Update ingest tutorial - Githubissues

View / edit / reply to this conversation on ReviewNB

ilan-gold commented on 2024-10-14T13:51:58Z ----------------------------------------------------------------

Would take a guess that the batch numbers are not right anymore or something as to why the mast is dropped below?

flying-sheep commented on 2024-10-15T08:33:41Z ----------------------------------------------------------------

I checked and nope, they stay the same, as they are manually assigned.

Seems like ingest doesn’t transfer the cell type in batch “2” anymore.

The only thing that changed in the notebook is using anndata.concat instead of anndata.AnnData.concatenate, but that’s not it.

review-notebook-app[bot] commented 1 month ago

View / edit / reply to this conversation on ReviewNB

ilan-gold commented on 2024-10-14T13:51:59Z ----------------------------------------------------------------

Are we ok with what's happening here? It looks like we lost a cell type and also some were re-assigned

flying-sheep commented on 2024-10-14T14:10:22Z ----------------------------------------------------------------

thanks, I saw that some numbers changed but totally missed that one cell type vanished!

flying-sheep commented 1 month ago

thanks, I saw that some numbers changed but totally missed that one cell type vanished!

View entire conversation on ReviewNB

flying-sheep commented 1 month ago

I checked and nope, they stay the same, as they are manually assigned.

Seems like ingest doesn’t transfer the cell type in batch “2” anymore.

The only thing that changed in the notebook is using anndata.concat instead of anndata.AnnData.concatenate, but that’s not it.

View entire conversation on ReviewNB

ilan-gold commented 1 month ago

Seems like ingest doesn’t transfer the cell type in batch “2” anymore.

Do we want to figure out why or no?

flying-sheep commented 1 month ago

I’m assuming it’s because of dependency changes or chance. The umaps are a bit different, idk

Before, 6/7 cells annotated as “mast” are mapped to “alpha” (1 to “mast”), now 7/7 are mapped to “alpha” :thinking:

ilan-gold commented 1 month ago

chance

If it's this we should probably set seeds then?

flying-sheep commented 1 month ago

I think we should either figure out how to reliably capture that cell type or decide we don’t care. I think that if setting a seed qualitatively changes the result, it doesn’t count as “reliable”.

Maybe “almost completely not captured” vs “completely not captured” is a sign that capturing this cell type isn’t stable, but maybe there’s a way to improve the amount of reliably identified mast cells.

ilan-gold commented 1 month ago

Maybe “almost completely not captured” vs “completely not captured” is a sign that capturing this cell type isn’t stable, but maybe there’s a way to improve the amount of reliably identified mast cells.

I would definitely agree with this sentiment. Not sure what to do about this PR specifically. I leave it up to you. I would say if "nothing" is the answer, maybe we could put a warning in the notebook about small cell clusters. Really not sure, though. I went through this issue with the leiden upgrade and eventually either landed on "let the change through because there's nothing we can do" or "found some dependency issue"

flying-sheep commented 1 month ago

Maybe we can wait for @Koncopd to come back from holiday and get a comment, I think he wrote the ingest code originally.

ilan-gold commented 1 month ago

@flying-sheep Sounds good to me

flying-sheep commented 3 weeks ago

@Koncopd could you please take a look?

flying-sheep commented 2 weeks ago

apparently not

scverse / scanpy-tutorials

Update ingest tutorial #139