SMILES Transformer

Toxisity prediction using latent representation of SMILES.

Requirement

This project requires the following libraries.

• NumPy
• Pandas
• PyTorch > 1.2
• tqdm
• RDKit

Dataset

Canonical SMILES of 1.7 million molecules that have no more than 100 characters from Chembl24 dataset were used.
These canonical SMILES were transformed randomly every epoch with SMILES-enumeration by E. J. Bjerrum.

Toxicity data from https://www.kaggle.com/datasets/fanconic/smiles-toxicity were used.

Pre-training

Pre-trained model is here.

Downstream Tasks

See experiments/ for the example codes.