Use model for making predictions for all drug combinations +protein info

[anisioti]

Hello,

I want to use the trained models to make predictions for all the possible drug combinations and the side effects, exactly as described also in the Decagon paper. From the files that I see on github this is not straightforward for me. I was wondering if you could provide some instructions on how this can be done.

Moreover, since I am interested in seeing the predicted side effects of the drug combinations of specific drugs and some of them don't exist in the dataset I wanted to add some of these drugs to the dataset and re-train the models with them.
To my understanding this is needed in order to make it possible to have predictions for these drugs (I think that we need at least some drug-protein information for them in order to make them part of the multimodal graph). My question is the following, is there a file where we can find the correspondance of protein and protein names, or some kind of protein info like the id in a database? Do you have any other suggestion on how to add new drugs to the dataset?

Thank you in advance,
Athina