search

mims-harvard / TxGNN

TxGNN: Zero-shot prediction of therapeutic use with geometric deep learning and clinician centered design
https://zitniklab.hms.harvard.edu/projects/TxGNN
MIT License
148 stars 33 forks source link

Creating a DGL graph with `create_dgl_graph` function #5

Open pgniewko opened 10 months ago

pgniewko commented 10 months ago

I've been looking into the code, and this line seems a bit puzzling (numerical experiments don't show any difference in the model's performance w/ and w/o this line). What is so special about the 'effect/phenotype' and why does it need to be set to 0? Isn't it replaced by a positive number anyways in the succeeding for loop? Are there any edge cases where even though the 'effect/phenotype' is not in the df, we still want to keep this node type in the graph?

l4b4r4b4b4 commented 1 month ago

exactly wondering about that Same loop. I did parallelize it, but it hangs at effect/phenotype :thinking: