Hi, I want to analyse time-series data. My design is as pasted below,

sample  group

H24_Untreated_R1    H24_Untreated

H24_Untreated_R2    H24_Untreated

H24_Untreated_R3    H24_Untreated

H24_Treated_R1  H24_Treated

H24_Treated_R2  H24_Treated

H24_Treated_R3  H24_Treated

H48_Untreated_R1    H48_Untreated

H48_Untreated_R2    H48_Untreated

H48_Untreated_R3    H48_Untreated

H48_Treated_R1  H48_Treated

H48_Treated_R2  H48_Treated

H48_Treated_R3  H48_Treated

I need to find genes changing in response to time. My script is as pasted below,

design <- model.matrix( ~0 + group, data = des)   #des is above dataframe
fit <- lmFit(ip,design)
cont.matrix <-makeContrasts(H24_Untreated_vs_H24_Treated = (groupH24_Treated - groupH24_Untreated),
                        H48_Untreated_vs_H48_Treated = (groupH48_Treated - groupH48_Untreated),
                        levels = design)
fit.cont <- contrasts.fit(fit, cont.matrix)
fit.cont <- eBayes(fit.cont)
limma.res <- topTable(fit.cont,n="Inf")

"limma.res" has values in pasted below columns H24UntreatedvsH24Treated, H48UntreatedvsH48Treated, AveExpr, F, P.Value, adj.P.Val

My question is , if we take the list of genes based on the adj.P.Val cut-off of interest, are those the genes which are changing in response to time? I am not interested in direction of regulation. Only want to find genes which are changing over time.

You have made two comparisons; treated vs untreated at each time point. And when you call topTable with no coefficient, it does an F-test that tests for evidence that either of those comparisons is different from zero. Since both comparisons are within a time point do you think you have tested for changes over time? You have to make comparisons between time points to test for differences between time points.

Your experiment is a bit more complex than that, however, and makes me think that what you say you want to test for might not actually be what you really want? In other words, if you do an experiment where you treat some, um, things, and then measure at two time points, and then you do the same thing without the treatment, it's more likely that you want to know if some genes have a time-dependent reaction to treatment. In which case you really want the interaction, which is

(groupH48_Treated - groupH48_Untreated) - (groupH24_Treated - groupH24_Untreated)

Which you can see is testing for any genes for which the differences at time 48 are larger or smaller than at time 24. Algebraically this is the same as

(groupH48_Treated - groupH24_Treated) - (groupH48_Unreated - groupH24_Untreated)

Which might be easier to interpret (but either contrast will result in the same exact genes).