Hi All,

I have samples from 45 individuals at 2 time points and these individuals were given a specific diet. The BMI of these individuals were recorded at both time points. The responders to the diet are the ones that show decrease in BMI from T1 to T2. I want to use continuous variable (BMI) to identify genes that are predictors of response to diet. I also have other co-variates that need to be controlled like age & gender. The metadata table looks like following

SubjID  SampleID    Time    Age Gender  BMI
Sub1    1   T1  23  M   29
Sub2    2   T1  19  F   31
Sub3    3   T1  28  M   32
Sub4    4   T1  45  F   33
Sub5    5   T1  67  F   29.5
Sub6    6   T1  43  M   32.5
Sub7    7   T1  36  M   24
Sub1    8   T2  23  M   26
Sub2    9   T2  19  F   28
Sub3    10  T2  28  M   30
Sub4    11  T2  45  F   28
Sub5    12  T2  67  F   30
Sub6    13  T2  43  M   31
Sub7    14  T2  36  M   32

I am not sure the design that I should go for. The one that I think is appropriate is

dds <- DESeqDataSetFromMatrix(countData=cts,colData=meta,design=~BMI+Gender+AGE+Time:BMI)

dds <- DESeq(dds,test="LRT", reduced=~Gender+AGE)

res=results(dds, contrast=list("T1.BMI","T2.BMI"))

Is this right to identify the genes that could be potential identifiers of responders to the diet as the BMI at time point 2 (T2) has decreased?

All suggestions are very welcome

Thank you

This is genes where the association of counts with BMI changes between the time points.

I will note that you should use centered and scaled numeric covariates in DESeq2:

dds$x_cs <- ( dds$x - mean(dds$x) ) / sd( dds$x )
design(dds) <- ~ x_cs + ...