liuganghuggingface/moldata-b3db_classification

TDC b3db_classification

Complete TDC source table: B3DB blood-brain barrier permeability.

Processed copy of b3db_classification 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: A curated diverse molecular database of blood-brain barrier permeability with chemical descriptors
• Redistributed by: Therapeutics Data Commons (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: 35% undefined stereo. Molecules with at least one unassigned stereocenter (BACE: 71%).
• info: 1351 stereoisomer sets. Groups of stereoisomers in the sampled structures; their labels can differ sharply.
• warn: 5% PAINS alerts. Molecules matching PAINS substructures, which often interfere with assays.
• good: 35.3% minority class. Median share of the rarer class across classification tasks.
• good: 6,167 labeled molecules. Median molecules with a label per task.
• good: 97% agree with bbb_martins. Blood-brain barrier penetration: on 1,642 shared molecules, 97% agree within same class.

Tables

Config	Splits	Rows	Targets
000_b3db_classification	full	6,167	Y (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("b3db_classification")

from datasets import load_dataset
ds = load_dataset("liuganghuggingface/moldata-b3db_classification", "000_b3db_classification")

Known issues

• TDC is a processed distribution of original studies. Preserve source-specific endpoint definitions and transforms; a matching molecule in another dataset does not imply the same measurement.
• This current registry entry was not matched to a task-specific section on the checked documentation page; the original study citation needs further verification.

License

Not stated by the source. Redistributed with the collection maintainer's approval; check the original source (https://tdcommons.ai/single_pred_tasks/adme/) for its terms before reuse.

Citation

Cite the original study and data source:

• Meng et al. A curated diverse molecular database of blood-brain barrier permeability with chemical descriptors (2021)
• Huang et al. Therapeutics Data Commons: Machine Learning Datasets and Tasks for Drug Discovery and Development (NeurIPS 2021)
• TDC b3db_classification task, original study references and recommended evaluation