Hi,

I have a RNAseq dataset with three conditions. I am using the raw counts reported by salmon in deseq2 to identify differentially expressed genes. From the preliminary analysis, there are several genes with significant padj values and extremely high log2fc values (>20). When looking into the raw counts for these genes, I see that majority of the samples have zero counts and only one sample having very high counts.

For example: CTRL 5 , 0 , 0 and treatment 0, 500, 0

I would expect deseq to flag these genes but instead they are reported to be significantly differentially expressed with huge fold changes.

Has someone experienced this before or have any advice on how do account for such genes with large variation in counts?

Thank you in advance for the feedback.

1) Check the section on prefiltering in the vignette.

2) If you say that you use salmon output directly then it is transcript level. Consider using tximport first.