ChemSearch DataSets

Original datasets used for publication in ChemSearch There are two data sets obtained from ChEMBL and Pubchem, respectively. The ChEMBL dataset was generated by downloading a bulk SDF file of all 2.3 million compounds as of February 2023. Molecules were parsed to SMILES strings using RDKit, with no additional standardisation performed. The collection of SMILES strings was randomly shuffled. A subset was then extracted from the collection by selecting the first 1,000,000 records.There are two data sets obtained from ChEMBL and Pubchem, respectively. The PubChem dataset was generated by downloading the SMILES strings of all 115 million compounds as of April 2023. The molecules were randomly shuffled, then split into chunks of 1,000,000 molecules. The shuffling process was performed to ensure each chunk was of approximately the same size in memory, and equally demanding computationally.