sentence mode

[biopzhang]

Hello GenePT developers,

Thank you for sharing the exciting work!

I'm curious about how to use your model. I wonder how each gene is tokenized in the model. To the best of my knowledge, GPT likely tokenizes many gene symbols, e.g. COL1A1, into a couple of subwords. I guess it does not really matters for tasks like cell type annotation or clustering. However, molecular-level investigation may suffer, for example, ligand-receptor. Could you please share your oppinion?

Thank you!
Pengzhi