Table 1 Model architectures considered for latent space evaluation
Name | Variational autoencoder | Architecture | Molecule representation | Enumerated |
|---|---|---|---|---|
SMILES-AE-can2can | No | GRU, 128 hidden and latent size, 3 layers, attention | SMILES | No |
SMILES-AE-enum2can | No | GRU, 128 hidden and latent size, 3 layers, attention | SMILES | Yes (input) |
SMILES-AE-can2enum | No | GRU, 128 hidden and latent size, 3 layers, attention | SMILES | Yes (output) |
SMILES-VAE-can2can | Yes | GRU, 128 hidden and latent size, 3 layers, attention | SMILES | No |
SELFIES-can2can | No | GRU, 128 hidden and latent size, 2 layers, attention | SELFIES | No |
SELFIES-enum2can | No | GRU, 128 hidden and latent size, 2 layers, attention | SELFIES | Yes (input) |
SELFIES-can2enum | No | GRU, 128 hidden and latent size, 2 layers, attention | SELFIES | Yes (output) |
SELFIES-VAE | Yes | GRU, 128 hidden and latent size, 2 layers, attention | SELFIES | No |