iNKT annotation?

[mihem]

@ncborcherding

Hi Nick,

sorry, not sure this is within the scope of scRepertoire, but in my current project I would like to annotate iNKT (and maybe MAIT). This is possible with marker genes, but more reliable when looking at the CDR3 region. The more recent versions of CellRanger seem to do that: https://www.10xgenomics.com/support/software/cell-ranger/latest/hidden/cr-5p-vdj-algorithm-inktmait. However, I am confused because I thought this needed VDJ (so e.g. not possible when 3' was used, but you can still run annotate then).

However, I have an ongoing project with an older CellRanger version, plus I would think that this would be a nice feature for scRepetoire

Thank you!
Mischko

[mihem]

Okay sorry, I think I now understood that this annotation is just part of cellranger vdj and has nothing to do with the more recent feature cell annotation https://www.10xgenomics.com/support/software/cell-ranger/latest/analysis/running-pipelines/cr-cell-annotation-pipeline.

However, nonetheless it would be nice if scRepertoire would be able to process this information. Since it's already in clonotypes.csv scRepertoire could just use this information? Or start from scratch based on the CDR3 region to identify iNKT? https://github.com/10XGenomics/enclone/blob/cellranger5.0/enclone/src/human_iNKT_CDR3.json

Thanks

[ncborcherding]

@mihem

Great suggestion - I actually wrote the iNKT and MAIT assigning functions for another project - but there is no reason not to update them and implement them into scRepertoire.

I will add this to the to-do list and update this issue as I make progress.

Thanks,
Nick

[mihem]

Great thanks.

[mihem]

@ncborcherding sorry, need to work on this soon. Could you maybe be so nice and share the functions you wrote (as you probably don't have time to include this in scRepertoire the next days I suppose ;) ).
I know this is a rather a question for 10X then for scRepertoire, but maybe you could help: The column iNKT looks like this in my sample:

So lots of (all of the first 50) clonotypes have iNKT evidence. But in the end I want to know: which of these are really iNKT? This is PBMC (of sick patients), so I would expect iNKT to be not more than a few percent of all T cells?

Thanks,
Mischko

[ncborcherding]

Hey @mihem,

I am more familiar with using the gene and amino acid length in terms of defining them. Here is the function I wrote - it will need a little work as it has some internal functionality, but you should be able to get it going:

scoreInvariant <- function(input.data, 
                           type = NULL,
                           species = NULL) {
  
  TCRS <- getTCR(input.data, chains = "both") #this is just pulling the data from the seurat object, so you can do it yourself from the meta data
  
  comp <- switch(type,
                "MAIT" = .MAIT.criteria,
                "iNKT" = .iNKT.criteria,
                stop("Please select either 'MAIT' or 'iNKT' for type."))
  
  cells <- unique(TCRS[[1]]$barcode)
  
  lapply(cells, function(x) {
    TRA.v <- TCRS[[1]][TCRS[[1]]$barcode == x,]$v
    TRA.j <- TCRS[[1]][TCRS[[1]]$barcode == x,]$j
    TRB.v <- TCRS[[2]][TCRS[[2]]$barcode == x,]$v
    TRA.length <- nchar(TCRS[[1]][TCRS[[1]]$barcode == x,]$cdr3_aa)
    if(any(is.null(c(TRA.v, TRA.j, length, TRB.v)))) {
      score <- 0
    } else {
      if (grepl(TRA.v, comp[[species]]$TRA.v) & 
          any(grepl(TRA.j, comp[[species]]$TRA.j)) & 
          any(grepl(TRB.v, comp[[species]]$TRB.v)) &
          TRA.length %in% comp[[species]]$TRA.length) {
        score <- 1
      } else {
        score <- 0
      }
    }
    score
  }) -> individual.scores
  output.scores <- data.frame(row.names =  TCRS[[1]]$barcode, score = unlist(individual.scores))
  colnames(output.scores)[1] <- paste0(type, ".score")
  return(output.scores)
}

.MAIT.criteria <- list(mouse = list(TRA.v = c("TRAV1"), 
                          TRA.j = "TRAJ33", 
                          TRA.length = 12, 
                          TRB.v = c("TRBV13", "TRBV19")), 
                       human = list(TRA.v = "TRAV1-2", 
                                    TRA.j = c("TRAJ33", "TRAJ20", "TRAJ12"), 
                                    TRA.length = 12,
                                    TRB.v = c("TRBV6, TRBV20")))

.iNKT.criteria <- list(mouse = list(TRA.v = "TRAV11", 
                                    TRA.j = "TRAJ18", 
                                    TRA.length = 15), 
                        human = list(TRA.v = "TRAV10", 
                                     TRA.j = "TRAJ18",
                                     TRA.length = c(14,15,16),
                                     TRB.v = "TRBV25-1"))

I'm on clinical service this week, would be happy to help out more when I catch my breath next week.

Nick

[mihem]

Thanks very kind.
So the iNKT criteria on these three genes?
So you don't use this list right?
https://github.com/10XGenomics/enclone/blob/cellranger5.0/enclone/src/human_iNKT_CDR3.json

Do you think this does not make sense or is not necessary?

Haha same here regarding clinical service.

[ncborcherding]

Oh interesting - thanks for the link. They are using previously published sequences for identifying cells. You could do that and replace the length check I suppose. My one question would be is that comprehensive? The way I envisioned using the scores is that as a permissive signal - like these cells qualify based on TCR, but that is not a guarantee.

I need to do some more reading in this area to see how to implement it.

Nick