Retrosynthesis-Reaction-Pathway

The web application generates complete breakdown pathways for the given product molecule.

For inference purpose a checkpoint created after 44000 steps is provided here.

Installations

1. rdkit
2. OpenNMT

Dataset Generation

Clone the OpenNMT repo

Follow the setup instructions in the repo.

Datasets can be downloaded here

Datasets should be plcaed in the data directory in the OpenNMT-py directory using a different data folder for each version of the dataset (no augmentation, 4x, 16x and 40x augmentation)

In the OpenNMT-py directory, run the following command:

python preprocess.py -train_src data/{dataset_directory}/{train_source}.txt \
-train_tgt data/{dataset_directory}/{train_targets}.txt -valid_src data/{dataset_directory}/{valid_source}.txt \
-valid_tgt data/{dataset_directory}/{valid_targets}.txt -save_data data/{dataset_directory}/{dataset_name} \
-src_seq_length 1000 -tgt_seq_length 1000 -src_vocab_size 1000 -tgt_vocab_size 1000 -share_vocab

Model Training

Move the model_config.yml file in this directory to the config directory in the OpenNMT-py directory.

Update the data and save_model fields for the dataset created above

In the OpenNMT-py directory, run the following command:

python train.py -config config/model_config.yml

Running the web app

1. Download the model in the Retrosynthesis-Reaction-Pathway directory. If you have created your own model then change the model path here.
2. Run python app.py and vsit the link displayed on the terminal/Command prompt.
3. Enter the correct SMILES in the text box and click on upload.
4. The molecule visualisation and the complete pathway will be shown.

Demo

TODO

• Improve the frontend.
• Add option to enter IUPAC name instead of SMILES as input.