NAG2G: Node-Aligned Graph-to-Graph Model

Welcome to the NAG2G (Node-Aligned Graph-to-Graph) repository! NAG2G is a state-of-the-art neural network model for retrosynthesis prediction.

🔥 Latest Updates

• 🗓️ May 2025 — 🌿 with_stereoisomerism branch is now publicly available

  📝 This branch was completed in 2023 but wasn’t released until now — opened in response to community interest.

  
  

  _{New in this branch:}

  • 💊 Enhanced Stereochemistry Support

    • Direct prediction of stereochemical features (e.g., chirality) from model outputs
    • No post-processing required for stereochemical reconstruction

  • ⇄ Unified Bidirectional Synthesis

    • Single model supports both retrosynthesis and forward synthesis

• 🗓️ August 2024 — 💻 Initial codebase released (main branch)

• 🗓️ February 2024 — 🧪 Paper published in JACS Au

• 🗓️ September 2023 — 📄 Preprint available on ArXiv

Environment Setup

To begin working with NAG2G, you'll need to set up your environment. Below is a step-by-step guide to get you started:

# Install Uni-Core
git clone https://github.com/dptech-corp/Uni-Core
cd Uni-Core
pip install .
cd -

# Install Unimol plus
cd unimol_plus
pip install .
cd -

# Install additional dependencies
pip install rdchiral transformers tokenizers omegaconf rdkit

Datasets and Pretrained Weights

You can obtain the dataset USPTO-50k and pretrained model weights for USPTO-50k from the Google Drive:

Model Validation

To validate the NAG2G model with the provided weights, follow the instructions below:

When using a dataset that does not include reactants, you need to modify the valid.sh script. Specifically, add the --no_reactant command in line 95 in the code.

When using your own dataset, please modify the data_path in the valid.sh script.

# Execute the validation script with the specified checkpoint file
sh valid.sh path2weight/NAG2G_unimolplus_uspto_50k_20230513-222355/checkpoint_last.pt

Data Preprocessing Instructions

If you need to regenerate the dataset, please refer to the code inside the data_preprocess directory.

cd data_preprocess
python lmdb_preprocess <input_csv> <output_lmdb>

Two sample CSV files are provided for reference:

• sample.csv: This sample includes given reactants.
• sample_without_reactants.csv: This sample does not include given reactants.

---

For any questions or issues, please open an issue on our GitHub repository.

Thank you for your interest in NAG2G!