ai4chems/nist-logv-pure

NIST log-viscosity — pure components (Chew et al. 2024)

A mirror of the publicly-releasable subset of the pure-component liquid viscosity dataset curated by Chew, Sender, Kaplan, Chandrasekaran, Chief Elk, Browning, Kwak, Halls & Afzal in J. Cheminformatics 16:31 (2024), PMC10938832.

All data credit goes to the original authors. This repo simply mirrors Additional file 2 of the paper into a more ML-ready Hugging Face dataset format. The 858 rows of the full 4,440-row curated set that are copyright-restricted are not redistributed here.

License

This dataset is published under the Creative Commons Attribution-NonCommercial 4.0 International (CC-BY-NC 4.0) license, the same license attached by the original authors. You may not use the data for commercial purposes, including (most commonly) training models that you then sell access to. Attribution to Chew et al. 2024 is required.

What's in the box

• 3,582 rows covering pure-liquid viscosities at multiple temperatures.
• 957 unique molecules (the paper reports 1,005; the remainder is in the copyright-withheld subset).
• T range: 227.45–404.10 K. Viscosity range: 0.10–26.52 cP.

The published parquet schema:

column	dtype	description
mixture_T_id	list[int]	[compound_id, round(T)] — join key.
value	float	log10(viscosity_cP) — the regression target.
T	float	Temperature (K).
P	float	Pressure (bar). NaN in this dataset — the source CSV doesn't carry pressure; assume ~ atmospheric.
cmp_ids	list[int]	Length-1 list with the compound's internal id (see compounds.csv).
cmp_mole_fractions	list[float]	[1.0].
reference_id	str	Source literature citation string for this row (~13 distinct sources across the dataset).
MD_density, MD_FV, MD_Rg, MD_SP_E, MD_SP_V, MD_SP, MD_HV, MD_RMSD	float	Per-row molecular-dynamics physics-informed descriptors computed by Chew et al.

compounds.csv at the repo root maps compound_id → (smiles, name). Compound ids are local to this dataset (assigned 0..956 by sorted unique SMILES); they do not correspond to the chemixhub compound ids used in ai4chems/nist-logv.

Loading

from datasets import load_dataset
from huggingface_hub import hf_hub_download
import pandas as pd

REPO = "ai4chems/nist-logv-pure"

ds = load_dataset(REPO, split="train").to_pandas()
compounds = pd.read_csv(hf_hub_download(REPO, "compounds.csv", repo_type="dataset"))

# Look up SMILES for compound 0
smiles = compounds.set_index("compound_id")["smiles"].to_dict()
print(ds.iloc[0]["cmp_ids"], "→", smiles[ds.iloc[0]["cmp_ids"][0]])

Source & citation

Please cite the original paper:

Chew, A.K., Sender, M., Kaplan, Z., Chandrasekaran, A., Chief Elk, J., Browning, A.R., Kwak, H.S., Halls, M.D., Afzal, M.A.F. Advancing material property prediction: using physics-informed machine learning models for viscosity. J. Cheminformatics 16, 31 (2024). https://doi.org/10.1186/s13321-024-00820-5

Companion comment paper:

Fleck, M. Comment on "Advancing material property prediction: using physics-informed machine learning models for viscosity". J. Cheminformatics (2025). https://doi.org/10.1186/s13321-025-01070-9

Source CSV: Additional file 2 of the J. Cheminformatics article (available from the publisher's supplementary materials page).