Open Reaction Database (ORD): Structured Organic Reaction Data with USPTO Grant Reactions
- Provider
- Open Reaction Database project (Kearnes, Coley et al.)
- License
- CC-BY-SA-4.0 (source)
- Size
- 101 files
- Format
- Parquet (in sampled files)
- Stored at
- Hugging Face (https://huggingface.co/datasets/open-reaction-database/ord-data)
- Access
- Free and public. No account with the data provider is needed.
- DOI
- 10.1021/jacs.1c09820
- Category
- Chemistry & Materials
- Source
- https://doi.org/10.1021/jacs.1c09820
About this dataset
The Open Reaction Database is an open-access repository of organic chemical reactions recorded in the ORD schema, a Protocol Buffers data model that captures reactants, reagents, catalysts, solvents, amounts, conditions (temperature, pressure, stirring, illumination, flow), workup steps, outcomes, yields, product characterisation and provenance for each reaction. Contributions include high-throughput experimentation (HTE) screens such as Buchwald-Hartwig and Suzuki coupling plates, data from published papers, and the reactions text-mined from United States patent grants by Lowe (USPTO, 1976 to 2016). This Hugging Face repository is the official data store for the ord-data GitHub project: 94 Parquet files (about 1.3 GB, measured 2026-11-15), one ORD dataset per file under data/ /, where each row is a serialized Reaction message. The default configuration holds 179,208 reactions from non-patent datasets; the uspto-grants configuration holds 1,771,032 patent reactions split into yearly datasets, and uspto-mit gives the train, validation and test splits used for reaction prediction. Load files with ord_schema.datasets.load_dataset and convert to JSON or pbtxt with the protobuf helpers. Used for reaction yield prediction, retrosynthesis, reaction condition recommendation and forward prediction. Data are released under CC-BY-SA-4.0 (code under Apache-2.0); cite Kearnes et al., J. Am. Chem. Soc. 2021.
How to load this dataset in Python
from datasets import load_dataset
ds = load_dataset("open-reaction-database/ord-data")
print(ds)Data structure
File types in sampled files: Parquet
Files
Example file: data/00/ord_dataset-00005539a1e04c809a9a78647bea649c.parquet
Analyze this dataset without downloading it
On Scigantic, Open Reaction Database (ORD): Structured Organic Reaction Data with USPTO Grant Reactions opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.
This page is also available as Markdown and as JSON.
Related Chemistry & Materials datasets
- Open Molecules 2025 (OMol25) Neutral Validation Set, ColabFit Conversion
- COCONUT 2.0 natural products, October 2026 release: 738,794 structures with organisms and NPClassifier classes, as Parquet
- Solution analyzer instrument output with ASM: 22 exports from Beckman Coulter PharmSpec, NovaBio Flex2, Roche Cedex Bio HT
- Graphene Design Universe 256K: Atomistic Porous and Patterned Graphene Geometries
- Binding affinity (SPR) instrument output with ASM: 15 exports from Cytiva Biacore Insight, Cytiva Biacore T200 Control, Cytiva Biacore T200 Evaluation
- Systematic Magnesium (Mg) Structure Set for Machine Learning Potentials and Defects: DFT (Poul et al. 2023, Edmond), ColabFit Conversion
- QM7b and AlphaML: LR-CCSD and Hybrid DFT Static Dipole Polarizabilities for 7,211 Molecules, ColabFit Conversion
- Open Molecules 2025 (OMol25) Validation Set, ColabFit Conversion