Seurat: Merge clusters

[kathirij]

Hello,
I am analyzing 10x dataset. I believe I have almost decided the correct resolution that correctly identifies unique populations in the my dataset (~3000 cells; 12 clusters; resolution 1.0). However, there are two sets of two clusters each that are similar to each other and I would like to merge them into single clusters. Decreasing resolution doesn't work because I loose some of the other interesting clusters.

I was hoping that there is a way to merge the clusters back instead of using lower resolution and subsetting data later on.

[amisharin]

Ok, I am not a member of Satija lab, but I had the same question. One approach is described in comments to pbmc33k dataset: https://www.dropbox.com/s/y6kwho066vugjrg/pbmc33k.R?dl=1 (or find the link on the "get started" page http://satijalab.org/seurat/get_started.html) via initial overclustering and then merging nodes of the cluster tree.
Also, in the main R code: https://github.com/satijalab/seurat/blob/master/R/seurat.R#L2422 (V1.4):
"Rename one identity class to another
Can also be used to join identity classes together (for example, to merge clusters)."
I have not tried the later function, let me know if you get it to work.

[kathirij]

Thanks @amisharin

The first approach you suggested seems pretty useful... I tried that and it worked in terms of merging clusters. Now, I have to play around with what biologically makes sense... Thanks!

[laijen000]

Hi! I am using Seurat v3 and after clustering and predicting likely cell identities, I want to merge some redundant clusters back together. Is there a way to do this in Seurat v3? The links above no longer seem to work. Thank you!