Reaction-conditioned De Novo Enzyme Design with GENzyme

GENzyme enables de novo design of catalytic pockets, enzymes, and enzyme-substrate complexes for any reaction. Simply change args.substrate_smiles and args.product_smiles in gen_configs.py to customized substrate SMILES and product SMILES, then run python generate.py, you can design your own enzymes.

GENzyme Paper at arxiv.

Requirement

python>=3.11
CUDA=12.1
torch==2.4.1 (>=2.0.0)
torch_geometric==2.4.0
torch_scatter==2.1.2

pip install mdtraj==1.10.0 (do first will install numpy, scipy as well, install later might raise dependency issues)
pip install esm==3.0.7.post1
pip install pytorch-warmup==0.1.1
pip install POT==0.9.4
pip install rdkit==2023.9.5
pip install biopython==1.84
pip install tmtools==0.2.0
pip install geomstats==2.7.0
pip install dm-tree==0.1.8
pip install ml_collections==0.1.1
pip install torchmetrics==0.11.4
pip install OpenMM
pip install einx
pip install einops

conda install conda-forge::pdbfixer

In case if you want to use the pocket-specific binding module, which is not necessarily installed for enzyme design:

For binding module, we use UniMol Docking v2, you need to install [UniCore](https://github.com/dptech-corp/Uni-Core)

Model Weights

You should download GENzyme checkpoint at Google drive. Once you download it, put it under genzyme_ckpt folder, namely genzyme_ckpt/genzyme.ckpt.

Notes before customized design

Please make sure you have ESM3 installed and have access to ESM3.
To customize catalytic reaction, remeber to change the subsrtate SMILES and product SMILES in gen_configs.py.
You may also change args.ptm_filter and args.plddt_filter in gen_configs.py for filtering enzymes.
GENzyme inference script generate.py is provided for your own design.

Inference script configs

args.pdb_name             #Enzyme PDB file for refinement/repurposing, set None if no PDB file available
args.substrate_smiles     #Input substrate SMILES
args.product_smiles       #Input product SMILES
args.n_pocket_res         #Number of catalytic pocket residues for design
args.n_protein_res        #Number of enzyme residues for design
args.num_pocket_design_t  #Number of inference steps (ODE steps for sampling)
args.n_sample_enzyme      #Number of enzymes
args.num_inpaint_t        #Number of pocket inpainting steps
args.ptm_filter           #pTM filtering
args.plddt_filter         #pLDDT filtering

Enzyme Refinement or Enzyme Repurposing

gen_configs.py contain all inference configurations and hyperparameters.
Put your pocket pdb file under data/ground_truth/pocket/ folder, put protein pdb file under data/ground_truth/protein/ folder.
In gen_configs.py, change args.pdb_name to your pdb file name. Also change args.substrate_smiles to one substrate SMILES, and args.product_smiles to one product SMILES, to customize reaction.
Run python generate.py for enzyme refinement and repurposing.
Output pockets and enzymes are saved under generated/ folder.

De novo Enzyme Design

gen_configs.py contain all inference configurations and hyperparameters.
In gen_configs.py, change args.pdb_name to one pdb file (set to None for de novo design args.pdb_name = None). Also change args.substrate_smiles to one substrate SMILES, and args.product_smiles to one product SMILES, to customize reaction.
Run python generate.py for de novo enzyme design.
Output pockets and enzymes are saved under generated/ folder.

Reproduce Enzyme Design

GENzyme reproduce script reproduce.py is provided.
Run python reproduce.py for reproduction.

Model Training

configs.py contain all training configurations and hyperparameters.
Train model using train.py for single GPU training. Run python train.py for training.

License

No Commercial use of either the model nor generated data, details to be found in LICENSE.

Citation

@misc{hua2024reactionconditionednovoenzymedesign,
      title={Reaction-conditioned De Novo Enzyme Design with GENzyme}, 
      author={Chenqing Hua and Jiarui Lu and Yong Liu and Odin Zhang and Jian Tang and Rex Ying and Wengong Jin and Guy Wolf and Doina Precup and Shuangjia Zheng},
      year={2024},
      eprint={2411.16694},
      archivePrefix={arXiv},
      primaryClass={q-bio.BM},
      url={https://arxiv.org/abs/2411.16694}, 
}

LiorZ/GENzyme