synrxn 1.0.0

Reaction Database for benchmarking

SynRXN

SynRXN is an open reaction benchmark repository for reproducible reaction-informatics evaluation.

SynRXN collects curated reaction datasets, canonical task folders, versioned data releases, and lightweight loading utilities for benchmarking atom-atom mapping, reaction classification, property prediction, reaction balancing, and synthesis/retrosynthesis workflows.

Highlights

• Five task families: aam, classification, property, rbl, and synthesis.
• Consistent tabular format: each dataset is a compressed CSV under Data/<task>/<name>.csv.gz.
• Stable identifiers: most curated rows use r_id; task-specific columns store reactions, labels, targets, splits, mappings, or references.
• Version-aware access: load data from Zenodo releases, GitHub tags, or exact Git commits.
• Reproducible benchmarking: use published splits when present, or generate deterministic repeated k-fold splits through synrxn.split.

Installation

SynRXN requires Python 3.11 or later.

pip install synrxn

Install optional dependencies when you need the broader tooling stack:

pip install "synrxn[all]"

For development:

git clone https://github.com/TieuLongPhan/SynRXN.git
cd SynRXN
python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"

Quick Start

from pathlib import Path
from synrxn.data import DataLoader

dl = DataLoader(
    task="classification",
    source="zenodo",
    version="1.0.0",
    cache_dir=Path("~/.cache/synrxn").expanduser(),
)

print(dl.available_names())
df = dl.load("schneider_b")
print(df.shape)
print(df.columns.tolist())

Use an exact commit for development snapshots you want to reproduce later:

from pathlib import Path
from synrxn.data import DataLoader

dl = DataLoader(
    task="property",
    source="commit",
    version="3e1612e2199e8b0e369fce3ed9aff3dda68e4c32",
    cache_dir=Path("~/.cache/synrxn").expanduser(),
    gh_enable=True,
)

df = dl.load("b97xd3")
print(df[["r_id", "ea", "dh"]].head())

Data Concept

The public data lives in Data/ and is grouped by benchmark task:

Folder	Purpose	Example datasets	Core columns
Data/aam/	Atom-atom mapping comparison	uspto_3k, golden, ecoli	ground_truth, mapper outputs, rxn
Data/classification/	Reaction class, template, and enzyme classification	uspto_50k_b, tpl_u, ecreact	rxn, labels, optional split
Data/property/	Reaction property prediction	b97xd3, rgd1, sn2	aam or rxn, target values, optional split
Data/rbl/	Reaction balancing and rebalancing	mos, mnc, mbs, complex	unbalanced rxn, balanced ground_truth
Data/synthesis/	Synthesis and retrosynthesis datasets	uspto_mit, uspto_50k, da	reactions, split/source metadata, optional reagents

Reproducible Splits

from pathlib import Path
from synrxn.data import DataLoader
from synrxn.split.repeated_kfold import RepeatedKFoldsSplitter

dl = DataLoader(
    task="property",
    source="zenodo",
    version="1.0.0",
    cache_dir=Path("~/.cache/synrxn").expanduser(),
)
df = dl.load("b97xd3")

splitter = RepeatedKFoldsSplitter(
    n_splits=5,
    n_repeats=2,
    ratio=(8, 1, 1),
    shuffle=True,
    random_state=1,
)
splitter.prepare_splits(df, stratify=None)
train_df, val_df, test_df = splitter.get_split(0, 0, as_frame=True)
print(len(train_df), len(val_df), len(test_df))

Documentation

• Documentation: https://synrxn.readthedocs.io/en/latest/
• Data release: https://doi.org/10.5281/zenodo.17297258
• Source code: https://github.com/TieuLongPhan/SynRXN
• Issues: https://github.com/TieuLongPhan/SynRXN/issues

Citation

If you use SynRXN in your research, please cite:

Tieu-Long Phan, Nhu-Ngoc Nguyen Song, and Peter F. Stadler. SynRXN: An Open Benchmark and Curated Dataset for Computational Reaction Modeling. Scientific Data 13, 625 (2026). https://doi.org/10.1038/s41597-026-07260-w

@article{phan2026synrxn,
  title = {SynRXN: An Open Benchmark and Curated Dataset for Computational Reaction Modeling},
  author = {Phan, Tieu-Long and Nguyen Song, Nhu-Ngoc and Stadler, Peter F.},
  journal = {Scientific Data},
  volume = {13},
  pages = {625},
  year = {2026},
  doi = {10.1038/s41597-026-07260-w},
  url = {https://www.nature.com/articles/s41597-026-07260-w}
}

License

This project is licensed under the MIT License. Dataset-specific terms are summarized in Data/LICENSE when applicable.

Acknowledgments

This project has received funding from the European Union's Horizon Europe Doctoral Network programme under the Marie Sklodowska-Curie grant agreement No. 101072930 (TACsy).