Question about feature selection

[AmosFong1]

Hello maintainers, I am thinking about excluding mitochondrial and ribosomal genes from my fast topics analysis, as these genes we are assuming are not relevant to our analysis. Without feature selection, the model seems to always describe 1-2 topics which are driven mainly by ribosomal or mitochondrial genes. Do you have any comments on advice for feature selection (although the vignettes recommend to use all genes).

[pcarbo]

@AmosFong1 These sort of data-preparation steps can be study-specific, but yes it is quite common to remove ribosomal and mitochrondial genes in single-cell studies because these genes are typically not helpful in understanding the underlying structure (e.g., the underlying cell types). Also, you should remove genes that are not expressed in any cells, or have expression in only a very small number of cells.

Hope that helps.