🗄️

hMOF-ML Dataset

Version v1.0 | ML-Ready Dataset for High-Throughput Screening

📊 Dataset Overview

A massive, machine learning-ready dataset of 137,953 hypothetical MOF structures with comprehensive structural and geometric features. This dataset enables high-throughput computational screening, material discovery, and gas adsorption prediction for various applications. Note: This dataset does not include hydrogen storage labels and is primarily designed for inference and candidate screening tasks.

Key Statistics

137,953

Hypothetical MOFs

102

Building Blocks

Screening Ready

JSON

ML-Ready Format

📊 Data Sources

🧱 Structural Data

From Northwestern University databases containing 137,953 hypothetically constructed MOF CIF files assembled from 102 building blocks. The hMOF-ML Database was originally developed by Christopher E. Wilmer and collaborators and is now curated by the Snurr Research Lab.

🔬 Geometric Features

From MOFXDB API including original geometric descriptors:

GSA — Gravimetric Surface Area (m²/g)
VSA — Volumetric Surface Area (m²/cm³)
VF — Void Fraction
LCD — Largest Cavity Diameter (Å)
PLD — Pore Limiting Diameter (Å)

🔧 Data Processing Pipeline

We performed comprehensive processing to transform raw hMOF-ML data into a ML-ready format:

⭐ 1. Automated Structure Data Extraction

Developed Pymatgen-based Python automation scripts
Automatic CIF parsing from Northwestern databases
Parsed and extracted: unit cell parameters (a, b, c, α, β, γ), atomic types and coordinates

⭐ 2. Derived Feature Calculation

Computed additional features to enrich the feature space:

Total metal atom count
Metal mass fraction
Metal molar fraction
Lattice volume

⭐ 3. Data Integration & Standardization

Unified JSON database containing:

Atomic composition features
Lattice parameter features
Pore geometry features

For a complete list of all accessible attributes and their descriptions, please refer to the Data Format documentation.

🎯 Applications

Machine Learning Inference: Direct use for model inference to predict various gas adsorption properties
High-Throughput Screening: Screen high-performance MOF candidates for specific applications
Gas Adsorption Prediction: Predict adsorption behavior for various gases (H₂, CH₄, CO₂, etc.)
Material Discovery: Discover novel MOF structures with optimized properties
Comparative Studies: Compare hypothetical structures against experimental MOFs

💡 Key Advantages

✅ ML-Ready

Structured JSON format with comprehensive features, ready for direct model inference

📊 Feature-Rich

Comprehensive structural and geometric features

🔬 Massive Scale

137,953 hypothetical structures for comprehensive screening

🎯 Ready-to-Use

Unified format optimized for inference tasks

⚠️ Important Note

This dataset does NOT include hydrogen storage labels (UG_at_PS, UV_at_PS, UG_at_TPS, UV_at_TPS). It is primarily designed for inference tasks and candidate screening. For training models on hydrogen storage prediction, please use the H2MOF-ML Dataset.

Download

The dataset is available in JSON format. You can download it via:

Browser: Visit the Hugging Face dataset page
Python: Use the download script from Quick Start guide

📚 Citation

If you use this dataset, please cite the following references:

Recent Curation (2023)

N. Scott Bobbitt, Kaihang Shi, Benjamin J. Bucior, Haoyuan Chen, Nathaniel Tracy-Amoroso, Zhao Li, Yangzesheng Sun, Julia H. Merlin, J. Ilja Siepmann, Daniel W. Siderius, and Randall Q. Snurr.
MOFX-DB: An Online Database of Computational Adsorption Data for Nanoporous Materials.
Journal of Chemical & Engineering Data, 2023, 68 (2), 483–498.
DOI: 10.1021/acs.jced.2c00583

Foundational Generation Study (2012)

Christopher E. Wilmer, Michael Leaf, Chang Yeon Lee, Omar K. Farha, Brad G. Hauser, Joseph T. Hupp, and Randall Q. Snurr.
Large-scale screening of hypothetical metal–organic frameworks.
Nature Chemistry, 2012, 4, 83–89.
DOI: 10.1038/nchem.1192

🔗 Related Dataset

For training machine learning models on hydrogen storage prediction with labeled GCMC data, see: H2MOF-ML Dataset