Home

Awesome

molecule_structure_generation

Target specific de novo drug design using Transformer neural network.

Folders:

/data The folder contains datasets used for training and evaluation of the model. To obtain these datasets we downoaded full version of BindingDB and ran BindingDB_dataset_preparation.ipynb (see scripts). Datasets contain only human proteins. Vocabulary file was used to encode data.

/data_4_organisms The folder contains extended datasets with proteins from human, bovine, rat and mouse and corresponding vocabulary file.

/scripts The folder contains IPython notebooks with raw data preparation, model training and decoding from it, model evaluation and results visualization.