This is a repository of our paper, "FragOPT: An ML-Driven Computational Workflow for Rational Fragments Optimization Toward Lead Compounds". (Link).
Run:
conda create -n FragOPT python=3.8.20
conda install numpy=1.24 pandas=2.0.3 scikit-learn=1.3 seaborn=0.13
conda install -c conda-forge openbabel charset-normalizer
pip install torch==2.3.1+cpu -f https://download.pytorch.org/whl/torch_stable.html
pip install tensorflow-gpu==2.10.0 xgboost-cpu==2.1.2 rdkit shap matplotlib prody
This will take care of installing all required dependencies.
The only required dependency is the latest Conda package manager, which you can download with the Anaconda Python distribution here.
You should prepare the data to train QSAR models for your protein.
The first column should be molecular SMILES, the second colomn should be the bioactivity class. you can refer to the files in dataset folder.
The QSAR model train with molecular fingerprint and bioacivity value of your protein. To do this, run:
python train_models.py
where the dataset_path in train_models.py should be change to your data path.
After trianed the QSAR models. You can use FragOPT to optimize the target molecule for its bioacivity and druglikeness.
python main.py -j config.json
The configuration of optmization could refer to the config.json.
The model of PD-L1(PDB ID: 5N2F) and COVID-19 (PDB ID: 7BQY) were pretrained in models folder, you can directly use them.