Differential expression of scRNA-seq after Seurat

Question

[lshepard]

Hi,

I am currently planning a scRNA-seq analysis and I would like to have some feedback from this community. As this is a Drop-seq run, I will be using the Seurat package to identify clusters of cells. However, the program does not perform differential expression among experimental conditions (only among populations of cells).

From what I gathered, bulk RNA-seq methods have been shown to perform well (DESeq2, EdgeR), along with scRNA methods like MAST. Recently, Seurat has included DESeq2 and MAST as part of the tests of expression for cell clusters. I am leaning towards DESeq2 for my approach as I like how the design can be constructed, however, I wanted to see if this sounds reasonable among users in here. As a further question, Seurat performs global normalization in the data, which confuses me a bit in the need to re-normalize in DESeq2 (although I think it is necessary as it normalizes for sequencing depth - thoughts?).

Thanks for the input.

Answer 1

[igor]

However, the program does not perform differential expression among experimental conditions (only among populations of cells).

You can set cell identities to anything. Then, differential expression can be performed between any of the identities.

You can run DESeq2 or MAST as part of Seurat's FindMarkers() function. That way, you don't have to worry about moving the data between different packages manually.

Check the Seurat differential expression vignette: http://satijalab.org/seurat/de_vignette.html

[lshepard]

I edited my reply because your answer made me think of the following:

Since you need to keep everything within a single object, I could merge the objects (control vs treatment) and follow this up with FindMarkers correct?

[igor]

I didn't realize your samples were in separate objects.

You can merge multiple Seurat objects into a single Seurat object. All the annotations in the meta.data slot should be preserved.

[lshepard]

Thanks, I don't know why I was thinking it was better to keep the treatment groups as separate objects. It makes more sense now to merge them. I appreciate the input!