search

ASKCOS / askcos-core

Python package for the ASKCOS platform for prediction of chemical reactivity
Other
15 stars 13 forks source link

tensorflow gpu #5

Open ydzdfci opened 2 years ago

ydzdfci commented 2 years ago

I'm running individual modules of template-free forward predictor.

  1. Is there an option or argument that I can run the script using gpu? I will run about millions building blocks and it will cost long time. Do you have any suggestions on large amount batch process? Thanks
  2. I got different results between individual modules and online predictor For example, react1: NNC(=O)CCc1ccc(S(=O)(=O)N2CCOCC2)cc1, react2: O=C(CSC1=CCS(=O)(=O)C1)Nc1ccccc1 From askcos.mit.edu, I input those two reactants in the fields and leave solvent blank. The result shows O=C(CSC1CCS(=O)(=O)C1)Nc1ccccc1 with prob of 0.8978. However, from python askcos/synthetic/evaluation/template_free.py 'NNC(=O)CCc1ccc(S(=O)(=O)N2CCOCC2)cc1.O=C(CSC1=CCS(=O)(=O)C1)Nc1ccccc1', I got NNC(=O)CCc1ccc(S(=O)(=O)N2CCOCC2)cc1 with prob of 0.9059. This predicted product is actually one of the reactant. Thanks