Feature Request: pseudoBulkDGE for Non-Integer Batch-corrected Data

[DarioS]

Some algorithms, such as RUV-III-NB, take integer counts as input and also output integer counts in gene space whereas others, such as scMerge, will produce logarithmic scale decimal numbers in gene space. May pseudoBulkDGE support this via a moderated t-test?

[LTLA]

Hm. While technically possible, I don't think highly of performing DE directly on batch-corrected values, either in bulk or in single-cell settings. I daresay it's even worse in single-cell as most of the popular batch correction algorithms are rather... aggressive. Nonetheless, if you have a good reason to do this, I'm happy to consider a PR.

[DarioS]

Ah, O.K. Perhaps the guardrail is beneficial. This leads me to wonder if scater has a saturation plot for cancer studies.