mean.fxn Decision Not Explained

[DarioS]

mean.fxn: Function to use for fold change or average difference calculation. If NULL, the appropriate function will be chose according to the slot used

Which slot relates to which automatic function choice? Be explicit, perhaps in Details section.

[lepriolc]

Hi @DarioS,

I am not part of Seurat team but you can find the mean.fxn function definition in the differential_expression.R script (https://github.com/satijalab/seurat/blob/master/R/differential_expression.R).
Basically, when using log-counts, this function is set to:

function(x) {
        return(log(x = rowMeans(x = expm1(x = x)) + pseudocount.use, base = base))
      }

Using counts, it is set to:

function(x) {
    return(log(x = rowMeans(x = x) + pseudocount.use, base = base))
  }

And, finally, when using scale.data slot, it is set to the rowMeans function.

When using an SCT assay, it appears that this function is not set correctly when using data from the counts slot. See more details in the issue #7095.

Christophe

[lucygarner]

I have found a bug in this.

If:

norm.command <- paste0("NormalizeData.", assay)
Command(seurat_all, command = norm.command, value = "normalization.method")

gives:

NormalizeData.RNA has not been run or is not a valid command.

mean.fxn is set to:

function(x) {
    return(log(x = rowMeans(x = x) + pseudocount.use, base = base))
}

even if slot = "data".

This can happen if you have run NormalizeData on a previous Seurat object, subset and renamed that Seurat object, and not re-run NormalizeData.

[saketkc]

Hi @DarioS and @lucygarner, I have just pushed 992c1a9 to develop branch which should fix this and related issues. Can you pull from the develop branch and re-open this issue if you are still facing issues?

devtools::install_github("satijalab/seurat", ref="develop")