Hello All,

I am trying to implement LIMMA room duplicate correlation strategy for my RNA-seq samples using blocking setup. I need to consider following points into account.

1. My samples are biologically paired and dependent also not all patients has its paired sample.

for instance, this is how my design table looks like,

 I have 82 samples with more than half of patients has its paired sample and few of them doesn't. According to Limma from user guide, I have adapted to the following workflow. But I have a question,   how limma voom fits into the workflow for a paired experimental design that uses duplicateCorrelation() to handle blocking on the same subject, how is the appropriate experimental design (taking subject blocking into account) fed to voom?

My design looks like,

combined_fac <- factor(paste( design_table$Condition))

design <- model.matrix(~0 + combined_fac)

rownames(design)<-colnames(data)

colnames(design)<-c("Non","PH")

## Then followed with estimating correlation, blocking is input with design and then the Contract for finding DE genes

y <- DGEList(data)
y <- calcNormFactors(y)
v <- voom(y, design)
corfit <- duplicateCorrelation(v, design, block =design_table$Patient

)
v <- voom(y, design, block =design_table$Patient, correlation =
corfit$consensus)
fit <- lmFit(v, design, block = design_table$Patient, correlation =
corfit$consensus)
cont_matrix <- makeContrasts( PHvsNon = PH-Non,  levels=design)

fit2 <- contrasts.fit(fit, cont_matrix)
fit2 <- eBayes(fit2)
result_double <- topTable(fit2, number = nrow(data), sort.by ="none")

I wanted to make the understanding the workflow is right with for my experiment.? or what is the appropriate way to run voom such that it takes all of the necessary design information into its account? 

Thanks

Time seems to be an important factor, so perhaps you should be doing:

combined_fac <- factor(paste0(design_table$Condition, ".", design_table$Time))

Even if you're only interested in DE between conditions, you need to include time in the design to avoid inflating the variance. It is simple to modify the contrasts to test for DE between conditions in the presence of time:

con <- makeContrasts( (PH.ID + PH.REL)/2 - (Non.ID + Non.REL)/2, levels=design)

In any case, your design matrix doesn't have a Phlike column, so I don't know what that is or where it comes from. I also don't know what model$Block is in the call to duplicateCorrelation, perhaps you mean design_tab$Patient.

Otherwise, the code looks fine.