This is an educational codebase for de novo molecule design with LSTM neural networks. The molecules are represented in SMILES strings and fed to the LSTM character-by-character. The LSTM is then trained to predict the next element in the sequence.

To train an LSTM, you need to install the following dependencies:

python==3.10.6
tensorflow==2.7.1
keras==2.7.0

In turn, you can run the standalone smiles_lstm.py file to train an LSTM model using the SMILES strings in data/train.txt. The code will save 256 SMILES designs into designs.txt.

Happy designing!