PreMut

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.

Installation

• 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

Make prediction

• To make a prediction, save the wild pdb file in a folder and create a folder to save the prediction.
• Run the following command

  python src/prediction.py wild_pdb_path mutation_info chain_id output_dir

• Here wild_pdb_path is the path to the structure file of the wild protein in the pdb format (e.g. /path/to/8b0s.pdb), mutation_info is the information regarding the wild residue, mutated position (0 indexed) and the mutated residue (e.g. C_144_A), chain_id is the id of the chain in the wild pdb file (e.g. A), and output_dir is the directory to save the predicted structure for the mutant as a pdb file.
• Example

  python src/prediction.py examples/8b0s.pdb C_144_A A predictions

Refine predictions

• ATOMRefine can be utilized to further refine the predictions made by PreMut. Install ATOMRefine from this link (https://github.com/BioinfoMachineLearning/ATOMRefine) and follow the instructions provided in that repository to apply refinement to the predictions made by PreMut.

Training

• Download and install TM-score and TM-align from this link: TM-score, TM-align
• Download all the files in the MutData2022 and MutData2023 datasets from the link PreMut. Uncompress the files, and then move the uncompressed folders to the installation directory of the PreMut repository.
• Run the following command to train the model

  
  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)