Masked Language Models are Fragment Based Drug Designers

Download, Process and Train FragmentBERT

bash run.sh

Steps to reproduce :

• fragment_tokenizer/tokenizer.py Defines Fragment Augmentation, Encoder and Decoder.
• moses_train_test_split.py Takes in moses_dataset.csv and produces train, test moses data.
• encode_moses.py Produces failed and augmented encoded moses smiles files.
• model/tokenizer.py Creates FragmentBERT Tokenizer.
• model.py Train the model and log to WandB.

Conda

conda env create -f environment.yml
conda activate fbdd