Suggestion: jitter & shuffle defaults FeatureScatter

[samuel-marsh]

Hi Seurat Team,

Just wanted to pop in potential suggestion for change in defaults on FeatureScatter based on recent issue #5875 where the jitter default can give false impression of data. In looking at this I started wondering whether changing the default behavior from jitter = TRUE and shuffle = FALSE might actually be more informative for the user. I'll describe my rationale and leave examples below and then leave it to your team. As I say just potential suggestion and I likely do not have the full rationale for the design choices so feel free to reject the PR.

In most cases on truly crowded plot the jitter being applied doesn't (at least to my eye) significantly alter the visualization. However, setting shuffle = TRUE does significantly alter things by reducing identity layering effects which can mask cell/identity location. I do concede that the interpretability of the shuffled plot is also difficult due to sheer number of points but overall I think the effect is better than shuffle = FALSE.

Using the SeuratData hcabm40K as moderately (compared to 100K+ cells at least) crowded dataset. I also cropped axes of plots below to increase resolution to see an effect of these parameters as raw dataset has some outliers.

Thanks again for everything you do!

Best,
Sam

library(tidyverse)
library(Seurat)
library(scCustomize)
library(patchwork)

hcabm40k <- hcabm40k.SeuratData::hcabm40k

hcabm40k <- Add_Mito_Ribo_Seurat(seurat_object = hcabm40k, species = "human")

p1 <- FeatureScatter(object = hcabm40k, cols = DiscretePalette(palette = "polychrome", n = 36), feature1 = "nCount_RNA", feature2 = "percent_mito", pt.size = 0.75, jitter = T) + NoLegend() + xlim(0, 75000) + ylim(0, 10)

p2 <- FeatureScatter(object = hcabm40k, cols = DiscretePalette(palette = "polychrome", n = 36), feature1 = "nCount_RNA", feature2 = "percent_mito", pt.size = 0.75, jitter = F) + NoLegend() + xlim(0, 75000) + ylim(0, 10)

wrap_plots(p1, p2)

p3 <- FeatureScatter(object = hcabm40k, cols = DiscretePalette(palette = "polychrome", n = 36), feature1 = "nCount_RNA", feature2 = "percent_mito", pt.size = 0.75, shuffle = F) + NoLegend() + xlim(0, 75000) + ylim(0, 10)

p4 <- FeatureScatter(object = hcabm40k, cols = DiscretePalette(palette = "polychrome", n = 36), feature1 = "nCount_RNA", feature2 = "percent_mito", pt.size = 0.75, shuffle = T) + NoLegend() + xlim(0, 75000) + ylim(0, 10)

wrap_plots(p3, p4)

[mojaveazure]

Hi Sam,

Thanks for the PR! We like setting jitter = FALSE by default, but we think we should keep shuffle = FALSE by default as well. Can you make that change? Then, we can get this merged in

[samuel-marsh]

Absolutely sounds good! Will update and push Monday AM.

Best,
Sam

[samuel-marsh]

Hi @mojaveazure

Just updated the PR switching shuffle to FALSE as well.

Best,
Sam

[saketkc]

Thanks @samuel-marsh!

[samuel-marsh]

No problem! Thank you guys too!