Entering edit mode
2.7 years ago
mark.pekarsky
▴
20
I need a statistically significant "decision" that DEGs from group A are different (or not) from group B (after drug treatment). For that, I have a mean fold change delta, calculated as the mean of 5 different treated samples minus the mean of the same, but untreated samples.
A.delta B.delta
gen1 4.9182385 4.5472662
gen2 3.6354761 3.4474844
gen3 3.7592630 3.2838886
gen4 3.9623264 2.8189166
gen5 3.4742171 3.1231050
In my vision, there is not much difference in fold change delta between A and B for the gen1 for example (4.91 is not very far from 4.54). Same for gen2. While for gen4 there might be something interesting. What kind of plot may be the most useful for that kind of comparison?
Is this RNA-seq?
Yes, this is data after DESeq2
if it is after DeSeq2 that tool already computes an FDR that tells you the statistical significance for each row, why do you need another test?
Usually the way to infer patterns is to select the DEGs from the results, scaling their normalized counts and putting it into a heatmap clustered by hclust (or similar methods).
If you want to know if A differs from B, it's simpler to just compare A to B. Can you compare them to each other directly, or is there a batch problem?
Thing is I would like to have a general idea about how population A is different from population B in terms of DEGs. Saying the downregulation/upregulation of the same gene is "greater" in population A than in B. So I can say A is more sensitive to a drug than B.
Comparing both to a control set of samples means that the uncertainty of those control samples is going to muddy the waters.
it is still unclear what are you after, hence people here are confused
When you ran DeSeq2 it gives you the genes/transcripts that are differentially expressed with an FDR. That FDR value is computed over all the transcript there: the group of genes under a threshold is the difference between A and B
The words you use seem incorrect, you cannot have a "greater downregulation" in A vs B, the downregulation consists of the difference between A and B
perhaps your confusion comes from trying to associate a meaningful value to seeing 4 in A and 8 in B. Those numbers: 4 and 8 would not be meaningful when presented alone. (If all you knew is that the value for transcript T in condition A is 8)
Only their ratio is 8/4 = 2 would be meaningful as the fold change difference for the transcript between conditions A and B.
What I think OP means is, they have something like compound A, compound B, and a control. So compound A makes a gene increase 4 fold compared to control, compound B makes it increase 3.5 fold, how do you determine if it's fair to say "A and B have the same effect on that gene" or "A increases that gene significantly more than B"?
Thank you for the reply.
It was a fold change (FC) measurement for 10 samples in total.
5 samples for group A (control and sample) and 5 samples for a group B (control and sample). For group A, I have calculated the mean control fc and mean sample fc, and then, had my delta fold change = mean_sample_fc - mean_control_fc. Same for group B. For every gene (
gen1, gen2, gen3... etc)That is the table that I have posted in the subject of this question.
Once again, I need genes (DEGs) that changed significantly differently between A and B. The example I have chosen,
gen1gene is the most upregulated in both groups, and its activation is SIMILAR (not statistically different) between A and B.You calculated, or DESeq calculated? And you really only have, like, 3 A treated versus 2 A controls? That's very underpowered. On the bright side, I think now you can use interactions, which will not just substract the one fold change from the other (it's not like you can't do that yourself), it will do that and give you a p-value as to if the difference is significant.
I have calculated mean 3A treated vs mean 3A control, one by one. Same for group B. This is the data from the table presented in this question.
Question is how DEGs from group A are different (or not) from DEGs from group B.
Are you using DESeq to calculate the fold change, or not? If not, you should be. Just counting things up in Excel is not statistically sound.
Once you got things into DESeq, you can compare fold changes of two separate things with interactions, as I already said.