liuganghuggingface/moldata-polyomics

PolyOmics

RadonPy Consortium computational database with many properties, material architectures, and simulation conditions; all_data is the merged tabular release.

Processed copy of polyomics 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://huggingface.co/datasets/yhayashi1986/PolyOmics
• Original study: Omics-scale polymer computational database transferable to real-world artificial intelligence applications
• Original study: PolyOmics dataset DOI

Data quality

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

• info: computed. Label origin from the source's tags.
• good: valid structures. Rows RDKit cannot parse (kept in quarantine).
• info: 4 stereoisomer sets. Groups of stereoisomers in the sampled structures; their labels can differ sharply.
• good: R² ceiling ≈ 1.00. Replicate SD is 0.00× the label SD, so no model can explain more than about 100% of the variance.
• bad: 14% label coverage. Median share of molecules labelled per task; missing labels are unmeasured, not negative.
• good: 13,590 molecules per task. Median molecules with a label per task.

Tables

Config	Splits	Rows	Targets
000_all_data	full	176,096	qm_total_energy_monomer1 (regression, not_verified), qm_homo_monomer1 (regression, not_verified), qm_lumo_monomer1 (regression, not_verified), qm_dipole_x_monomer1 (regression, not_verified), qm_dipole_y_monomer1 (regression, not_verified), qm_dipole_z_monomer1 (regression, not_verified), qm_polarizability_monomer1 (regression, not_verified), qm_polarizability_xx_monomer1 (regression, not_verified) and 112 more

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("polyomics")

from datasets import load_dataset
ds = load_dataset("liuganghuggingface/moldata-polyomics", "000_all_data")

Known issues

• Not every property is available for every record.
• Simulation validity fields and conditions must be filtered for each task.
• all_data overlaps subtype files, so redundant subtype copies are omitted.
• Atomistic snapshot archive (~97 GB complete repository) is catalogued separately and not fully mirrored.

License

CC-BY-4.0

Citation

Cite the original study and data source:

• Omics-scale polymer computational database transferable to real-world artificial intelligence applications
• PolyOmics dataset DOI