Large chem-informatics dataset of 7'131'914'291 small molecules with up to 50 heavy atoms, aggregated from PubChem, GDB13, and Enamine REAL, and enriched with machine-searchable representations of every molecule, growing over time. Those currently span 28 billion binary structural fingerprints computed with RDKit and CDK - MACCS with 166 dimensions, PubChem with 881, ECFP4 with 2048, and FCFP4 with 2048 - as well as 3D conformer geometries generated with ETKDG and optimized with MMFF94. Distributed as Parquet shards with pre-constructed USearch indexes for real-time similarity search and clustering across billions of molecules, hosted anonymously on the AWS Open Data platform, and accompanied by a bigger-than-memory Python pipeline with optional GPU acceleration for reproducing every stage.

If you use this dataset, please cite it as below.