AqSolDB collects experimental aqueous-solubility measurements for 9,982 drug compounds. Each row contains a chemical name, a SMILES representation, and Y, the aqueous solubility expressed as logS, or logarithmic molarity. section 4.1.1 section 4.1.2
Use scaffold splitting to keep structurally similar molecules in the same partition, producing a more demanding test of generalization than random splitting. Inspect the target distribution before modeling: values span roughly -12 to 1.97 logS, with extreme observations that may require chemical or statistical outlier review. section 4.1.1 section 4.1.2
Listing 4.1 Load AqSolDB with Scaffold-based Splits
from tdc.single_pred import ADME data = ADME(name = 'Solubility_AqSolDB') split = data.get_split(method = 'scaffold') train_df, valid_df, test_df = split['train'], split['valid'], split['test']
Load AqSolDB through the Therapeutics Data Commons API and obtain train, validation, and test data using scaffold-based splits. section 4.1.1
Practical preparation can include:
- Treat logS differences as multiplicative: one log unit represents a tenfold solubility change. section 4.1.2
- Filter molecules to a single fragment with RDKit when disconnected structures are inappropriate for the modeling task. section 4.1.2
