Why isn't it possible to run solo when there are categorical_covariate_keys?

[alexlenail]

I have a large experiment in which multiple mice were each separately sequenced in multiple 10X channels. So each mouse has multiple channels. I created by model with:

scvi.model.SCVI.setup_anndata(
    adata,
    batch_key='channel',
    categorical_covariate_keys=["Mouse"],
)

And now I want to remove doublets, but just found out

ValueError: Initializing a SOLO model from SCVI with registered categorical covariates is currently unsupported.

Should I leave out the "Mouse" covariate? I don't understand what I'm supposed to do here.

[ori-kron-wis]

Hey,

Sorry for late response.
Yes, leave out the mouse covariate. Our SOLO model does not support this option now.

When performing batch integration, both the channel ID and the donor ID (or any other relevant source of variation) should ideally be considered as part of the batch correction process, depending on your analysis needs.
For that manner, you can generate a new column in your concatenated adata as the batch key which is a concatenation between "channel" and "Mouse" and use that in the SCVI setup anndata to be later used in SOLO.
e.g:
'donor_1_channel1'
'donor_1_channel2'
'donor_2_channel1'
etc..