Hi folks,

I want to use MAST for DGE analysis on dataset processed with Seurat. I wrote a small function to convert seurat dataset to sca:

seu2sca <- function(seu_obj, freq_expressed = 0.2, group, level_group){
  sce <- SingleCellExperiment(list(counts = GetAssayData(seu_obj, slot = "counts", assay = "RNA")), 
                              colData = seu_obj@meta.data)
  logcounts(sce) <- log2(counts(sce) + 1)
  sca <- SceToSingleCellAssay(sce)
  expressed_genes <- freq(sca) > freq_expressed
  sca <- sca[expressed_genes,]
  cdr <-colSums(assay(sca)>0)
  sca@colData[["cngeneson"]] <- scale(cdr)
  cond <-factor(colData(sca)[[group]])
  cond <-relevel(cond,level_group)
  sca@colData[["cond"]] <- cond
  sca
}

However, I am not sure about following statement from the vignette:

... MAST assumes that log-transformed approximately scale-normalized data is provided.

In the function above raw counts are log transformed, but im not sure what is meant by "approximately scale-normalized". Can you please help me.

Kind regards, Amel

If you go through the vignette this seems the usual normalized counts on log2 scale, see https://www.bioconductor.org/packages/release/bioc/vignettes/MAST/inst/doc/MAST-interoperability.html

Another reference is the conquer GitHub https://github.com/csoneson/conquer_comparison/tree/master/scripts which contains the code from a comprehensive scRNA-seq DE benchmark (https://pubmed.ncbi.nlm.nih.gov/29481549/) where they also use logCPM/TPM.