How to create annotation file to use scSorter tool?

Question

[Chris]

Hi all,

I got stuck at the step annotation file when following this tutorial:

https://cran.r-project.org/web/packages/scSorter/vignettes/scSorter.html

How do we get 2 columns: type and marker when we are trying to identify cell type?

[jv]

pretty sure you have to define your own cell type markers or at least build a table of markers from published/public resources. For instance, in the scSorter paper methods the authors state

"Rosenberg data: This data was created by Rosenberg et al. [20] using the SPLiT-seq protocol to analyze cells from the mouse brain and spinal cord. It profiled 27,096 non-neuronal cells from different cell types: Oligo (4,294 cells), OPC (5,793 cells), Immune (621 cells), Vascular (659 cells), VLMC (1,474 cells), Astrocyte (13,481 cells), Ependyma (518 cells), and OEC (256 cells). We used the cell type definition as well as marker genes defined in the original study [20]."

and

"TM Pancreas data: This data was a mouse atlas created by the Tabula Muris Consortium [21]. The cells were sorted using fluorescence-activated cell sorting (FACS) and sequenced by Smart-seq2 protocol. 1564 cells with valid cell type annotation from a pancreas tissue were used for our analysis. They included cells from Pancreatic A (390 cells), Pancreatic B (449 cells), Pancreatic D (140 cells), Pancreatic PP (73 cells), Pancreatic Acinar (182 cells), Pancreatic Ductal (161 cells), Pancreatic Stellate (49 cells), Endothelial (66 cells), and Immune (54 cells). Marker genes for these cell types were extracted from the original study [21]."

[Chris]

Could you give a simple example for first-time users to know how to do it? It seems scCATCH is easier to use compared with scSorter. How about the cells from humans?

[jv]

Could you give a simple example for first-time users to know how to do it?

I'm not quite what it is that you are asking for an example of. Do you mean an example of how to use scSorter? Or how to define a set of gene markers from other data? If the latter, I don't think a "simple" example is all that easy to provide but a good starting point would be differential expression based marker gene discovery methods like Seurat FindMarkers or what's described in the OSCA documentation.

If you are interested in cell type predictions you may instead want to consider SingleR or CellO which are reference based cell type classifiers and don't require a user defined set of marker genes.