Title: Using Deep Learning to Predict Toxicity and Lipophilicity from Molecular Fingerprints and 2D Structures.
Please download the sdf datafiles from the following links and add them to /data
folder:
Python version supported: Python 3
.
Setup Steps:
- Install miniconda (please see instructions in https://docs.conda.io/en/latest/miniconda.html )
- Create a new environment:
conda create -n myenv python
- Activate conda environment:
source activate myenv
- Install RDKit (necessary for sdf files preprocessing):
conda install -c rdkit rdkit
- Install requirements:
pip install -r requirements.txt
The best performing models are the following:
-
Lipophilicity predictors:
- fingerprints input
fc_nn_6l_logp.py (LogP)
fc_nn_6l_exp_logp.py (experimental LogP) - 2D molecular images input
incep_resnet_compact_v4_logp.py (LogP)
incep_resnet_compact_v4_exp_logp.py (experimental LogP)
- fingerprints input
-
Toxicity predictors:
-
fingerprints input
fc_nn_tox21.py -
2D molecular images input
incep_resnet_tox21_t.py
-
General instructions set Train=True
in main function for training and Train=False
for running in test mode specifying a previously saved weight file.
N.B: We include some of the best performing models weights files in weights/
folder.
You can see come of the previous run results (metrics images and history files) in the output/
folder.