Question about normalization

Hi, and thanks for a great software package! I have a couple of questions related to the normalization procedure in scVelo.

• When determining whether the data should be normalized: https://github.com/theislab/scvelo/blob/dea2f64c0eec4566ecfacef928c21dbed77316ec/scvelo/preprocessing/utils.py#L295-L296 the logic is different for dense and sparse matrices (and sometimes gives different conclusions). What is the reason behind this choice?

• I’m using counts estimated by alevin as input to scVelo. These are not necessarily integers, and thus in some cases scVelo concludes that the data is already normalized. My solution was to run the four filtering/normalization steps separately, and enforce the normalization. I was wondering if the enforce argument of normalize_per_cell() perhaps could be propagated to filter_and_normalize(), to allow running all steps in one.

Thanks!

[csoneson]

Thank you! That sounds great. I think it’s hard to detect by just looking at the counts (there could be many or few non-integer values). Perhaps one could imagine something similar to the log argument of filter_and_normalize, allowing the user to specify whether the data has been normalized or not, which would override the internal checks?

[csoneson]

Yes, this seems like a good solution. Thank you!