lianghsun / drug_discovery

This project uses a Variational Autoencoder (VAE) to generate SMILES strings for novel compound generation. The VAE model is trained on a dataset of existing chemical compounds and can generate new, valid SMILES strings, which may represent potentially new and useful chemical entities.

Prerequisites

• Install the required packages by running the following command:

  pip install -r requirements.txt

Dataset

• Download the SMILES dataset from Kaggle ZINC 250k. Change the file extension to .smi and remove the header row.
• Place the prepared .smi file in the datasets folder.

Preprocessing

• Use preprocess.ipynb and run the notebook to preprocess the .smi file and obtain the tokenization mapping table.

Running the Main Script

We have provided a token file for this project, so you can skip the Load Training Data section and proceed with running the rest of the code.

Note: Due to the nature of Variational Autoencoders (VAE), there might be instances where new compounds are not generated (sampling problem). If this happens, please run the code multiple times to obtain a valid compound.

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.