This project used a vanilla transformer and Graph transformer to predict a molecule as a string of smiles by sampling from a distribution and by providing the disease protein to get the corresponding drugs. Using the decoder only transformer to decode a smiles of molecules to high-dimensional representation. There is also another models that can be used using UI. By uploading the 3D protein structure we can generate the drugs candidate. The candidate can further be optimized for the docking score.
Download the files from this repo then runs the code below
python3 train.py --path 'data/smiles.csv' --epoch=10 --ts 0.1
python3 sample.py
Given the protein structure the model will generate candidate drug.
### Step Two