AUCell comparability for integrated data

[ashenflower]

Discussed in #476

^{Originally posted by ashenflower January 21, 2025}
Hello everyone,

I’m new to pySCENIC and single-cell analysis, so I apologize in advance if my question seems naive. I ran pySCENIC on a matrix that comes from the integration of data from multiple datasets, with rows representing different cells, columns representing different genes, and each cell contains the raw counts . My data are grouped by cell type, and I aim to identify regulons specific to certain cell types.

To do this, I’ve been using the RSS metric and Z-scores calculated column-wise (i.e., for each regulon). However, I now have a concern: is this approach correct? Both RSS and Z-scores are based on the AUC matrix, but I’m worried that my results might not be valid because the pySCENIC paper states:

"The metric used by AUCell to quantify the activity of a predicted regulon in individual cells of an experiment is an unnormalized enrichment score. Therefore, it can only be used to compare the activity of a regulon across cells of the same experiment."

My question is: does "same experiment" refer to the same pySCENIC run, or does it mean data from the same dataset?

I know there is a similar issue, but it looks like a different case (i.e. comparing statistics made on results obtained on different data on indipendent runs of pySCENIC)

Thank you in advance for your help!