Skip to main content

Table 4 Best models from the ChEMBL benchmark for both SMILES variants

From: Randomized SMILES strings improve the quality of molecular generative models

SMILES Time % Valid % Unique FCD
Canonical 131:32 98.26 34.67 0.0712
Rest. Random. 84:22 98.33 64.09 0.1265
  1. SMILES SMILES variant, Time time used to train the model hhh:mm, % Valid Percent of valid molecules, % Unique Percent of unique molecules in a 2 billion SMILES sample, Fréchet ChemNet Distance (FCD) between the validation and a sample of 75,000 molecules (FCD)