Hi,

Following a single-cell RNA-seq workshop, I created a Seurat object (my_data), normalized the data, and then tried to identify highly variable genes using two different R packages: Seurat and M3Drop.

variable_genes_Seurat <- my_data %>%
  FindVariableFeatures(selection.method = 'vst', nfeatures = 2000) %>%
  VariableFeatures()

variable_genes_M3Drop <- my_data %>%
  GetAssayData('counts') %>% # unnormalized
  NBumiConvertData() %>%
  NBumiFitModel() %>%
  NBumiFeatureSelectionCombinedDrop(ntop = 2000) %>%
  rownames()

I compared the results and found out that the gene lists were pretty different and shared only 588 of 2000 genes.

shared_variable_genes <- intersect(variable_genes_Seurat, variable_genes_M3Drop)
length(shared_variable_genes)

I wonder why the results are so different, and which feature list - shared_variable_genes / variable_genes_Seurat / variable_genes_M3Drop - I should use for dimensionality reduction and clustering.

Thank you!