I have the following experimental design and I´m a bit lost:
I have subgingival samples collected from 30 patients. Two samples were taken from each patient on a first visit (V1):
A sample from a tooth that has not received any treatment (V1-C)
A sample from the treated tooth (V1-T)
(In each patient sample C and T were taken at the same time)
Sampling was repeated for all patients two months after first visit (V2).
In other words, I have 120 samples: 30 V1-C, 30 V1-T, 30 V2-C and 30 V2-T and I am interested in knowing if there are differences between treatment and control in both alpha and beta diversity.
I´m quite unsure what´s the proper statistical analysis/methods for this study because it involves either paired and longitudinal samples. I have read the q2-longitudinal tutorial but it is not clear to me. What´s the best way to analyse these samples?
My best recommendation is to contact a statistician and see if you can collaborate on this project. I think the level of support you need verges into authorship territory, especially since you may have a doubly nested design.
However, there are somethings you might wish to consider in your conversations:
Is an individual's oral microbiome more similar overall, and so do you need to account for that in your model? (I would guess yes, but I mostly work in the gut)
What are the kind of things you want to compare? The time point? The tooth? Figuring that out will help you model better.
Do you have a hypothesis around what you expect to change? Abundance? Richness? Both? Do you have the correct metrics to test that?
Thanks for your answer. I just want to compare the tooth: treatment (T) vs control (C). I had thought to use qiime longitudinal pairwise-differences (for alpha diversity metrics) and qiime longitudinal pairwise-distances (for beta diversity metrics). First, I would perform the T vs C comparison for visit 1 (V1) and then for visit 2 (V2). I don't know if this approach is correct...
I think there are two questions, one that can be easily answered here and one that's a little harder.
So, are q2-longitudinal pairwise-differences and pairwise-distances good ways to get differences between timepoints or along a gradient? Yes. You just have to make sure that your data is coded in a way that the model will accept the gradient. (You may need to code C=0 and T=1 in a numeric column, for example).
I don't think I'm in a position to judge how correct your biological question is, and whether the model you're working with is the best way to address that question. I would recommend working with a biostatistican to make sure your modeling is correct, and inviting them to be a co-author on your publication. This is past the point of support I can provide here.
forum!