This repository contains the 33 data sets, code and scripts needed to generate the compound activity prediction models reported in:

KekuleScope: prediction of cancer cell line sensitivity and compound potency using convolutional neural networks trained on compound images

Isidro Cortés-Ciriano and Andreas Bender

Journal of Cheminformatics, 2019, 11:41

KekuleScope depends on the following python libraries:

scipy==1.0.0
numpy==1.15.1
joblib==0.11
ipython==7.1.1
Pillow==5.3.0
profilehooks==1.10.0
rdkit==2009.Q1-1
scikit_learn==0.20.0
torch==0.4.1.post2
torchvision==0.2.1

To install these libraries run:

pip install -r requirements.txt

To obtain help on how to run the models run:

python Kekulescope.py --help

It is assumed that the user has access to at least 1 GPU card.

Please contact Isidro Cortés Ciriano, PhD (isidrolauscher at gmail.com) for further information or suggestions.