Telling apart control and treatment groups in Seurat Visualizations

Question

[Aaron]

Hi,

I saw on one of the Seurat data visualization tutorials that if you have a dataset you generated from an experiment, you can split a dataset into the control and the treatment. For example, if you have the following dataset where the metadata is clearly split into groups, you can create visualizations for each group like this:

FeaturePlot(pbmc3k.final, features = c("MS4A1", "CD79A"), split.by = "groups")

I'm having trouble finding how my dataset is split into groups - there doesn't appear to be a feature of the metadata that indicates whether the data is from the control or the treatment group, but the the Seurat data analysis functions seems to recognize which cells they were collected from. When I run this function:

"cluster1.markers <- FindMarkers(seurat.obj, ident.1 = 1, min.pct = 0.1, test.use = "roc", logfc.threshold=0.1)
head(cluster1.markers, n = 1000)"

Seurat returns values for both "pct.1" and "pct.2," which leads me to believe that it can tell apart cells in the control group from cells in the treatment group. (Although the "avg_logFC" for all entries is "-Inf," which I'm not sure if it's concerning.) I've looked in the output of "[e-mail]" but haven't found any thing that might seem to indicate control or treatment.

In a dataset without clearly labelled control and treatment groups, I'd like to know how I can tell apart the control and treatment group?

EDIT: "orig.ident" only has one value

Answer 1

[rpolicastro]

In your FindMarkers command pct.1 represents the percentage of cells in cluster 1 that have at least one read for that feature, and pct.2 represents the percentage of all other cells that have at least one read for that feature.

If there is only one orig.ident in the metadata and no other columns specifying sample type then the Seurat object likely only contains one sample.

Answer 2

[swbarnes2]

which leads me to believe that it can tell apart cells in the control group from cells in the treatment group.

Are you absolutely sure that you experiment has a control and treated? If you don't specify a group, it's likely comparing between the calculated clusters.

Answer 3

[seidel]

In a dataset without clearly labelled control and treatment groups, I'd like to know how I can tell apart the control and treatment group?

You can't. If the data are not labelled, knowing which came from control or treatment is a matter of either guesswork, extrapolation, or imputation. As @swbarnes mentions, pct.1 and pct.2 have nothing to do with treatment vs control. Treatment and Control are specific experimental terms describing an experimental parameter applied to cells. The experiment either consists of unique populations of cells exposed to some treatment or control, of which each would generate a dataset, that could be combined at some point, but this would be indicated, or it doesn't. Alternatively, single datasets can be split into groups with some property - but that property would be some characteristic of the data itself.