getMarkerFeatures() all groups vs specific group

[wuv21]

Hi! Just a conceptual question regarding getMarkerFeatures() (and to make sure I'm understanding this function correctly):

I understand that getMarkerFeatures() as used below will return a list of markers that are unique to each group in the groupBy column data. If I understand correctly, the background group is set to all other cells not in the current group being examined.

getMarkerFeatures(proj, groupBy = "myGroup")

However, if I run the below code instead - should I expect the same list of significant markers as compared to markers that are specific to Group_Level_1 from the above code? When I try it out on my data, I don't see the same list of significant markers. Any clarification on this would be greatly appreciated!

getMarkerFeatures(proj, groupBy = "myGroup", useGroups = "Group_Level_1")

Thanks!

[rcorces]

Have you used set.seed() to set a constant seed? There are randomized operations in this function.

[wuv21]

My debugger is acting up right now, but I think something in this chunk is causing the discrepancy.

ArchR/R/MarkerFeatures.R

Line 641 in 21099d6

matchList <- lapply(seq_along(useGroups), function(x){

[rcorces]

interesting. I am transferring this to an issue and will look into it.

[rcorces]

Moved to #1467

[anastasiya-pendragon]

Hello there! Just wanted to give you an idea of why it works for the tutorial data and not in your dataset.

EDIT: just now saw that you already tried it with a different cluster.
So only going to leave this reference:
You can read my pull request here - about markerFeatures potential bug, there I explain where this difference comes from and how to potentially fix this. TLDR: it has to do with the random function that is used for the background cell selection.

ArchR/R/MarkerFeatures.R

Line 675 in 21099d6

sx <- sample(seq_len(nrow(knnx)), nrow(knnx))

The code in this line will have different random selections, depending on how many times it's called in a method. I.e. when you run your C10 cluster separately it's a different random seed than when you run it as the 10th cluster when you do not specify the usegroups.

Hope it helps or at least clarifies some things!
Best.
Anastasiya

[wuv21]

Hello there! Just wanted to give you an idea of why it works for the tutorial data and not in your dataset.

EDIT: just now saw that you already tried it with a different cluster. So only going to leave this reference: You can read my pull request here - about markerFeatures potential bug, there I explain where this difference comes from and how to potentially fix this. TLDR: it has to do with the random function that is used for the background cell selection.

ArchR/R/MarkerFeatures.R

Line 675 in 21099d6

sx <- sample(seq_len(nrow(knnx)), nrow(knnx))

The code in this line will have different random selections, depending on how many times it's called in a method. I.e. when you run your C10 cluster separately it's a different random seed than when you run it as the 10th cluster when you do not specify the usegroups.

Hope it helps or at least clarifies some things! Best. Anastasiya

Thanks @anastasiya-pendragon! Very helpful - will try it out and tag you in the issue that Ryan opened for this.