Entering edit mode
17 days ago
seaweed96
•
0
I have three conditions
- Treatment 1
- Treatment 2
- Mock
While I have been doing pairwise comparisons across the three I now need to compare treatment 1 vs treatment 2 and then use that to compare to mock. Treatment 2 is a negative analog of treatment 1, this explains why I want to "subtract" this effect in addition to mock. It would be great to get some advice on how to approch this. Logically I would expect that contrasting the counts of both treatments previous to finding the log2FoldChange to mock would make sense, but I would appreciate the advice.
yes, I think your approach is approriate,
using a design with all conditions Mock, T1, T2 will compute the overall normalization factors, then a contrast with T1 - T2 vs M would do it.
I'm curious how you would propose for that to work from a technical stand-point and what the resulting biological interpretation of that would actually be. I just want to make sure the poster doesn't end up doing a contrast like (T1-(T2-M))-M, which will very likely not be what they want since it's actually just finding whether T1-T2 is significantly different from 0 (or some other fold-change).
I don't know what the OP intends to do, or whether that makes sense.
The way I read it, T2 may have an independent effect and they might want to remove that. In that case the fold changes of (T1-T2) vs M could be computed with a contrast formula - I think ... but writing R contrasts is not something I regularly do and the whole thing is not intuitive to me, so I always have to do double checking that it does what I think.
But conceptually, if T2 has an independent, additive effect on the fold change, a subtraction should work.
A subtraction only works if it's (A) a confounder affecting T1 and not mock or (B) if treatment 1 is otherwise a mixture with treatment 2. In that case then this basically devolves to an incomplete factorial design and you're totally right that subtraction (i.e., looking at the interaction term) makes total sense.
It may be helpful to know more specifically about what you mean by treatment 2 being a negative analog of treatment 1. Are these expected to have opposite changes on the same pathway? Is there any estimate of the relative magnitude of these changes relative to each other? I'm trying to visualize how the statistics for this would actually work if one were to subtract another effect before the comparison. What would happen if Treatment 2 decreases expression of a gene that is already maximally expressed in the Mock condition and then we "subtract" that out of T1? Would that then not result in this subtracted result being artificially different from mock? While that does tell you that a given gene is affected by perturbations in the pathway, I would think it's then easy to draw the wrong biological conclusion regarding the nature of that change.
Thanks for taking the time to respond.
In this case the negative analog is a chemical with a similar structure (isomer) but that has been found to have no biologically significant roles. The idea behind using such treatment was to account for reponse to chemical treatment that is non-specific. Mock is just water, without chemical and would therefore represent general perturbations from doing a treatment.
From analyzing T1-M and T2-M and comparing the DEGs of the two data sets they overlap around 10%. This is why I considered doing (T1-T2)-M. With regards to your point that i may artificially reduce expression from T1-T2, im not sure this would be a problem, this subtraction is artificial in nature and already functions a bit like a mock condition, the extra comparision to actual mock is to further reduce the noise generated by treatment.
Ah, you'll want to simply do T1-T2. I don't think the comparison to mock will further reduce your noise, if anything it will have the opposite effect if you try to do this directly in the model.