liuganghuggingface/moldata-polyid

PolyID

NREL (now National Laboratory of the Rockies) public constructed-polymer training input with experimental Tg and Tm labels, stereochemistry and monomer metadata.

Processed copy of polyid 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: https://github.com/NatLabRockies/polyID
• Original study: PolyID: Artificial Intelligence for Discovering Performance-Advantaged and Sustainable Polymers

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).
• info: 17 stereoisomer sets. Groups of stereoisomers in the sampled structures; their labels can differ sharply.
• warn: 73% label coverage. Median share of molecules labelled per task; missing labels are unmeasured, not negative.
• bad: 124 molecules per task. Median molecules with a label per task.

Tables

Config	Splits	Rows	Targets
000_stereopolymer_input_nopush	full	229	Tg (regression, C as explicit Tg_units column), Tm (regression, C as explicit Tm_units column)

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("polyid")

from datasets import load_dataset
ds = load_dataset("liuganghuggingface/moldata-polyid", "000_stereopolymer_input_nopush")

Known issues

• Public code input is smaller than the approximately 1,800-polymer study described by the Notre Dame index.
• Representations include constructed stereopolymer/oligomer structures, not necessarily minimal repeat units.
• Same repeat unit may differ in stereochemistry, molecular weight, and experimental conditions.

License

BSD-3-Clause repository

Citation

Cite the original study and data source:

• PolyID: Artificial Intelligence for Discovering Performance-Advantaged and Sustainable Polymers