Hello,

I am having trouble finding a proper design AND get all of desired contrasts for my study.

Data description

Two groups (5 and 7 subjects), three time points, gut microbiota analysis, no missing data

Group G: T0 and T2 normal diet, T4 no gluten Group noG: T0 normal diet, T2 and T4 without gluten

To put it in parallel:

• Time4: Ta__Tb__Tc__Td
• G:____Ta__Tb__Tc__x : Ta,Tb
• noG:__x__Tb__Tc__Td : Tb

where in bold are time points with normal diet for two groups (problem with visualisation so I had to note it aside).

Important inter-subject effect to account for.

Constrasts of interest

1. (G+noG)Tb vs (G+noG)Tc to see the effect of diet change
2. Ta vs Tb for G (paired analysis) : see the effect of normal diet (no change)
3. Tc vs Td for noG (paired analysis) : see the long term effect of no gluten

Proposed designs

a) ~ Group + Group:Subject.n + Group:Time (Time: T0/T2/T4)

Here I can account for subjects and difference in time effect on group (since T2 will be with or w/o gluten depending on the group). As contrast, I should be able to answer the questions 2. and 3., however impossible to answer the 1st one. If I replace Time with three points with Time6 with four points to be able to answer Q1, even after removing 0 rows from design matrix I have error messsage "matrix not full rank".

*Subject.n: subjects are renamed to have identic names in both groups as suggested in tutorial "Analysing RNA-seq data with DESeq2".

b) ~ Subject + Time4

Design accounts for subjects but not in paired manner, so not possible to answer Q2 and Q3 but easily possible to answer Q1.

Despite reading numerous tutorials and threads, I am not sure that my logic is correct and if there is a way to have one design, which I would prefer and what is suggested in general, to answer all of my questions.

Any advice or help is much appreciated since I really cannot think of a proper way out of this !