A cross-attention-empowered SE(3)-equivariant graph neural network architecture for predicting enzyme substrate specificity

This package is tested with Python 3.10.12 and CUDA 12.1 on Ubuntu 22.04. Please note that the versions of PyTorch and PyTorch-Geometric depend on the CUDA version installed on your local server.

1. Install python packages by using Anaconda / Miniconda conda env create -f environment.yml
2. Install pytorch-geometric and related package pip install torch_geometric pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.2.0+cu121.html

The following script trains an EZSpecificity model using Halogenase data. python main_specificity_ss.py --config_path $LOCAL_ROOT_DIR/src/Configs/Train/random_0.yml

The following script generates predictions by an EZSpecificity model trained on Halogenase data. python main_specificity_ss_eval.py --log_dir $LOCAL_ROOT_DIR/saved_model/finetune_halogenase_random_0/run_0

1. Please follow the instruction in the paper to generate predicted 3d structure of enzymen-substrate complex
2. Follow example.ipynb to complete two tasks: feature preparation and prediction generation.