liuganghuggingface/moldata-bace

BACE-1 inhibitor potency and binary activity

Complete untransformed DeepChem source table for BACE-1 inhibitor potency and binary activity. MoleculeNet is a benchmark view of the original experimental or quantum source.

Processed copy of bace from the molecular datasets collection. Structures are canonicalized and validated; labels are never averaged or filled. Prefer the original source for citation.

Where the data comes from

• Original data: published with the original study
• Original study: Subramanian et al. Computational modeling of beta-secretase 1 (BACE-1) inhibitors using ligand based approaches (2016)
• Redistributed by: MoleculeNet (DeepChem) (dataset page)

Data quality

Grade: care (rough; rules on the hub page).

• info: experiment. Label origin from the source's tags.
• warn: multi-source. Compiled from several labs or publications; assay protocols can differ (Walters 2023: 45% of same-compound IC50 pairs from two papers differ by >0.3 log units).
• good: valid structures. Rows RDKit cannot parse (kept in quarantine).
• bad: 71% undefined stereo. Molecules with at least one unassigned stereocenter (BACE: 71%).
• info: 28 stereoisomer sets. Groups of stereoisomers in the sampled structures; their labels can differ sharply.
• good: 0% PAINS alerts. Molecules matching PAINS substructures, which often interfere with assays.
• good: 45.7% minority class. Median share of the rarer class across classification tasks.
• good: 100% label coverage. Median share of molecules labelled per task; missing labels are unmeasured, not negative.
• good: 1,513 molecules per task. Median molecules with a label per task.
• good: source split. The source defines a split; shared structures between its parts leak information.

Tables

Config	Splits	Rows	Targets
000_bace	test, train, validation	1,513	pIC50 (regression, -log10(IC50 in mol/L)), Class (classification, binary)

data/ holds parseable rows with identical rows collapsed. observations/ holds every source row, including unparseable structures, for auditing. Added columns: SMILES_original, SMILES_canonical (RDKit; stereo, salts, and * kept), structure_key (JSON list of canonical components; use it to deduplicate or group), structure_status, wildcard_atom_count, source_row_first, observation_count, source_dataset, source_table. All original columns are strings, so censored values such as <10 survive; cast labels yourself.

Load

from torch_molecule.datasets import load_dataset
data = load_dataset("bace")

from datasets import load_dataset
ds = load_dataset("liuganghuggingface/moldata-bace", "000_bace")

Known issues

• pIC50 and Class represent related versions of the same endpoint; do not count them as independent assays. Preserve the supplied Model split column when reproducing that protocol.

License

Not stated by the source. Redistributed with the collection maintainer's approval; check the original source (https://github.com/deepchem/deepchem/blob/master/deepchem/molnet/load_function/bace_datasets.py) for its terms before reuse.

Citation

Cite the original study and data source:

• Subramanian et al. Computational modeling of beta-secretase 1 (BACE-1) inhibitors using ligand based approaches (2016)
• Wu et al. MoleculeNet: a benchmark for molecular machine learning (2018)
• Official DeepChem loader and task definitions