nf-core/rnaseq produces non-integer counts in salmon.merged.gene_counts.tsv

[SteveYost]

Hi all, this is my first post here. I'm relatively new to RNASeq, and I'm trying out BioConductor pcaExplorer. In pcaExplorer the "generate the dds and dst objects button" produces an error:
"Error in DESeqDataSet: some values in assay are not integers", when running DESeqDataSetFromMatrix under the covers, because DESeqDataSetFromMatrix does indeed require integer counts. Is RNASeq's use of Salmon producing non-integer counts because it's doing normalization? If so, how can I get raw gene counts from the RNASeq pipeline?

This looks helpful: https://bioconductor.org/packages/re.../tximport.html
but I'm not sure how I would get that result into pcaExplorer.

[fchatonnet]

Dear Steve Yost,

Salmon actually produces transcript level estimates rather than gene level raw counts with are used by DESeq2 or EdgeR to perform differential analysis. I can suggest to use either HTSeq-counts or featureCounts with the relevant GTF/GFF reference file to get raw counts directly usable by DESeq2.
If you want to use Salmon estimates, you can follow the procedure described in the tximport package manual that you found.
First generate a count table by following this step:

https://bioconductor.org/packages/release/bioc/vignettes/tximport/inst/doc/tximport.html#Salmon

Then produce a dds by using the following command using the DESeqDataSetFromTximport() function from the DESeq2 package:

https://bioconductor.org/packages/release/bioc/vignettes/tximport/inst/doc/tximport.html#DESeq2

Good luck!

[SteveYost]

Thank you, fchatonnet!