Actually the limma guide is right and I am wrong. For a moment I forgot the precise meaning of ":" in context of R's formula system.
You still need to be careful with this, as the limma guide also states quite clearly. Kasper On Tue, Jul 9, 2013 at 9:01 PM, Steve Lianoglou <lianoglou.st...@gene.com>wrote: > Hi Kasper, > > On Tue, Jul 9, 2013 at 5:51 PM, Kasper Daniel Hansen > <kasperdanielhan...@gmail.com> wrote: > > This is side-stepping the question, but I am not aware that it ever makes > > sense to include the "A:B" term in a design matrix without also including > > the main effects of A and B (here I include obvious extensions such as A > + > > A:B + C where B is a coarser factor than C, so here the main effects of B > > are essentially included). Of course, the A+A:B+C example seems to also > > fail in the DESeq2 code you quote, but these examples are rare in comp > bio. > > > > If you're just fitting a model like > > A + A:B > > you're almost certainly doing something wrong from a statistical point of > > view. > > Thanks for the input/guidance -- I'm always happy to get some > linear-modeling-schooling. > > If that's the case, does that mean that I'm interpreting the advice > from the limma user's guide incorrectly? Page 44 in the "Nested > Interaction Formula" section: > > > http://bioconductor.org/packages/release/bioc/vignettes/limma/inst/doc/usersguide.pdf > > Aren't the coefficients extracted for cellA:treatment1, > cellA:treatment2, ..., cellD:treatment4 that you get when modeling > this way exactly the fold changes for the effect of the treatment > within each cell type? > > Thanks, > -steve > > -- > Steve Lianoglou > Computational Biologist > Bioinformatics and Computational Biology > Genentech > [[alternative HTML version deleted]] _______________________________________________ Bioc-devel@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/bioc-devel