Statistical comparison

It looks like I do not have any significant statistic difference at all in my dataset when I compare the different groups; not even a single cluster shows difference between groups. It also looks like that all the values of adjusted p-values are never higher than 0.6-0.7.
I really do not understand if my hypothesis has been brutally not proven, or if I am doing something wrong.


One explanation is that the lack of significant results could be due to low N and/or high level of variation between replicates of your group(s). Did you check this? For example, you could plot some genes on the violin plot in the Plots & Tables module, and show all samples - this will show how much variation in gene expression there is across all samples.

Perhaps @Sara_Bioinformatics might have more thoughts on this?

Hi Salvo,

I just wanted to add something to what Vicky said. Mostly questions to better understand your situation:

  • Are these results consistent across all comparisons, or you see this issue only when comparing specific groups?
  • If I remember correctly, you run the same analysis using a different tool. If that is correct, did you observe different results for the same analysis using a different tool?
  • If you’ve obtained different results with another tool, are you seeing the same genes appearing in the list but with high p-values, or are the results completely different?

Dear @Vicky_PM_at_Parse and @Sara_Bioinformatics ,
first of all: thank ou for your replies.

The sample groups are “asymmetric”, because I have 4 samples for one group, 3 for the second one, and 2 for the third one.
I compared some genes (those that were either down- or up-regulated, considering just the log fold change) and I have a mixed situation, with some violin plots showing genrs for all individual samples (all of the 9) and some variations, and some that would show up only for some. I hope I made sense, because I am still learning how to interpret these plots.

The results are consistent across all comparisons I did, and the genes that were showing up as up-regulated or down-regulated (p-value sometimes significant) with Loupe were very different: one or two genes show up again in Cellenics as either down- or up-regulated, with no significance at all.

I’d like to specify that the not significant values regard the adjusted p-values. The “classic” p-values show some significant values

Thanks for the additional details. It is possible that the removal of ambient RNA via CellBender could have significantly influenced the differential gene expression results. This procedure aims to eliminate “noise” in the data, potentially leading to different outcomes in comparison to analyses performed without this step.

However, given the discrepancy you’ve observed, and to confirm the results you see in Cellenics and ensure there are no inaccuracies or potential bugs in the differential expression analysis process (which I think it’s unlikely because DE analysis it’s a well-established feature in Cellenics), I would suggest performing a cross-validation step.
You can do this by downloading the RDS object from Cellenics and performing a differential expression analysis in RStudio using a package of your choice (e.g. limma, or presto). Using different tools and comparing the results can help to cross-validate the results obtained from Cellenics. Also note that as every tool relies on specific algorithms and statistical models, there may be inherent differences in results between tools.

Alternatively, we can assist you in this task through a small consultancy project. We could run the differential expression analysis with a different tool on your behalf.
Please don’t hesitate to contact us if you would like to discuss this option further.