DILI is a comprehensive repository aimed at enhancing the early detection of Drug-Induced Liver Injury (DILI) through the integration of predicted in vivo and in vitro data. This project utilizes advanced machine learning models and chemical informatics to predict the likelihood of DILI for various compounds.
- Integration of in vivo and in vitro data for improved DILI prediction.
- Utilizes pharmacokinetic parameters, structural fingerprints, and physicochemical parameters.
- Nine proxy-DILI labels from datasets such as mitochondrial toxicity, bile salt export pump inhibition, and preclinical rat hepatotoxicity studies and two pharmacokinetic parameters.
- Available as online tools, local implementation and a Python library.
You can install the DILI Predictor using pip: pip install dilipred. Please use Python <3.12, >=3.9
You can also build from source using python-poetry:
- Clone the repository:
git clone https://github.com/Manas02/dili-pip.git - Navigate to the project directory:
cd dili-pip/ - Install dependencies:
poetry install - Activate the virtual environment:
poetry shell - Build the project:
poetry build
To get started with the CLI, use: dili -h
Select from the sidebar to predict DILI for a single molecule.
Here's a basic example of how to use DILIPredictor as a Python library:
from dilipred import DILIPRedictor
if __name__ == '__main__':
dp = DILIPRedictor()
smiles = "CCCCCCCO"
result = dp.predict(smiles)
print(result)Download key files from https://github.com/srijitseal/DILI/raw/main/local_implementation.zip and run locally!
If you prefer to use the predictor online via Uppsala University SciLifeLab Serve: https://dili.serve.scilifelab.se/
If you prefer to use the predictor online via streamlit: https://dilipredictor.streamlit.app/
data/: Contains datasets used in the project.00_Create_livtox_dili_datasets.ipynb: Notebook to curate DILI datasets.01_Split_Train_Test_Datasets.ipynb: Notebook to split datasets into training and testing sets.02_Explore_Chemical_Spaces_PCA_TSNE-.ipynb: Notebook to explore chemical spaces using PCA and t-SNE.03a_Create_Cmax_models.ipynb: Notebook to create Cmax models.03b_predCmax_DILI_Scores.ipynb: Notebook to predict Cmax DILI scores.03c_Create_combination_models_v2.ipynb: Notebook to create combination models.04_Compare_Performance_NCV.ipynb: Notebook to compare model performance using nested cross-validation.05a_Find_AUC_Regions_v2.ipynb: Notebook to find AUC regions.05c_Compare_FPR.ipynb: Notebook to compare false positive rates.08_check_chemistry_12trainingcompounds.ipynb: Notebook to check chemistry of 12 training compounds.08_check_chemistry_4compounds.ipynb: Notebook to check chemistry of 4 compounds.Cmax_processed.csv: Processed Cmax data.Heldout_test_predictions_earlydetection.csv: Predictions for held-out test data for early detection.Test_data_DILIst.csv: Test data for DILI.all_features_desc.csv: Description of all features.mordred_desc.csv: Description of Mordred features.test_data_heldouttest_DILIst_223.csv: Held-out test data for DILI.test_data_ncv_DILIst_888.csv: Nested cross-validation test data for DILI.
If you use DILI Predictor in your work, please cite:
Improved Early Detection of Drug-Induced Liver Injury by Integrating Predicted in vivo and in vitro Data; Srijit Seal, Dominic P. Williams, Layla Hosseini-Gerami, Manas Mahale, Anne E. Carpenter, Ola Spjuth, Andreas Bender bioRxiv 2024.01.10.575128; doi: https://doi.org/10.1101/2024.01.10.575128
This project is licensed under the MIT License. See the LICENSE file for details.
Developed and maintained by Srijit Seal and contributors.
For any questions or issues, please open an issue on the GitHub repository.