Hi Daria. Thanks for sharing the code with the drug design community!
In your associated article from Scientific Reports, there are a few ROC-AUC curves provided for evaluating the discriminating ability of SMINA docking software.
I had a few questions regarding how the thresholding works and what are the class labels for the datasets compared in the figures. It would be really useful if you can share the scripts for the curve calculation from docking scores or if you can explain how the curves are plotted from the docking scores along with the associated hypotheses.

Hello, I read your paper and want to check your model structures.

However, the model_training_and_decoding.ipynb file returns "reach display cannot be rendered" error message.

Could you check the file please?

I tried to replicate the code, but with the current versions of the used libraries, there are a lot of incompatibility issues.
Could you please fix the version of the dependencies in the code, such that it is possible to run it.

Eg.
"pip install tensor2tensor==1.15.7" instead of "pip install tensor2tensor"

Hello @dariagrechishnikova, in data_4_organisms , the number of unique train proteins in dataset 1, i.e., train_4_org_1042_104/proteins_train_4_org_corrected , is 1514 and, the number of unique test proteins in dataset 1, i.e., test_4_org_1042_104/proteins_test_4_org_corrected , is 99, instead of 1042 and 104 respectively. Requesting for the correct dataset. Thanks!