What to think about when treating scRNA-Seq data as bulk RNA-Seq data

Question

[chilifan]

We got a bunch of scRNA-seq data and I've done a few tests on them. Now my boss wants me to treat them as bulk RNA-seq data and and compare the differences between treatments. Just as a test. I have not worked with bulk RNA-seq before, so I'm wondering: do you have any opinions on what I should do with these data sets? I have the digital gene expression matricies like this:

          Cell1      Cell2      Cell3             ->        Sample 1
Gene 1        1          2          0                              3              
Gene 2        0          4          0                           1.33
Gene 3        0          0          1                           0.33

My plan is just to add all the cells for one gene and divide all genes with the number of cells, like I show in Sample 1. Should I do some normalization first, scaling, something else? Are there any reasons why this would not be doable?

[Arup Ghosh]

If your data is from 10x genomics you can do it in Loupe cell browser just combine the two h5 files from different treatments(if the analysis is done separately) and do differential gene expression analysis b/w groups. Else you can opt for Seurat and follow the PBMC tutorial.

[chilifan]

But these suggestions still treat my samples as single cell samples, right? My question regards which pre-processing steps I need to take if I'm just averaging my two samples to contain one vector of gene expression each, and compare these two. Starting out from a digital gene expression matrix. :) or am I misunderstanding your answer? :):)

[danvoronov]

I was wondering if it was possible to use one or more of the FASTQ files (if you have 10x genomics), try FASTQC to see if there are any adaptors and bad reads, and then try to use it straight to map onto genome as a normal bulk RNA-seq, or Trinity if no genome is available? If you only have one sample per condition you may not be able to compare them beyond log fold changes since most tools require replicates. Otherwise you can count and sum every occurence of gene in each condition and analyse them, without normalization at this step. If you have replicates tools like DESeq2 can normalize themselves.

[ATpoint]

Please open a new question for this.

[shoujun.gu]

You can use scDE tools, such as DESingle, etc. Bulk analysis usually do not consider dropouts.