I'm hoping to quantitatively show differences in total RNA expression for a gene in a cluster of interest between different experimental groups. My exported average RNA values for each experimental group in my macrophage cluster are shown below:

00 d Control: 10.43

00 d PLX5622: 14.48

07 d Control: 16.08

07 d PLX5622: 17.12

28 d Control: 15.90

28 d PLX6722: 19.40

However, the violin plot for Cd68 in the macrophage cluster looks like this

VlnPlot(combine.combined, features = ("Cd68"), pt.size = 1, idents = "Macrophages", group.by = "orig.ident", split.by = NULL, assay = "RNA"):enter image description here

The violin plot suggests that the Control groups actually have more total CD68 RNA in them than in the PLX5622 groups, which isn't what the average RNA values show. Any code you can share showing me how to get total RNA per group would be an amazing help!

Thanks so much!

  • $\begingroup$ IMO I think you've the wrong type of plot. You need to plot expression versus frequency (histogram). Expression differences in that data are specialist area of statistics - it is doable.However I could be wrong. $\endgroup$
    – M__
    Jul 27, 2020 at 23:35
  • $\begingroup$ A violin plot is much like a histogram on its side. Except that the widths of the violin plot are all scaled to be the same. $\endgroup$
    – swbarnes2
    Jul 28, 2020 at 0:40
  • 1
    $\begingroup$ It looks to me like you have a just a tiny number of cells making up your 00 d PLX5622 sample... Having inconsistent numbers of cells is going to bias the results. I would say by eye that the controls are lower in all cases except day 07 (where they are about the same). $\endgroup$
    – story
    Jul 28, 2020 at 9:00
  • $\begingroup$ What do you mean show differences in "total" RNA expression for a gene? All isoforms of the gene? Putting it in plain English, what is your hypothesis? "Cells treated with PLX5622 have higher gene expression of CD68"? If so, do a differential expression analysis between your groups, considering your biological replicates $\endgroup$
    – csgroen
    Jul 28, 2020 at 9:53
  • $\begingroup$ Single cell experiments tend not to have biological replicates the way a bulk-RNASeq experiment would. If anything, each cell is a biological replicate, with the usual caveats about assessing expression in a single cell. $\endgroup$
    – swbarnes2
    Jul 28, 2020 at 20:40

2 Answers 2


(I currently cant comment as I used up all my rep for a bounty.)

Could you explain how you calculate the average? In you plot the averages are between an expression level of 2 and 3, while you report values over 10. Given the data you show, there is no significant difference and I would guess p-values are above 0.8!

And you mention that the violin plots suggest the control groups have higher expression.. but how do you see that? For me they are identical with minimal trends that the PLX5622 groups are higher.

And to address your question regarding showing the total: No, you would want to show the mean or median, as the total is pointless, given deviating sample sizes.

EDIT: Overall, I have a feeling you uploaded the wrong picture!

EDIT2: I think I got your point now: The difference in mean is caused by an increased number of outliers in your controls at the value 0. Despite those, the means are identical and this is what can be seen through the violin plots. You should rather try to diminish the effect of these outliers.


To me, it looks like the widest parts of the plots are higher in the treated. I don't see why you are so sure that the violin plots are hugely different from the calculated averages.

Either way, the differences are really small. Maybe not statistically significant.

BTW, in the future, you might want to disguise your treatment and gene of interest when on a public forum.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.