liuganghuggingface/moldata-moleculeace_chembl239_ec50

MoleculeACE CHEMBL239 EC50

Activity-cliff prediction benchmark for CHEMBL239 using EC50 activity labels from ChEMBL. Includes the original train/test assignments and activity-cliff annotations.

Part of the Molecular Datasets collection.

Get started

from datasets import load_dataset
dataset = load_dataset("liuganghuggingface/moldata-moleculeace_chembl239_ec50", "000_CHEMBL239_EC50")

Where the data comes from

• Original data: https://github.com/molML/MoleculeACE
• Original study: Exposing the Limitations of Molecular Machine Learning with Activity Cliffs
• Built on: ChEMBL

Tables

Config	Splits	Rows	Targets
000_CHEMBL239_EC50	test, train	1,721	y [pEC50/pKi] (regression; negative log10 molar potency)

data/ contains rows with readable molecular structures; identical observations are collapsed. observations/ contains all source rows.

Source label columns are text, including values such as <10. SMILES_original retains the input representation; SMILES_canonical provides a canonical representation. structure_key groups molecular components, and structure_status records parsing outcomes.

Data quality

• experiment: Source label origin.
• valid structures: Molecular structure parsing results.
• 34% undefined stereo: Source structures with unspecified stereochemistry.
• 2% PAINS alerts: Molecules matching PAINS substructures, which often interfere with assays.
• 1,721 labeled molecules: Median molecules with a label per task.
• source split: Source split assignments and shared-structure counts.

Dataset notes

• Activity-cliff labels depend on the original similarity and potency thresholds.
• Targets pooled from ChEMBL can contain heterogeneous assays and measurement conditions.
• The dataset includes its original training and test assignments.

License

MIT (benchmark repository); underlying ChEMBL data CC BY-SA 3.0

Citation

Cite the original study and data source:

• Exposing the Limitations of Molecular Machine Learning with Activity Cliffs