About

• A repository for our paper titled "Substrate-Aware Zero-Shot Predictors for Non-Native Enzyme Activities".
• See Zenodo for data

Environments

• The main environment is substrate_aware.yml
• The coves.yml, esmif.yml, and plip.yml files are conda environment for the COVES, ESM-IF, and PLIP zero-shot calcualtion, respectively.
• Frozen versions of the dependencies can be found under envs/fronzen/.

Vina

• Install vina following the instructions in the documentation here
• Create a seperate conda environment for vina

conda create --name vina_env python=3.9.7 -y
conda activate vina_env
conda install -c conda-forge numpy openbabel pdbfixer scipy rdkit openmm biopython -y
pip install meeko
conda install -c conda-forge mdanalysis
pip install numpy==1.24.4

• Run the zero-shot prediction using the following command

~/miniconda3/envs/vina/bin/python -m tests.test_zs-vina

LigandMPNN

• Get code and model parameters

git clone https://github.com/dauparas/LigandMPNN.git
cd LigandMPNN
bash get_model_params.sh "./model_params"

• Modify the paths in substrate_aware.zs.ligandmpnn and/or tests.test_zs-ligandmpnn accordingly.

FlowSite

• Get code and moedel parameters

git clone https://github.com/HannesStark/FlowSite.git
cd FlowSite
mkdir model_params
cd model_params

• Download the model parameters in a zip file based on the repo from here.
• Unzip as needed and check that there are subfolders called lf5t55w4 and b1ribx1a
• Modify the path in substrate_aware.zs.ligandmpnn and/or tests.test_zs-ligandmpnn accordingly.
• Set up the conda environment

conda create -n flowsite python=3.10
conda activate flowsite
pip install torch==2.1.0+cu121 torchvision==0.16.0+cu121 torchaudio==2.1.0+cu121 -f https://download.pytorch.org/whl/torch_stable.html
pip install torch-scatter -f https://data.pyg.org/whl/torch-2.1.0+cu121.html
pip install torch-sparse -f https://data.pyg.org/whl/torch-2.1.0+cu121.html
pip install torch-cluster -f https://data.pyg.org/whl/torch-2.1.0+cu121.html
pip install torch-spline-conv -f https://data.pyg.org/whl/torch-2.1.0+cu121.html
pip install torch-geometric
pip install rdkit
pip install pyyaml wandb biopython spyrmsd einops biopandas plotly prody tqdm lightning imageio datasets
pip install e3nn

• Troubleshoot

pip uninstall numpy
pip install numpy==1.23.5

Datasets

The data/ folder is organized as follows:

• lib/ contains CSV files with sequence (AAs), fitness, and selectivity (when applicable) data.
  • Each file is named using the format {enzyme}_{substrate}.csv.
  • Note: ParLQ-a is sometime abbreviated as ParLQ
• seq/ contains FASTA files for each enzyme sequence.
• structure/ contains the structural files in PDB format.
  • ParLQ.pdb is taken from Yang and Lal et al. (2025).
  • PfTrpB.pdb corresponds to chain A of PDB entry 5DW0.
  • Rma.pdb corresponds to PDB entry 3CP5.
  • apo/ contains apo (substrate-free) structures in PDB format.
  • clean/ contains cleaned structures (solvent removed) in PDB format.
  • docked/ contains docked structures in CIF format, generated using AF3 or Chai.
    • See docked/README.md for additional information on the docked structures.
• evmodel/ contains EVcouplings model files (.model) for each enzyme, which are used to calculate EVmutation scores.
• evmodel_dets/ contains detailed metadata for the corresponding EVcouplings models.

Preprocessing

• Run preprocess_all from substrate_aware.preprocess to preprocess the data.

Zero-shot prediction

• Each zero-shot predictor corresponds to a module in substrate_aware.zs.
• run_all_combzs in substrate_aware.zs can be used to combine all generated zero-shot scores for each dataset.

Analysis

• The analysis scripts are located in substrate_aware.analysis.

Contact

• Francesca-Zhoufan Li