🎯 Projecting Molecules into Synthesizable Chemical Spaces (ICML 2024)
ChemProjector and its derivatives (e.g. SynFormer) have been deprecated. Please refer to our latest model PrexSyn: Efficient and Programmable Exploration of Synthesizable Chemical Space:
- Code: https://github.com/luost26/prexsyn
- Paper: https://arxiv.org/abs/2512.00384
- Documentation: https://prexsyn.readthedocs.io/
Please clone the repository with the --recurse-submodules flag to include the third-party submodules.
git clone --recurse-submodules https://github.com/luost26/ChemProjector.git# Install conda environment
conda env create -f env.yml -n chemprojector
conda activate chemprojector
# Install ChemProjector package
pip install -e .The default CUDA version is 11.8. If you have to use a different version, please modify the env.yml file accordingly.
We provide preprocessed building block data. You can download it from here and put it in the data directory.
However, the data is derived from Enamine's building block catalog, which are available only upon request.
Therefore, you should first request the data from Enamine here and download the US Stock catalog into the data directory.
Then run the following script which will check whether you have a copy of the Enamine's catalog and unarchive the preprocessed data for you:
python unarchive_wizard.pyYou may also process the building block data by yourself. Please refer to the scripts/preprocess_data directory for more details.
You can download the trained weights from here and put them in the data/trained_weights directory.
| Item | Link | Destination |
|---|---|---|
| Preprocessed building blocks and reaction templates (password protected) | https://drive.google.com/file/d/1scui0RZ8oeroDAafnw4jgTi3yKtXxXpe/view?usp=drive_link | data/processed.zip |
| Enamine building block US stock catalog (to unzip the password protected preprocessed files) | https://enamine.net/building-blocks/building-blocks-catalog | data/Enamine_*.sdf |
| Trained weights | https://drive.google.com/drive/folders/1T9f9MsEAsAjPV8GR0pXimHKCvq97SIzs?usp=drive_link | data/trained_weights/original_default.ckpt, data/trained_weights/original_split.ckpt |
You can create a list of SMILES strings in CSV format (example: data/example.csv) and run the following command to project them into the synthesizable chemical space.
python sample.py \
--model-path data/trained_weights/original_default.ckpt \
--input data/example.csv \
--output results/example.csv| Test set | Command |
|---|---|
| Test split | ./scripts/synthesis_planning_test_split.sh |
| ChEMBL | ./scripts/synthesis_planning_chembl.sh |
| Enamine REAL | ./scripts/synthesis_planning_enamine.sh |
Please refer to the scripts/sbdd directory for details.
Please refer to the scripts/goal_directed directory for details.
python train.py ./configs/original_default.yml@inproceedings{luo2024chemprojector,
title={Projecting Molecules into Synthesizable Chemical Spaces},
author={Shitong Luo and Wenhao Gao and Zuofan Wu and Jian Peng and Connor W. Coley and Jianzhu Ma},
booktitle={Forty-first International Conference on Machine Learning},
year={2024}
}