Hi everyone,

I have protein peak intensities that I am using limma to analyze. My design is 3 concentrations over 3 timepoints with repeated measures in that the same individual is tested at each of the concentrations at each timepoint (i.e. using blood in cell culture wells for exposure).

My replicate size is only 5 due to limitations, and I have used the duplicateCorrelation function for blocking by individual. I have made sure that the input data and metadata are in the same order. Some of my raw p values are quite low (<0.001) but the FDR is continually around 0.98-0.99. I understand it is better to determine DE proteins (or genes) using FDR. When I look at boxplots of proteins I expect to be different, I do see visual differences in the medians but this is not captured by the FDR. However, I am wondering if I am capturing the variation with the repeated measures correctly. Or perhaps it is a consequence of having so few replicates over ~2500 proteins + thus multiple comparison tests.

I normalized my data and transformed by log10. Then used this code, following 9.7 Multi-level Experiment from the limma user guide:

Treat <- factor(paste(m$Concentration,m$Time,sep="."))
design <- model.matrix(~0+Treat)
colnames(design) <- levels(Treat)
corfit <- duplicateCorrelation(d,design,block=m$Subject)
corfit$consensus
[1] -0.0266345
fit <- lmFit(d,design,block=m$Subject,correlation=corfit$consensus)
contrast.matrix <- makeContrasts(
    CvHtime0 = H.0-C.0,
    CvHtime4 = H.4-C.4,
    CvHtime8 = H.8-C.8,
    CvLtime0 = L.0-C.0,
    CvLtime4 = L.4-C.4,
    CvLtime8 = L.8-C.8,
     levels=design)
fit2 <- contrasts.fit(fit, contrast.matrix)
fit2 <- eBayes(fit2, robust=TRUE, trend=TRUE)
topTable(fit2, coef = "CvLtime0", adjust="BH")
summary(decideTests(fit2))

Thank you for any advice!