DeepChemStable: chemical stability prediction using attention-based graph convolution network
Python 3.6.3
Tensorflow 1.1.0
RDkit 2018.03.4
Autograd 1.2
Numpy 1.14.2
Pandas 0.23.4
The trained model weights are stored in fingerprint_variables.bin, prediction_variables.bin.
Use the predict.py to predict.
The predictions are saved in the file results.csv.
The visualization of predictive unstable compounds with hightlighted unstable fragment are saved in the folder figures/.
>python predict.py *yourfilepath* *amount*
Examples: >python predict.py test.csv 5
Data file format:
Datafile should be CSV file;
The header must be "substance_id, smiles, label";
In "label" column, use "0" for all compounds so the visualization can implement.