My goal is to explore if the ratio of ASVs classified within certain taxa are different between two conditions.
For example, evaluating Clostridiales vs Pseudomonadales, i intend to check if the ratios (Clostridiales_ASVs_condition_A/Clostridiales_ASVs_condition_B) and (Pseudomonadales_ASVs_A/Pseudomonadales_ASVs_B) are different. Importantly, I have to say that these groups aren't necessarily differently abundants (are the ones I'm interested in).
I think it will also be very interesting if these ratios could be calculated with random subsets of ASVs from these groups (bootstrap of 1000 times maybe?) to give some statistic significance?
Is there any option within qiime2 that allows this kind of aproach? I looked at qiime2 gneiss but I think this only concerns differently abundants' ratios.
Dear @colinbrislawn , thanks for your quick answer, and sorry for my delayed one!
If I undersand correctly, a differential abundance aproach (ancom or gneiss or others) will give me those taxa whose abundance change between the two coniditions (A and B). However, I am interested in comparing these changes between taxons (I already know they are differentials). For example, the reduction I see in Clostridiales between conditions A and B is bigger or smaller than the other taxa combined?
I am not sure how can these differential abundance aproaches provide this information. For example, if I want to see if Clostridiales reduce more than other taxa I could compare their p-value or ancom-W? In this case I have the isue that I'll have my statistics for Clostridiales, but I won't have a value for "the rest of the taxa"
I got you. As an example, you want to compare the magnitude of change in Clostridiales to the magnitude of change in Firmucates to show which changes more.
I'm not an expert on stats, so it may be time to reach out to a card-carrying statistician.
My gut feeling is that this is a comparison of means. Commonly, we would compare observed mean differences to randomized mean differences from a null distribution. But here you want to compare the two observed mean differences between two taxa. Because Microbiome Datasets Are Compositional, common methods that assume independent sampling are not a great fit.
This is an excellent question! Let's see what others have to say.