Modeling Diverse Chemical Reactions for Single-step Retrosynthesis via Discrete Latent Variables

This is the code of paper Modeling Diverse Chemical Reactions for Single-step Retrosynthesis via Discrete Latent Variables. Huarui He, Jie Wang, Yunfei Liu, Feng Wu. CIKM 2022.

Reproduce the Results

1. Environmental setup

Please ensure that conda has been properly initialized, i.e. conda activate is runnable. Then

bash -i scripts/setup.sh
conda activate retro

2. Data preparation

Download the raw (cleaned and tokenized) data from Google Drive by

python scripts/download_raw_data.py --data_name=$DATASET

where DATASET is one of [USPTO_50k, USPTO_DIVERSE]
Run create_1toN_map.ipynb in data/ to derive 1-to-N answer dict for the each dataset. It is okay to only download the dataset you want.

Then run the preprocessing script by

sh scripts/preprocess.sh $DATASET

3. Model training and validation

Run the training script by

export CUDA_VISIBLE_DEVICES=7
sh scripts/train_g2s.sh "cvae-gru-K20" "USPTO_DIVERSE" "20"

Optionally, run the evaluation script by

sh scripts/validate.sh "cvae-gru-K20" "USPTO_DIVERSE"

Note: the evaluation process performs beam search over the whole val sets for all checkpoints. It can take tens of hours.

4. Testing

Then run the testing script by

sh scripts/predict.sh "cvae-gru-K20" "USPTO_DIVERSE"

which will first run beam search to generate the results for all the test inputs, and then computes the average top-k accuracies.

File tree

RetroDCVAE
├─ README.md
├─ data
│  ├─ USPTO_50k
│  │  ├─ src-train.txt
│  │  ├─ ...
│  │  └─ tgt-test.txt
│  ├─ USPTO_DIVERSE
│  │  ├─ src-train.txt
│  │  ├─ ...
│  │  └─ tgt-test.txt
│  └─ create_1toN_map.ipynb
├─ models
│  ├─ VAR
│  │  ├─ utils.py
│  │  └─ var_dec.py
│  ├─ attention_xl.py
│  ├─ dgat.py
│  ├─ dgcn.py
│  ├─ graph2seq_series_rel.py
│  ├─ graphfeat.py
│  ├─ model_utils.py
│  └─ seq2seq.py
├─ predict.py
├─ preprocess.py
├─ scripts
│  ├─ download_checkpoints.py
│  ├─ download_raw_data.py
│  ├─ predict.sh
│  ├─ preprocess.sh
│  ├─ setup.sh
│  ├─ train_g2s.sh
│  └─ validate.sh
├─ train.py
├─ utils
│  ├─ chem_utils.py
│  ├─ data_utils.py
│  ├─ parsing.py
│  ├─ rxn_graphs.py
│  └─ train_utils.py
└─ validate.py

Citation

If you find this code useful, please consider citing the following paper.

@inproceedings{CIKM22_RetroDCVAE,
  author={Huarui He and Jie Wang and Yunfei Liu and Feng Wu},
  booktitle={Proc. of CIKM},
  title={Modeling Diverse Chemical Reactions for Single-step Retrosynthesis via Discrete Latent Variables},
  year={2022}
}

Acknowledgement

We refer to the code of Graph2SMILES. Thanks for their contributions.