RXNEmb is a novel reaction-level embedding descriptor generated via the pre-trained model RXNGraphormer. It captures chemical bond formation and cleavage patterns, enabling data-driven reaction classification, mechanistic interpretation, and reaction space visualization.

🚀 Overview

The preprint paper is available at https://arxiv.org/pdf/2601.03689 .

🔧 Installation

pip install rxnemb

💡 Generate RXNEmb Descriptor

Basic Usage

Example 1: Generate reaction embeddings from SMILES list

from rxnemb import RXNEMB

# Initialize with CPU (default)
# To use GPU: device='cuda' or device='cuda:0'
generator = RXNEMB(device='cpu')

# prepare reaction SMILES
# here are some examples
rxn_smiles_lst = [
    "O=C(O)c1ccccc1.[Cl-]>>O=C(Cl)c1ccccc1",  
    "C1CCOC1.C1CCOC1.O.O=C(O)C(Br)c1ccccc1>>OCC(Br)c1ccccc1",
    # ... more 
]

# output
rxn_emb = generator.gen_rxn_emb(rxn_smiles_lst)
print(rxn_emb.shape) # torch.Size([2, 768])

Use Customized Model

If you want to train your own model from scratch, please refer to the instructions in the RXNGraphormer repository.

After training is done, you can load the model by specifying the path to the directory containing your model weights and configuration file:

from rxnemb import RXNEMB

generator = RXNEMB('/path/to/your/own/model')

📚 Citation

If you use RXNEmb in your research, please cite:

@misc{RXNEmb_liu2026,
      title={A Pre-trained Reaction Embedding Descriptor Capturing Bond Transformation Patterns}, 
      author={Weiqi Liu and Fenglei Cao and Yuan Qi and Li-Cheng Xu},
      date = {2026-01-07},
      eprint={2601.03689},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      doi = {10.48550/arXiv.2601.03689},
      url={https://arxiv.org/abs/2601.03689}, 
      pubstate = {prepublished},
}