QSARBoost is a Python framework for developing gradient boosting–powered QSAR models for molecular bioactivity prediction and virtual screening. This repository applies QSARBoost to the repurposing of DrugBank compounds as SIRT2 inhibitors, but it can be adapted for any regression-based QSAR task.
Python 3.x
RDKit – Molecular descriptor/fingerprint generation
Scikit-learn – Machine learning utilities
CatBoost / LightGBM / XGBoost – Gradient boosting algorithms
Optuna – Hyperparameter optimization
Pandas – Data manipulation
Matplotlib – Visualization
Data Curation – Import, clean, and preprocess chemical datasets from ChEMBL or other sources.
Molecular Featurization – Generate ECFP4 fingerprints or other descriptors using RDKit.
Gradient Boosting Models – Support for multiple boosting algorithms: CatBoost, LightGBM, XGBoost, GradientBoostingRegressor.
Model Selection & Optimization – Hyperparameter tuning via Optuna.
Applicability Domain (AD) – Williams plot & Mahalanobis distance methods for prediction confidence.
Virtual Screening – Apply trained models to large chemical libraries (e.g., DrugBank).
Install required packages:
pip install rdkit-pypi scikit-learn pandas matplotlib catboost lightgbm xgboost optuna
git clone https://github.com/yboulaamane/QSARBoost.git
This project is released under the MIT License.
Repurposing DrugBank Compounds as SIRT2 Inhibitors via QSAR Modeling with Gradient Boosting Algorithms and All-Atom Molecular Simulations (manuscript in preparation).
For questions or feedback, please contact: boulaamane.yassir@etu.uae.ac.ma