search

mperozek11 / scSHARP_tool

Repository for TSP 2022
MIT License
1 stars 0 forks source link

Train GCN with an external consensus prediction #7

Open tvegawaichman opened 4 months ago

tvegawaichman commented 4 months ago

Hello! I'm interested in applying the GCN approach for label propagation for non-consensus prediction but using an external method to calculate the consensus. I would like to know if it is possible to run the GCN using the query matrix and the predictions array as inputs.

Thanks!

Tomas

lewinsohndp commented 2 months ago

Hello! To clarify, you are looking to run the GCN for label propagation, but with your own cell type label predictions? If so, you can do this by setting the preds_path argument when initiating the scSHARP model. In this case, scSHARP expects a csv with an index corresponding to cell id and header corresponding to the name of the cell type annotation tool. See an example here: https://github.com/mperozek11/scSHARP_tool/blob/main/simulations/splat_0.7/preds.csv. Let us know if you have any more questions!

tvegawaichman commented 2 months ago

Hello! Thanks for your reply. I have my own prediction tools and an additional column with Consensus calls but use a different method for handling "No consensus" calls, so I want to use that Consensus column as input for the GCN propagation instead of the calling method used in scSHARP.

lewinsohndp commented 2 months ago

Ok. Currently there is no way to do this, but I am happy to add this functionality. Would inputting an array corresponding to each cells label (either a confident cell type, or unconfident call) work for you?

tvegawaichman commented 2 months ago

Yes, that would be amazing! Thank you very much.