I have a RNAseq data for repeated measures cross-over design. See below for the study design.

ID  Group   Time
1   SLEEP   BASE
1   SLEEP   POST
1   CON     BASE
1   CON     POST
2   SLEEP   BASE
2   SLEEP   POST
2   CON     BASE
2   CON     POST

Therefore, there is a dependence between measurements in the sleep group timepoints (BASE v POST) (repeated measures) but also dependence between the measurements between the groups (SLEEP v CON) as they the same participants (cross-over). I have done alot of reading and believe dream (within the variancePartition package) is the way to go.

I have followed the pipeline for edgeR-Limma https://www.bioconductor.org/packages/devel/workflows/vignettes/RNAseq123/inst/doc/limmaWorkflow.html e.g. filtered lowly expressed genes, normalised gene distributions (Calc Norm Factors with method="TMM") Then proceeded to the dream vignette https://www.bioconductor.org/packages/devel/bioc/vignettes/variancePartition/inst/doc/dream.html where I need to design the model.matrix. My questions are: 1) Transcriptome changes within SLEEP group (Sleep BASE v Sleep POST) 2) Transcriptome changes within CON group (CON BASE v CON POST) 3) Interaction: differences in the response of the transcriptome between sleep and CON.

This is where I would appreciate some guidance as to most appropriate way to code the random effect and more specifically how to code the ID.

Question 1) Can you run a crossover random effect in dream?

form <- ~ Group*Time + (1+Group|ID)
#Estimate weights using LMM of dream
vobjDream = voomWithDreamWeights(x, form, ss)

`

OR would a more simple random effect be more appropriate to account for the dependence between the 2 measurements within each group (SLEEP v CON)

form <- ~ Group*Time + (1|ID)

My second Question is regarding the coding of the ID, should it be individual for each timepoint for example:

1_SLEEP_B
1_SLEEP_P
1_CON_B
1_CON_P
2_SLEEP_B
2_SLEEP_P
2_CON_B
2_CON_P

or should it be coded so that the model recognises the paired measurements at baseline and post in each group

1_SLEEP
1_SLEEP
1_CON 
1_CON

If this has already been answered please let me know the post so I can check it out.

This looks like a standard complete block design with participants as blocks and four treatments. The standard analysis would be

design <- model.matrix(~ID + Treatment)

where ID is a factor with 10 levels (participants) and Treatment is a factor with four levels. In other words, it is the same as a paired-comparison design but with four treatment levels instead of two.

There's no need for a random effect. The standard analysis already fully accounts for dependence between repeated measures on the same participant.

Your experiment doesn't appear to be a cross-over design. Cross-over designs adjust for the order in which treatments are presented to participants, but there's nothing in your post to suggest that the order is important or that the order is varied between participants.