What is the appropriate method for clustering a small number of cells?

0

Entering edit mode

3.8 years ago

david.f.stein ▴ 10

I have a count matrix of single cell data with 122 cells. I am trying to process it with the standard Seurat workflow, however, the UMAP reduction is yielding strange results. Basically, no clusters are forming. I'm wondering if ~100 cells is too low for UMAP. Does anyone know what might be going on here? I can provide my code if necessary

Edit: I found if I raise the "resolution" argument in the "FindClusters" function then clusters are picked up. I'll have to read more about what the "resolution" means in this context.

RNA-Seq next-gen gene genome • 1.6k views

ADD COMMENT • link 3.8 years ago by david.f.stein ▴ 10

0

Entering edit mode

Did you try "pca" instead of "umap" ?

ADD REPLY • link 3.8 years ago by GouthamAtla 12k

0

Entering edit mode

I ran PCA, TSNE, and UMAP. None seem to result in clusters on visualization

ADD REPLY • link 3.8 years ago by david.f.stein ▴ 10

0

Entering edit mode

It might be biological then. Did you check what are the top variable genes using function like rowVars ?

ADD REPLY • link 3.8 years ago by GouthamAtla 12k

0

Entering edit mode

I did, finding variable features seemed to work as usual/expected. Is there something I should be looking for in particular? I just saw on a github issue on the Seurat repository that graphical clustering tends to perform poorly for around <100 cells, so maybe I'm just out of luck

ADD REPLY • link 3.8 years ago by david.f.stein ▴ 10

Login before adding your answer.