Question: Limma time course experiments
0
9 months ago by

In limma manual, the example of time course experiment (Section 9.6) the target frame showed 2 replicates of 0hr wt and mutant but why it was not considered in the levels or in design? Can anyone please explain the discrepancy? The example, "we consider here a two-way experiment in which time course profiles are to be compared for two genotypes. Consider the targets frame

FileName Target
File1 wt.0hr
File2 wt.0hr
File3 wt.6hr
File4 wt.24hr
File5 mu.0hr
File6 mu.0hr
File7 mu.6hr
File8 mu.24hr


The targets are RNA samples collected from wild-type and mutant animals at 0, 6 and 24 hour time points. This can be viewed as a factorial experiment but a simpler approach is to use the group-mean parametrization.

> lev <- c("wt.0hr","wt.6hr","wt.24hr","mu.0hr","mu.6hr","mu.24hr")
> f <- factor(targets$Target, levels=lev) > design <- model.matrix(~0+f) > colnames(design) <- lev > fit <- lmFit(eset, design)  Thanks limma • 190 views ADD COMMENTlink modified 9 months ago • written 9 months ago by shuvadeep.maity0 Answer: Limma time course experiments 0 9 months ago by United States James W. MacDonald51k wrote: You can often answer your own question by processing things using R and seeing what happens. As an example: > targets <- data.frame(FileName = paste0("File", 1:8), Target = paste0(rep(c("wt.","mu."),each = 4), rep(c(0,0,6,24), 2), "hr")) > targets FileName Target 1 File1 wt.0hr 2 File2 wt.0hr 3 File3 wt.6hr 4 File4 wt.24hr 5 File5 mu.0hr 6 File6 mu.0hr 7 File7 mu.6hr 8 File8 mu.24hr > targets$Target
[1] wt.0hr  wt.0hr  wt.6hr  wt.24hr mu.0hr  mu.0hr  mu.6hr  mu.24hr
Levels: mu.0hr mu.24hr mu.6hr wt.0hr wt.24hr wt.6hr
> lev <- c("wt.0hr","wt.6hr","wt.24hr","mu.0hr","mu.6hr","mu.24hr")
> f <- factor(targets\$Target, levels = lev)
> f
[1] wt.0hr  wt.0hr  wt.6hr  wt.24hr mu.0hr  mu.0hr  mu.6hr  mu.24hr
Levels: wt.0hr wt.6hr wt.24hr mu.0hr mu.6hr mu.24hr


So the duplicates ARE accounted for - the lev object is just used to reorder the factor levels.

The duplicates aren't incorporated in the contrasts matrix, because the contrast is made between the coefficients, rather than the observations. In other words, in this model the coefficients are the mean of each group, and the contrasts are the difference between those means.

Just to make sure, the correlation between the duplicates can be modeled with the duplicateCorrelation function, right?

Hypothetically yes, but you need a reason to want to do that sort of thing, and there isn't a compelling reason here. In other words, you use duplicateCorrelation when you have a reason to believe that there will be correlations between observations that you want to control for. So if the time course involved taking samples from the same animals at each time (which is apparently not the case for this example), then the repeated measurements from each animal could be correlated.

In this case the only duplicates are for the 0 hr WT and MU animals, and unless they were measured twice (why?) there is no expectation that they should be correlated, and duplicateCorrelation should return a value very close to zero.