Hey all,

I think I'm misunderstanding the F-statistic and/or the use of contrasts. This is a time series experiment. I'm interested in any genes that are differentially expressed between any two time points, so I did this:

cont.dvsd <- makeContrasts("DAY_2-DAY_1", "DAY_3-DAY_2", "DAY_4-DAY_3", "DAY_5-DAY_4", levels=design)
fit2 <- contrasts.fit(fit, cont.dvsd)
fit2 <- eBayes(fit2)
res <- topTable(fit2, adjust="BH", number=Inf)
res['SLC38A6',]

        DAY_2.DAY_1     DAY_3.DAY_2     DAY_4.DAY_3     DAY_5.DAY_4  AveExpr    F         P.Value      adj.P.Val
SLC38A6 0.9278499       2.389819        2.844422        2.12539      0.2260567  13.16474  1.314994e-08 1.893592e-05

I thought the F-statistic incidates whether the gene is differentially expressed in any of the contrasts that were used. However, when I set up the contrasts differently (now comparing all timepoints to baseline), the F-statistic (and also p value and adjusted p value) comes out the same:

cont.dayzero <- makeContrasts("DAY_2-DAY_1", "DAY_3-DAY_1", "DAY_4-DAY_1", "DAY_5-DAY_1", levels=design)
fit2 <- contrasts.fit(fit, cont.dayzero)
fit2 <- eBayes(fit2)
res <- topTable(fit2, adjust="BH", number=Inf)
res['SLC38A6',]

        DAY_2.DAY_1     DAY_3.DAY_1     DAY_4.DAY_1     DAY_5.DAY_1     AveExpr    F         P.Value     adj.P.Val
SLC38A6 0.9278499       1.461969        0.4546035       -0.7190325      0.2260567  13.16474  1.314994e-08    1.893592e-05

This is true for all genes. How is this possible? I expect there to be genes that are DE between day 4 and day 1 that do not show up in a comparison between day3-day4 or day4-day5 (as in my first setup).

This result is completely normal. The first set of contrasts can only be all zero if all the time points are equal. The same is true of the second set of contrasts. Therefore they give the same F-test. Indeed, any set of 4 non-redundant contrasts between the 5 time-points will give the same F-test. In all cases, the F-test will test for any differences between the 5 time-points, exactly like a one-way anova.

I expect there to be genes that are DE between day 4 and day 1 that do not show up in a comparison between day3-day4 or day4-day5 (as in my first setup).

If genes are DE between day 4 and day 1, then they must be also be DE in at least one of the contrasts in your first setup. It is not possible for day 4 and day 1 to be different, but for day 2 to be equal to day 1, day 3 to be equal to day 2 and day 4 to be equal to day 3.