmc --target_type dna --pdb_target_ids C,D --gpu_id 0 --design_samples 5 --suffix 1
--alphafold_dir: Path to your AlphaFold3 installation (default: ~/alphafold3)--af3_docker_name: Name of your AlphaFold3 Docker container--af3_database_settings: Path to AlphaFold3 database--af3_conda_env: Name of your AlphaFold3 conda environment
If you want to disable af3 cross validation add flag --run_alphafold False
- If binder does not form a highly compact structure, increase num_intra_contacts e.g. (default) 2 -> 4
- If target does not form interaction with binder, increase num_inter_contacts e.g. (default) 2 -> 4
- If generated binders have all alpha helices and you want to design beta sheets, change e.g. helix_loss_max 0.0, helix_loss_min = -0.3 to helix_loss_max -0.3, helix_loss_min = -0.6
We installed trajectory visualization based on LogMD (https://github.com/log-md/logmd, implemented for Boltz diffusion trajectory https://colab.research.google.com/drive/1-9GXUPna4T0VFlDz9I64gzRQz259_G8f?usp=sharing#scrollTo=4eXNO1JJHYrB)
If you want to enable visualization of the trajectory, you need to set --save_trajectory True. However, be cautious that if you are just optimizing with distogram (--distogram_only True), it will take more time since it also runs the diffusion modules to get actual xyz coordinates.
Configure your molecular design parameters:
config = {
# Optimization parameters
'mutation_rate': 1,
'learning_rate_pre': 0.2, ## Pre_iteration stage
'learning_rate': 0.1, ## Soft, temp, hard stages
# Iteration stages
'pre_iteration': 30, # Initial logits optimization
'soft_iteration': 75, # Logits to Softmax optimization
'temp_iteration': 45, # Softmax Temperature annealing
'hard_iteration': 5, # Final hard encoding optimization
'semi_greedy_steps': 0, # MCMC based on iPTM score
# Algorithm settings
'design_algorithm': '3stages',
}BoltzDesign1 supports sequence optimization using:
- Use case: Protein-protein interface design
- Use case: Protein-ligand and non-protein biomolecule interfaces
- Interface residues (< 4 Å) are fixed during design
- Non-interface residues are redesigned
- Custom interface definitions can be specified
Final structures are validated using AlphaFold3 for:
- Structure quality assessment
- Confidence scoring
- Cross-validation against design targets
- Chai-1: All-atom structure prediction
- AlphaFold: Protein monomer and multimer structure prediction
After running the pipeline in boltzdesign.py, high-confidence designs can be found in:
your_output_folder/ligandmpnn_cutoff_(interface threshold)/03_af_pdb_success
The designs are saved along with high_iptm_confidence_scores.csv, which contains the iPTM and pLDDT scores for each design.
- AlphaFold3 integration for validation pipeline
- Boltz1x Integration
- Multi Chains Design - Currently supporting single chain design
- Multi-state optimization - Alternating conformations
- Specificity enhancement - Target selectivity
- RNA MSA Generation - Multiple sequence alignments
- Get Colab version of MSA extraction from ColabNuFold (https://colab.research.google.com/github/kiharalab/nufold/blob/master/ColabNuFold.ipynb#scrollTo=KDs4o5Bv35MI)
- Input Support for DNA and RNA Modifications
- Advanced Filtering:
- Docking score integration
- Metal coordination prediction
- DNA/RNA specificity scoring
- Enhanced Scoring: Currently uses Rosetta scores (from [BindCraft])
License: MIT License - See LICENSE file for details Citation: If you use BoltzDesign1 in your research, please cite:
@article{cho2025boltzdesign1,
title={Boltzdesign1: Inverting all-atom structure prediction model for generalized biomolecular binder design},
author={Cho, Yehlin and Pacesa, Martin and Zhang, Zhidian and Correia, Bruno E and Ovchinnikov, Sergey},
journal={bioRxiv},
pages={2025--04},
year={2025},
publisher={Cold Spring Harbor Laboratory}
}
Questions or Collaboration: [email protected]
Issues: Please report bugs and feature requests via GitHub Issues
EXPERIMENTAL SOFTWARE: This pipeline is under active development and has NOT been experimentally validated in laboratory settings. We release this code to enable community contributions and collaborative development. Use at your own discretion and validate results independently.