Accurate prediction of the structure of any protein mutant with a single-site mutation with equivariant graph neural networks. PreMut takes as input a wild-type protein structure and a single-site mutation to predict the structure of the mutated protein with the mutation.
Download or clone this repository
Setup environment by running the following command
conda env create -f environment.yml
Some packages may fail to install. In that case, use the following commands to install these packages necessary for full environment installation.
Firstly, activate the created environment
conda activate PreMut
pip install the following packages
pip install pytorch-lightning==1.8
pip install biopandas
pip install rmsd
python src/prediction.py wild_pdb_path mutation_info chain_id output_dir
python src/prediction.py examples/8b0s.pdb C_144_A A predictions
python src/train.py
* After the completion of training, the model weights are stored in a folder titled Checkpoints.
* You can select the model with the best validation performance from the Checkpoints folder to test.
## Evaluation
* Evaluation script is provided to check the reported performance in the paper.
* Make sure the files are downloaded, uncompressed and moved to the root of the directory as instructed in the previous section.
* Install SPECS from following the instructions from this link [SPECS](http://watson.cse.eng.auburn.edu/SPECS/).
* To get the evaluation metrics on MutData2022_test dataset, run this command
python Evaluation.py MutData2022
* To get the evaluation metrics on MutData2023_test dataset, run this command
python Evaluation.py MutData2023
## Acknowlegements
The EGNN model code is partially adapted and built upon the source code from the following project [egnn](https://github.com/vgsatorras/egnn). We thank all the contributors and maintainers.
## Reference
Sajid Mahmud, Alex Morehead, Jianlin Cheng. Accurate prediction of protein tertiary structural changes induced by single-site mutations with equivariant graph neural networks. bioRxiv. (https://doi.org/10.1101/2023.10.03.560758)