FindConservedMarkers vs FindMarkers vs FindAllMarkers Seurat

Question

[Payal]

Can someone please explain the difference between FindConservedMarkers and FindMarkers.

As per definition:

FindConservedMarkers- Finds markers that are conserved between the groups

1. But does that mean that the genes are similarly expressed between groups/conditions or genes are differentially expressed between groups/conditions?
2. Also is this supposed to be for all clusters or a single cluster? If all clusters, according to the function we are identifying them for one cluster (ident.1) or max two clusters (ident.2) at a time right? Then, what ? Do I look for similar genes for all the clusters?

FindConservedMarkers(object, ident.1, ident.2 = NULL, grouping.var,
    assay = "RNA", slot = "data", meta.method = minimump,
    verbose = TRUE, ...)

FindMarkers - Finds markers (differentially expressed genes) for identity classes

I don't understand what's the significance of this?

FindAllMarkers - Finds markers (differentially expressed genes) for each of the identity classes in a dataset

What is this doing and how is this different from FindConserved Markers?

It would be very helpful if someone can explain this with a biological use cases.

Thanks in advance!
Payal

Answer 1

[jared.andrews07]

FindMarkers will find markers between two different identity groups - you have to specify both identity groups. This is useful for comparing the differences between two specific groups.

FindAllMarkers will find markers differentially expressed in each identity group by comparing it to all of the others - you don't have to manually define anything. Note that markers may bleed over between closely-related groups - they are not forced to be specific to only one group. This is what most people use (and likely what you want).

FindConservedMarkers will find markers that are conserved between two groups - this can be useful if you want to find markers that are conserved between a treated and untreated condition for a specific cell type or group of cells. It means they are differentially expressed compared to other groups, but have similar expression between the two groups you're actually comparing.

[Payal]

Thank you! This is helpful.

I have a combined dataset (control vs treatment ). I was following this - https://satijalab.org/seurat/v3.1/immune_alignment.html tutorial. I got confused after the FindConservedMarkers step.

In general on what basis should we select the marker genes to draw the feature plot or identify the clusters? In my case, should I be looking at the genes with low p-value and high log fold change difference between control and treatment for say cluster 1?

[jared.andrews07]

This depends on your experimental setup, research question, and cells. It's difficult to answer without more information. If it seems like cell types are clustering together, despite the treatment, then I think your approach would be valid. If not, you may want to assign cell types to your cells and then compare between treatment and control for each given cell type.

[m.sadman.sakib]

Hi Jared! Thanks a lot for this explanation! I am trying to identify differential expression in each cell cluster for control vs treatment. For this, which function will be appropriate? Thanks a lot in advance!

[jared.andrews07]

See their DE vignette. Generally, you'll want to use FindMarkers, though in your case you'll probably want to loop through each cluster and subset to your cell of interest before comparing the conditions. Or combine the cluster/treatment info into a single column in the metadata and use that to specify the two groups.

[camillab.]

Hi jared.andrews07 !

I have a question regarding FindMarker function. I have an integrated dataset (ctrl vs treatment) and I want to find the DEGs per cluster following this tutorial from seurat

so I would do FindAMarkers but I notice differences based on the additional parameter so I am not sure why and maybe you have a clue on it. Basically if I do :

cluster1.markers <- FindMarkers(pbmc, ident.1 = 1, logfc.threshold = 0.25, test.use = "roc", only.pos = FALSE)

I got these 5 top 5 genes (tot 2367 genes):

    myAUC avg_diff power avg_log2FC pct.1 pct.2
Wfdc2   0.989 2.411825 0.978   3.479528 1.000 0.596
Bpifa1  0.986 5.553996 0.972   8.012722 0.976 0.030
Ly6a    0.976 2.895342 0.952   4.177095 0.966 0.068
S100a11 0.970 1.763789 0.940   2.544610 0.997 0.622
Gsto1   0.962 2.613521 0.924   3.770513 0.929 0.012
S100a6  0.961 2.183622 0.922   3.150301 0.997 0.379

But If I do:

cl1 <- FindMarkers(pbmc, ident.1 = 1, grouping.var = "orig.ident", verbose = FALSE) 

with orig.ident identifying control and treatment post integration I get :

     p_val avg_log2FC pct.1 pct.2 p_val_adj
Aox3          0   1.348781 0.511 0.010         0
Serpinb11     0   2.251726 0.534 0.001         0
Pax1          0   2.030149 0.688 0.003         0
Bpifa1        0   8.012722 0.976 0.030         0
Ecm1          0   2.674723 0.561 0.016         0
Aqp3          0   1.746973 0.513 0.004         0

And if I plot the top 2 genes per analysis (Wfdc2, Bifal and Aox3 and Serpinbl1) I get suggesting that the most DEGs from the 1st analysis is not that specific for clustert1

So which code should I use if I want to find the DEGs per cluster? I would expect that selecting the log2fc threshold and the method (in this case ROC) would give me more specific genes.

[jared.andrews07]

Neither of those are testing for what you think. Take a close look at the FindMarkers examples, which should help you figure out what you need.

If you still run into issues, open a new question, piggybacking on old questions is generally discouraged as it makes information difficult to find.

[camillab.]

Oh sorry, I thought it was a stupid question (eg. something that I have may missed) that's why I didn't open a new question. I have already look into it and obviously I didn't find very explanatory otherwise I would not have bothered you! but I will try again! Thanks!

Answer 2

[Payal]

I found this link. Its very well explained there too - https://hbctraining.github.io/scRNA-seq/lessons/sc_exercises_integ_marker_identification.html