Several questions

[KevinMaroney]

Hi Aleksandr,

First of all I want to say I think you're doing crazy important stuff. I've been suggested to use SingleR, but as I've found it's not accurate ~1/3-1/2 of the time, especially with my cell type which could include NKT-like aB T cells that are relatively new and not well-described. SingleR is also based off of bulk RNA-seq sorted data which, as you stated in your paper, is probably not the best reference for scRNA-seq. But I wanted to try your automated classification method (as opposed to just using my CiteSeq memory and cell type markers I selected to assign based on high vs low quantiles and was halted by a few problems.

First, I just used your basic code you have at the start:

source("https://raw.githubusercontent.com/IanevskiAleksandr/sc-type/master/R/gene_sets_prepare.R"); source("https://raw.githubusercontent.com/IanevskiAleksandr/sc-type/master/R/sctype_score_.R")

    #Download scType curated references
    gs_list = gene_sets_prepare("https://raw.githubusercontent.com/IanevskiAleksandr/sc-type/master/ScTypeDB_short.xlsx", "Immune system")
    es.max = sctype_score(scRNAseqData = [my Seurat object dataset], scaled = TRUE, gs = gs_list$gs_positive, gs2 = gs_list$gs_negative)

This is the error I got, which indicates to me that there's been some change in Seurat object convention that doesn't allow conversion of the expression data for each cell to matrix format (or I'm doing something else stupid): Warning: Renaming features in v3/v4 assays is not supported Error in Z[cell_markers_genesscore[jj, "gene"], ] * cell_markers_genes_score[jj, : non-numeric argument to binary operator In addition: Warning message: In sctype_score(scRNAseqData = Sig_Increase_ClonotypePair_Seurat, : scRNAseqData doesn't seem to be a matrix

Second, when looking at your gene list (scTypeDB_short.xslx or _long.xslx) I saw that it only goes down to the level of NKT-like CD8, memory CD8, or effector CD8, but not terminal effector, effector memory, etc. Are you planning to expand this at some point? I'm doing the best with SingleR and my own metadata annotation that I can, but was really excited by your tool, as most people just filter out anything they don't like based on SingleR, but SingleR isn't often very accurate anyway because of its reliance on bulk RNA data so I believe this is bad science.

Thank you for your work. Hoping I'm making some stupid mistake or missing something obvious.

[IanevskiAleksandr]

Dear @SomeGuy3865,

The issue arises due to the update from Seurat version 4 to Seurat 5. We plan to update our code next week to ensure compatibility with both versions.

We are indeed developing ScType 2.0, with an anticipated release later this year. This update will expand the range of cell types and tissues covered and include improved marker genes for enhanced accuracy.

Best regards, Aleksandr

[IanevskiAleksandr]

Dear @SomeGuy3865 ,

We have now updated scType to support Seurat v5.

BR, Aleksandr