liuganghuggingface/moldata-opc2025

NeurIPS Open Polymer Challenge 2025

MD-simulated glass transition temperature, fractional free volume, thermal conductivity, density, and radius of gyration of polymers from the NeurIPS 2025 Open Polymer Challenge. Holds the released test labels as the private and public test parts.

Part of the Molecular Datasets collection.

Get started

from datasets import load_dataset
dataset = load_dataset("liuganghuggingface/moldata-opc2025", "default")

Where the data comes from

• Original data: https://www.kaggle.com/datasets/alexliu99/neurips-open-polymer-prediction-2025-test-data
• Original study: Open Polymer Challenge: Post-Competition Report

Tables

Config	Splits	Rows	Targets
default	private_test, public_test	3,502	Tg Glass transition temperature · simulated (regression; °C), FFV Fractional free volume (regression; units not specified), Tc Thermal conductivity · simulated (regression; W/(m·K)), Density Polymer density · simulated (regression; units not specified), Rg Radius of gyration · simulated (regression; units not specified)

data/ contains rows with readable molecular structures; identical observations are collapsed. observations/ contains all source rows.

Source label columns are text, including values such as <10. SMILES_original retains the input representation; SMILES_canonical provides a canonical representation. structure_key groups molecular components, and structure_status records parsing outcomes.

Data quality

• computed: Labels are computed (for example DFT or molecular dynamics).
• all structures readable: All 3,502 SMILES were read with RDKit.
• 32.9% measured per task: In a typical task, 32.9% of molecules have a value. A missing value means not measured, not inactive.
• 14 extreme values: 14 of 4,524 values lie more than 6 robust standard deviations from the median of their target (on a log scale when values span 1,000-fold or more). They are often typing or unit mistakes, and sometimes real; look at them before training.
• 239 molecules per task: A typical task has 239 labeled molecules. Results on fewer than 1,000 molecules vary a lot between splits.
• source split: private test / public test: The source divides the data into private test (3,207 rows), public test (295 rows). Use this division to compare with published results.

Dataset notes

• The competition training set is available only on Kaggle after accepting the competition rules, so it is not included.
• Training on these labels makes results incomparable with competition scores.
• Computational labels.

License

Dataset-wide license: unspecified. Original source.

Citation

Cite the original study and data source:

• Open Polymer Challenge: Post-Competition Report
• Official challenge site