FeaturePlot() with split.by does not seem to have the same scaling for all subplots

[abayegan]

# insert reproducible example here

[abayegan]

I believe the scales in different feature subplots with split.by are not the same which might be misleading especially that the scales are hidden by default. For instance running
p <- FeaturePlot(obj,features,reduction = 'umap',split.by = "condition")
gives the following plot for my data:

but running:
for(i in 1:3){p[[i]] <- p[[i]]+theme_void()} p
yields:

As you can see the scale of the middle plot is from (0,125) while the other two are from (0-80) so the colors are not comparable. This is not an extreme case but still not accurate.

[samuel-marsh]

Hi,

Not member of dev team but hopefully can be helpful. What version of Seurat are you using? This issue was addressed in release 3.2.3 (https://github.com/satijalab/seurat/blob/master/NEWS.md#seurat-323---2020-12-14).

Best,
Sam

[abayegan]

Hi,

Not member of dev team but hopefully can be helpful. What version of Seurat are you using? This issue was addressed in release 3.2.3 (https://github.com/satijalab/seurat/blob/master/NEWS.md#seurat-323---2020-12-14).

Best, Sam

Thank you @samuel-marsh! I was not aware of the keep.scale parameter.

[abayegan]

Setting keep.scale="all" unifies the scales in my specific example with only one feature

[samuel-marsh]

Are you plotting more than one gene/feature? The default should keep scale the same within feature when split. Also be advised that setting to all may result in suboptimal scales when plotting multiple features

[abayegan]

Are you plotting more than one gene/feature? The default should keep scale the same within feature when split. Also be advised that setting to all may result in suboptimal scales when plotting multiple features

I am using Seurat 4.0.3 . That does not seem to be the case. By default the splits have different scales even for the same feature as you can see in my example. It should not be the case according to the documentation here.

As you mentioned setting keep.scale="all" does not seem reasonable to be applied to more than one feature.

[samuel-marsh]

hmmm actually you're right... Something must have gotten lost in recent versions. Hopefully devs can chime in here.

Reproducible example below.

pbmc <- pbmc3k.SeuratData::pbmc3k.final
pbmc <- UpdateSeuratObject(pbmc)

[email protected]$test[[email protected]$seurat_annotations == "B"] <- "Group1"
[email protected]$test[[email protected]$seurat_annotations == "NK"] <- "Group2"

pbmc_subset <- subset(x = pbmc, idents = c("B", "NK"))

FeaturePlot(object = pbmc_subset, features = "MS4A1")
FeaturePlot(object = pbmc_subset, features = "MS4A1", split.by = "test") & theme(legend.position = "right")

*Edit: @andrewwbutler tagging you since you helped with original PR. Thanks!!

[abayegan]

Thanks for generating this example!

[olneykimberly]

so what is the solution?

[samuel-marsh]

Hi @olneykimberly for now (or for future) you can use FeaturePlot_scCustom from scCustomize package. See this vignette: https://samuel-marsh.github.io/scCustomize/articles/Gene_Expression_Plotting.html for more info. When using split.by it will preserve scale across plots.

Best,
Sam

[samuel-marsh]

For future reference this issue was also fixed within Seurat with 4.1 release.

https://github.com/satijalab/seurat/blob/master/NEWS.md#seurat-410-2022-01-14

[ShanshuoZHU]

Hi @olneykimberly for now (or for future) you can use FeaturePlot_scCustom from scCustomize package. See this vignette: https://samuel-marsh.github.io/scCustomize/articles/Gene_Expression_Plotting.html for more info. When using split.by it will preserve scale across plots.

Best, Sam

Hi @samuel-marsh , now I am trying to use FeaturePlot_scCustom and I am usingsplit.by and only have 1 feature. I find that when set 'na_cutoff = NA', the scale is not kept the same across plots for some features and all the scale legends will be displayed. but if donot set 'na_cutoff = NA', the scale will be preserved and only 1 scale legend displayed. why the na_cutoff setting affects the scale/legend and how can I avoid that? Thanks a lot.