liuganghuggingface/moldata-openadmet_pxr_challenge

OpenADMET PXR challenge

OpenADMET pregnane X receptor challenge data with molecular structures, experimental endpoint labels, and source training/test partitions.

Part of the Molecular Datasets collection.

Get started

from datasets import load_dataset
dataset = load_dataset("liuganghuggingface/moldata-openadmet_pxr_challenge", "pxr_challenge_96_compound_uscale_semi_pure_train")

Where the data comes from

• Original data: https://huggingface.co/datasets/openadmet/pxr-challenge-train-test

Tables

Config	Splits	Rows	Targets
pxr_challenge_96_compound_uscale_semi_pure_train	train	96	Corrected Semi-Pure pEC50 (log) (regression; -log10(M)), Semi-Pure Emax normalized (regression; dimensionless normalized Emax), Semi-Pure Product Yield (%) Semi-Pure Product Yield (regression; percent)
dose_response	train, test	8,683	none
pxr_challenge_test_blinded	test	513	none
pxr	train, test_phase_1, test_phase_2	4,652	pEC50 PXR activation (pEC50) (regression; units not specified), Emax_estimate (log2FC vs. baseline) Emax estimate (log2FC vs. baseline) (regression; units not specified), Emax.vs.pos.ctrl_estimate (dimensionless) Emax.vs.pos.ctrl estimate (dimensionless) (regression; units not specified)
pxr_challenge_counter_assay_train	train	2,859	pEC50 PXR activation (pEC50) (regression; units not specified), Emax_estimate (log2FC vs. baseline) Emax estimate (log2FC vs. baseline) (regression; units not specified), Emax.vs.pos.ctrl_estimate (dimensionless) Emax.vs.pos.ctrl estimate (dimensionless) (regression; units not specified)
pxr_challenge_htchem_libraries_train	train	456	Corrected Crude pEC50 (log) (regression; -log10(M)), Crude Emax normalized (regression; dimensionless normalized Emax), Crude Product Yield (%) Crude Product Yield (regression; percent)
pxr_challenge_single_concentration_train	train	21,003	log2_fc_estimate log2 fc estimate (regression; units not specified)
pxr_challenge_structure_test_blinded	test	184	none

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: Labels are measured in a lab.
• all structures readable: All 38,446 SMILES were read with RDKit.
• 41.9% undefined stereo: 41.9% of molecules have a stereocenter whose configuration is not given, so mirror-image forms cannot be told apart.
• 1 of 34 stereoisomer sets differs: Found 34 groups of molecules that differ only in stereochemistry. In 1, the values are 10-fold apart (log-scale labels), more than one standard deviation apart, or in different classes. That can be a real effect or a mistake in the source; a model that ignores stereochemistry treats these molecules as the same.
• 1.1% PAINS alerts: 1.1% of molecules contain a PAINS substructure, which often looks active in screening assays without real activity.
• 100% measured per task: In a typical task, 100% of molecules have a value. A missing value means not measured, not inactive.
• 408 extreme values: 408 of 44,334 values lie more than 6 robust standard deviations from the median of their target (on a log scale when values span 1,000-fold or more). They are often typing or unit mistakes, and sometimes real; look at them before training.
• 456 molecules per task: A typical task has 456 labeled molecules. Results on fewer than 1,000 molecules vary a lot between splits.
• source split: train / test / test phase 1 / test phase 2: The source divides the data into train (35,698 rows), test (2,235 rows), test phase 1 (253 rows), test phase 2 (260 rows). Use this division to compare with published results.

Dataset notes

• The PXR-null counterscreen differs from receptor induction.
• Records include well-level curves, crude reaction products, and purified molecules with assay and purity context.
• 2026-09-02 corrected ten semipure compound structures; older versions can have wrong constitutional isomers.
• Unblinded phase1/phase2 test tables supersede withheld labels for their respective subsets.

License

cc-by-4.0

Citation

Cite the original study and data source:

• Original release dataset card, assay methods and changelog
• OpenADMET publications and representative uses