Hi

I have a very simple experiment. We have 4 time points including baseline and just are asking the question which genes are differentially expressed over this time course - in response to a vaccination at baseline. I have done a paired t test using different contrasts, but how do I fit a regression and find differentially expressed genes this way?

I have found the section in the limma manual, but here the question is different and they are comparing between two different groups? I suppose it is possible to modify this code to fit my problem, but I do not know exactly how....

library(splines)

X <- ns(targets$Time, df=5)

Group <- factor(targets$Group)

design <- model.matrix(~Group*X)

fit <- lmFit(y, design)

fit <- eBayes(fit)

Thanks,

Chris

If everybody got the vaccination, then you don't have a 'Group', you just have the time. In other words, the example you are quoting is looking at the interaction between the spline fit and the time variable, trying to find genes that have different patterns of expression over time, dependent on their group membership.

You appear to have a much simpler experiment, where everybody gets the vaccination and you are just looking for genes that respond to the vaccine in some manner. With three time points after vaccination, there are limited numbers of patterns that can evolve. Rather than using splines (Gordon already intimated at this in your original question), which is like hunting pheasants with an AK47, you could probably just fit time and time^2 (or even time^3 if you expect genes to for example go up, then down, then back up, but really?) and then do an ANOVA type test on both coefficients. Or you could do it step-wise, looking for genes significant in time^2, then exclude those and look for genes significant in time.

If there aren't too many significant genes, you could just plot them (or do something like using ReportingTools to make a sweet HTML table with plots that show the expression pattern), or you could do something like what maSigPro does, where you use kmeans to find the genes that all have the same pattern type, and infer something about each group.