Seurat workflows and compatibility with targeted gene panels

[jhb1980]

Hi Team (cc community)!

With BD and 10X Genomics offering off-the-shelf gene targeting panels for scRNA analysis (typically ~1,000 genes), I was wondering if you could comment / have experience with the suitability of the "default" Seurat framework (normalization / SCTransform, PCA, clustering etc) for working with these datasets...?

Edit: see for example the public 10X neuroscience panel data on their support site. They detect a little over 1,100 (targeted) genes, but the median per-cell sits at 310 genes, presumably reflecting the differences in cell types. I'm not sure how many "housekeeping" genes are included in these panels...

Thanks!

[saketkc]

Hi @jhb1980,

Thanks for your question and thanks for bringing our attention to the newly released data! We will probably have a detailed vignette on these type of datasets soon, but I did some quick analysis below.

There are indeed some house keeping genes in the panel (and they exhibit the usual technical noise):

SCTransform's variance stabilization works:

With most parameters as defaults, the UMAP:

object <- SCTransform(object = object, method="glmGamPoi",  variable.features.n=500, verbose = FALSE)
object <- RunPCA(object = object verbose = FALSE)
object <- RunUMAP(object = object dims = 1:20, verbose = FALSE)
object <- FindNeighbors(object = object, dims = 1:20, verbose = FALSE)
object <- FindClusters(object = object, verbose = FALSE)