Normalisation of reference dataset for scRNA cell type annotation

Question

[user31888]

I have scRNA data in a form of a SingleCellExperiment object that I previously filtered, normalised, clustered, with genes selected.

I am trying to assign a cell type to each of my cells in this SingleCellExperiment object using scmap (here) and the Human Primary Cell Atlas (HPCA) as a reference (taken from the SingleR package).

Following the scmap vignette, I created a SingleCellExperiment object out of the HPCA dataset, selected features and created indexes.

I am now ready to project my scRNA data into these indexes.

However, is it correct to project scRNA read counts (or CPM) on HPCA reference indexes knowing that HPCA came from affymetrix data normalized differently (HPCA values are robust multi-array average (RMA) expression measure)?

[GouthamAtla]

I would be a bit skeptical to annotate single cell data using bulk RNA data generated from primary cells that too by using arrays. It may not work. Why don't you use marker genes to annotate the cells ?

[user31888]

Do you mean using cell type signatures and scoring their enrichment with GSEA or gsva for example?

[GouthamAtla]

by using expression of well known marker genes per cell type.

[user31888]

I came across this recent database, CellMarker. Do you think a simple GSVA using this database could be accurate? I need to get a score of some sort to be able to discriminate cell types.