DotPlot avg. expression & scaling: Do calls to FetchData() retrieve @scale.data on SCT and RNA assays?

[adRn-s]

I'm working on a dataset where all cells have the expression of a given marker. I want to show this, per cluster, on a DotPlot; but leaving the default scale=TRUE argument gives me negative values in 2 out of 8 clusters. If I understand correctly, this means the gene has lower than average expression in these, when compared to the remaining clusters. Yet, I want to avoid the possible confusion for collaborators, reviewers, and ultimately, readers of the paper.

I'm not scaling the values (switching to FALSE). This is clear to me after some trials, I want the plot to be clear. Now, I've read the source code to get a grasp of what it means... I was afraid that turning this off would instead get me the geometric mean.

Since FetchData() has no layer argument defined, that's going to be @scale.data ... Right? Please confirm. And, looking to what DotPlot does, these are then potentiated (expm1), arithmetic mean is calculated, and log1p-transformed. Which is fine, but it took me a while to digest.

For the record, some grievances I had, were:

• Shouldn't DotPlot call AverageExpression? If that were the case, I'd like to switch the layer there to have maximum control on what I'm plotting (e.g. via ellipsis, ...). By default, the function assumes data values need exponentiation (cfr. details at the reference manual)... anyway, averaging is not done by using this function and I found that to be confusing at first, so I wanted to complain :P bla bla DRY principles.

• I'm using RNA assay, but the same observation applies to SCT, which is what I'm actually going to use across the whole pipeline for finding markers and identification of clusters' cell type. But for this plot, I had set off scaling and switched over to RNA. Else, I'd get zeroes (which, to my situation, is as bad as having the negative values).

Any advice will be appreciated. But mainly, I need to confirm the SeuratObj::FetchData behaviour. Thanks in advance!

[adRn-s]

I copied over the source code and adjusted to replace base R scale() with my own MinMax function, mimicking what scanpy does (although iirc they forgot to divide by max-min so both libraries are wrong under my perspective)