Applying CellTypist to scVI batch-corrected data

[mark0428]

Hi,

I'm wondering whether CellTypist can be applied directly to scVI batch-corrected data.

By default, the normalized output from scVI uses library_size=1, meaning all expression values are scaled between 0 and 1. However, CellTypist expects log1p-normalized expression scaled to 10,000 counts per cell, and otherwise raises the error:

ValueError: 🛑 Invalid expression matrix in both .X and .raw.X, expect log1p normalized expression to 10000 counts per cell

To address this, I can set library_size=1e4 when extracting normalized expression from scVI. My question is whether it is reliable to apply CellTypist to such rescaled scVI outputs, or if CellTypist should only be used with conventionally normalized (non-scVI) data.

Thanks in advance!

[mark0428]

I'm also wondering whether data that have been transformed to zero mean and unit variance per gene are applicable to CellTypist. Thanks!