Validation report: extend_v1

Model: extend_v1

Samples Generated: 500

Generated On: 2026-01-05 00:28

Generation Quality Metrics

Uniqueness	Novelty	SELFIES fidelity
0.9960	0.6426	0.0152

Descriptor Statistics

Descriptor	Average	Minimum	Maximum
MW	246.2045	97.1600	496.0100
LogP	1.9321	-2.2100	7.6400
SA_Score	3.5984	1.5400	6.9800
QED	0.6220	0.1950	0.9430
Fsp3	0.5701	0.0000	1.0000
RotatableBonds	3.0820	0.0000	12.0000
RingCount	2.5800	0.0000	9.0000
TPSA	47.8160	0.0000	138.8300
RadicalElectrons	0.0120	0.0000	2.0000

Descriptor Distributions

Development Notes

Model was trained on SMILES patterns encoded into SELFIES. Training data was obtained from QM9 and ZINC datasets. Training was stopped when a plateau in loss was observed even with a reduced learning rate of 1e-5.

PyTorch dataset was defined as follows:

class ChempleterDataset(Dataset):
   """
   PyTorch Dataset for SELFIES molecular representations.

   :param selfies_file: Path to CSV file containing SELFIES strings in a "selfies" column.
   :type selfies_file: str
   :param stoi_file: Path to JSON file mapping SELFIES symbols to integer tokens.
   :type stoi_file: str
   :returns: Integer tensor representation of tokenized molecule with dtype=torch.long.
   :rtype: torch.Tensor
   """

   def __init__(self, selfies_file, stoi_file):
      super().__init__()
      selfies_dataframe = pd.read_csv(selfies_file)
      self.data = selfies_dataframe["selfies"].to_list()
      with open(stoi_file) as f:
            self.selfies_to_integer = json.load(f)

   def __len__(self):
      return len(self.data)

   def __getitem__(self, index):
      molecule = self.data[index]
      symbols_molecule = ["[START]"] + list(sf.split_selfies(molecule)) + ["[END]"]
      integer_molecule = [
            self.selfies_to_integer[symbol] for symbol in symbols_molecule
      ]
      return torch.tensor(integer_molecule, dtype=torch.long)