ChemCensor DataBase

This repository provides the SQLite database used by ChemCensor, a precedent-based framework for evaluating the reaction chemical plausibility.

ChemCensor scores a proposed reaction by separating the reaction center (what changes during the reaction) from the functional-group context (what must be tolerated), and then checking whether similar reaction-center and functional-group compatibility patterns are supported by documented synthetic precedents. The resulting ChemCensor Score is an integer confidence level from 0 to 5, where higher values indicate stronger support from precedent chemistry.

The database is distributed as a compressed .zip archive containing a .sqlite file.

---

What is inside

The SQLite database stores precomputed ChemCensor knowledge used for reaction plausibility scoring. The knowledge base was constructed from the USPTO-full reaction dataset introduced by D. Lowe and includes reactions extracted by text-mining from United States patents published between 1976 and September 2016.

The DB is compatible with chemcensor>=1.0.0.

The database includes:

• reaction-center precedent information;
• functional-group compatibility information;
• relational mappings required by the ChemCensor scoring pipeline.

During ChemCensor scoring, the database is queried to determine whether a proposed retrosynthetic step is supported by previously observed reaction-center precedents and functional-group compatibility patterns.

Download and unpack

Download the archive from this Hugging Face repository and unzip it:

unzip ChemCensor-DB-U2-1.0.0.sqlite.zip

After unpacking, you should obtain a SQLite database file, for example:

ChemCensor-DB-U2-1.0.0.sqlite

---

Use with ChemCensor

The database is intended to be used together with the ChemCensor source code:

git clone https://github.com/insilicomedicine/ChemCensor.git
cd ChemCensor

Place the unpacked .sqlite file in the location expected by the ChemCensor configuration, or provide the database path according to the repository instructions.

A typical workflow is:

1. download this dataset;
2. unzip the .sqlite database;
3. install ChemCensor from the GitHub repository;
4. point ChemCensor to the unpacked database file;
5. run ChemCensor scoring for candidate reactions.

Please refer to the ChemCensor repository for the exact command-line or Python API usage.

---

Citation

If you use this database or ChemCensor in your work, please cite:

@misc{zagribelnyy2026singleanswerenoughrethinking,
      title={When Single Answer Is Not Enough: Rethinking Single-Step Retrosynthesis Benchmarks for LLMs}, 
      author={Bogdan Zagribelnyy and Ivan Ilin and Maksim Kuznetsov and Nikita Bondarev and Roman Schutski
                and Thomas MacDougall and Rim Shayakhmetov and Zulfat Miftakhutdinov
                and Mikolaj Mizera and Vladimir Aladinskiy and Alex Aliper and Alex Zhavoronkov},
      year={2026},
      eprint={2602.03554},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2602.03554}, 
}

License

This database is part of ChemCensor and is distributed under the same terms as the ChemCensor software. By downloading or using this dataset, you agree to the ChemCensor License Agreement (independent benchmarking and evaluation purposes only).

You may use the database solely to assess, compare, validate, or report the performance of models, algorithms, or methodologies together with ChemCensor, as described in the ChemCensor repository and paper. Use in products, automated pipelines, training or fine-tuning of models, redistribution, or incorporation into operational workflows is not permitted without prior written permission from Insilico Medicine AI Limited.

Insilico retains all intellectual property rights in the database. For the full definitions, permitted and prohibited uses, and legal terms, see the LICENSE file in the ChemCensor GitHub repository.