smiles-autoencoder

LSTM-based autoencoders for SMILES strings

Installation

$ pip install smiles-autoencoder

or

$ git clone https://gitlab.com/tjkessler/smiles-autoencoder
$ cd smiles-autoencoder
$ pip install .

Usage

One-hot encoding

from smiles_autoencoder.encoding import SmilesEncoder


smiles: List[str] = [...]

encoder = SmilesEncoder()
encoder.fit(smiles)

encoded_smiles: numpy.ndarray = encoder.encode_many(smiles)
# encoded_smiles.shape == (n_smiles_strings, sequence_length, n_unique_characters)

Autoencoding

import torch
import torch.nn as nn

from smiles_autoencoder.model import LSTMAutoencoder


encoded_smiles = torch.tensor(encoded_smiles, dtype=torch.float32)

autoencoder = LSTMAutoencoder(
    input_size=encoded_smiles.shape[2],
    hidden_size=64,
    latent_size=12,
    num_lstm_layers=1
)

opt = torch.optim.Adam(autoencoder.parameters(), lr=0.001)
loss_crit = nn.L1Loss(reduction="sum")

for epoch in range(8):

    for enc_smiles in encoded_smiles:

        opt.zero_grad()
        pred = autoencoder(enc_smiles)
        loss = loss_crit(pred, enc_smiles)
        loss.backward()
        opt.step()

Decoding predictions

pred_smiles: torch.Tensor = autoencoder(encoded_smiles[0])
pred_smiles: str = encoder.decode(torch.round(pred_smiles).detach().numpy().astype(int))

Contributing, Reporting Issues and Other Support:

To contribute to smiles-autoencoder, make a pull request. Contributions should include tests for new features added, as well as extensive documentation.

To report problems with the software or feature requests, file an issue. When reporting problems, include information such as error messages, your OS/environment and Python version.

For additional support/questions, contact Travis Kessler ([email protected]).