metadata

title: PROTAC-Degradation-Predictor
emoji: 🧬
colorFrom: pink
colorTo: green
sdk: gradio
sdk_version: 4.37.2
app_file: app.py
pinned: false
license: mit

PROTAC-Degradation-Predictor

A machine learning-based tool for predicting PROTAC protein degradation activity.

📚 Table of Contents

Data Curation
Installation
Documentation and Usage
Training
Citation
License

📝 Data Curation

The code for data curation can be found in the Jupyter notebook data_curation.ipynb.

The folder data/studies contains the training and test data used in each study reported in our paper. The label column that is used for predictions is named "Active (Dmax 0.6, pDC50 6.0)" and contains binary values.

🚀 Installation

To install the package, open your terminal and run the following commands:

pip install git+https://github.com/ribesstefano/PROTAC-Degradation-Predictor.git

The package has been developed on a Linux machine with Python 3.10.8. It is recommended to use a virtual environment to avoid conflicts with other packages.

🎯 Documentation and Usage

The package documentation can be found here. For a walkthrough on how to use the package, please refer to the tutorial notebook protac_degradation_predictor_tutorial.ipynb.

After installing the package, you can use it as follows:

import protac_degradation_predictor as pdp

protac_smiles = 'Cc1ncsc1-c1ccc(CNC(=O)[C@@H]2C[C@@H](O)CN2C(=O)[C@@H](NC(=O)COCCCCCCCCCOCC(=O)Nc2ccc(C(=O)Nc3ccc(F)cc3N)cc2)C(C)(C)C)cc1'
e3_ligase = 'VHL'
target_uniprot = 'P04637'
cell_line = 'HeLa'

active_protac = pdp.is_protac_active(
    protac_smiles,
    e3_ligase,
    target_uniprot,
    cell_line,
)

print(f'The given PROTAC is: {"active" if active_protac else "inactive"}')

This example demonstrates how to predict the activity of a PROTAC molecule. The is_protac_active function takes the SMILES string of the PROTAC, the E3 ligase, the UniProt ID of the target protein, and the cell line as inputs. It returns whether the PROTAC is active or not.

The function supports batch computation by passing lists of SMILES strings, E3 ligases, UniProt IDs, and cell lines. In this case, it returns a list of booleans indicating the activity of each PROTAC.

📈 Training

Before running the experiments reported in our work or train on your custom dataset, here are some required steps to follow (assuming one is in the repository directory already):

Download the data from the Cellosaurus database and save it in the data directory:

wget https://ftp.expasy.org/databases/cellosaurus/cellosaurus.txt data/

Make a copy of the Uniprot embeddings to be placed in the data directory:

cp protac_degradation_predictor/data/uniprot2embedding.h5 data/

Create a virtual environment and install the required packages by running the following commands:

conda env create -f environment.yaml
conda activate protac-degradation-predictor

The code for training the PyTorch models can be found in the file run_experiments_pytorch.py.

(Don't forget to adjust the PYTHONPATH environment variable to include the repository directory: export PYTHONPATH=$PYTHONPATH:/path/to/PROTAC-Degradation-Predictor)

Training on Custom Dataset

For training a model on a user-provided dataset, please refer to the guide reported in this README.

📄 Citation

If you use this tool in your research, please cite the following paper:

@article{Ribes_2024,
   title={Modeling PROTAC degradation activity with machine learning},
   volume={6},
   ISSN={2667-3185},
   url={http://dx.doi.org/10.1016/j.ailsci.2024.100104},
   DOI={10.1016/j.ailsci.2024.100104},
   journal={Artificial Intelligence in the Life Sciences},
   publisher={Elsevier BV},
   author={Ribes, Stefano and Nittinger, Eva and Tyrchan, Christian and Mercado, Rocío},
   year={2024},
   month=dec, pages={100104}
}

The directories logs and reports contain the logs and reports generated during the experiments reported in the paper. Additionally, in reports, one can find the pickled Optuna studies for the reported experiments.

The directory models contains the trained models for the experiments reported in the paper.

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.