Hello everyone,
I would like to analyze the effect of a treatment on psychological outcomes.
In this project we studied the gut microbiota as a possible mediator of the effect of the treatment on psychological outcomes.
To analyze the microbiota composition at baseline and its evolution during the treatment I carried out a CLR transformation on my microbiota data using your tutorial. We want to use PLS-DA with the shift of gut microbiota composition (with other biological parameters) to select the important variables to describe.When we use asv-relative abundance and clr we obtained quite different VIP scores for the delta (but not for the baseline data).
Does it make sense to subtract transformed data to create a delta for the microbiome variables?
I’m quite lost if anyone can help me 
Thanks for your help.
Camille
Welcome @Cam_Amad!
We have also struggled with that question for a while - and still have not found any answer. Even if we CLR transform the data (and basically project the data out of the simplex), comparing one time point to another for the same taxa might not be completely ‘correct’ given the compositional aspect of the data in any case. I assume you are calculating the delta based on the CLR data. It would seem normal for me though that a PLS-DA on baseline vs a PLS-DA on delta value would bring different results? Or am I understanding the question incorrectly?
You should consider applying a multilevel decomposition to directly take into account the time structure of the data (there has been a few posts on that forum + some tutorial on our website). You could also look at the timeOmics package (in bioC) and this paper. Perhaps this type of analysis would also help answering your question.
Kim-Anh