Hello all,

I am working with microarray RNA data and using the voom/limma pipeline. I have repeated measures, so I am utilizing the duplicateCorrelation to account for correlated measurements.

My main question stems from the fact that even after a Bonferroni correction, I am still finding that ~50% of my 800 markers are significant.

The variable of interest Time is coded 0/1. demo.long contains my demographic information, mRNA are the raw counts and have been converted into a DGEList object. I was told that you can perform voom twice to get more accurate estimates when dealing with repeated measurements.

Here is my code:

design <- with(demo.long, model.matrix(~ Time))
mRNA <- calcNormFactors(object = mRNA, method = "TMM")
colnames(mRNA$counts) <- gsub("B|-REF", "", colnames(mRNA))
v <- voom(mRNA, design, plot = F, normalize.method = "quantile")

dc <- duplicateCorrelation(v, design, block = as.factor(demo.long$ID))
vf <- voom(v, design = design, block = as.factor(demo.long$ID), correlation = dc$consensus.correlation, normalize.method = "quantile")

fit.ref <- lmFit(v, design = design, block = as.factor(demo.long$ID), correlation = dc$consensus.correlation)
fit.refEb <- eBayes(fit.ref)
tT.ref <- topTable(fit.refEb, 
                    coef = 2, 
                    n = dim(mRNA)[1], 
                    sort = "logFC", 
                    adjust.method = "bonferroni")

I am curious if there is anything fundamentally wrong with the code above and if there is a glaring reason that could explain the high rate of significant markers.

Thank you.