bii-dpi / dlscore-cnn

Predicting protein-ligand binding affinities using Convolutional Neural Networks (CNN)

This is an implementation of the CNN network architecture described in the following paper

KDEEP: Protein–Ligand Absolute Binding Affinity Prediction via 3D-Convolutional Neural Networks
José Jiménez , Miha Škalič , Gerard Martínez-Rosell , and Gianni De Fabritiis
DOI: 10.1021/acs.jcim.7b00650

Requirements

• Tensorflow : pip install tensorflow-gpu or pip install tensorflow
• Keras: pip install keras
• Scikit-learn: pip install -U scikit-learn
• oddt: conda install -c oddt oddt
• tqdm: pip install tqdm
• htmd:

conda config --add channels acellera 
conda config --add channels psi4 
conda install htmd 

Training the model (by generating augmented data on the fly)

• First extract the voxel features using the script feature_extraction_htmd.py insdie the directory training_with_htmd. The script will create the file called data.h5 inside the dataset dir
• Run the script train.py to train the model. Modify the code in the method main() to enable/disable data augmentation and tweak other training parameters.

Training the model (using saved augmented data)

• Make sure npy_data directory is inside the dataset directory.
• Run train2.py to train the model.