Answer: about two way anova on time and treatment
I usually start this kind of analysis, as you did, with per-gene two-
ANOVAs. I then compute p-values for the significance of the complete
ANOVA (in effect comparing the full model with the null model), and
some kind of FDR estimate to set a cutoff on the p-values for
With this approach, I only decide on the relative importance of th
factors for the genes that have already been identified as
in the sense that the model explains some part of the expression data.
For the interpretation step, you have a couple of choices. One idea
to use Tukey's test for honestly significant difference (again per-
but without any other additional correction for multiple testing) to
which groups differs in mean expression.
One of the things we've tried recently (see Nakamura et al, Clin
Res, 2007) is to cluster the significant genes and compute average
expression profiles for each cluster to see which of the effects are
driving the expression.
James Anderson wrote:
> I have affy data for time and treatment, I did a two way anova for
each gene. What's the correct way to evaluate whether time effect is
more imporant or treatment effect is more important? I think averaging
the variance component of each gene is not a very good way, since not
all genes are equally important, in addition, most of the genes are
noisy genes. I am thinking of doing a PCA and take the first several
components, instead of doing anova on each gene, I can do two way
anova on the scores of the first several component, then I can use a
weighted average (weight is determined by the amount of variance each
PC captured), is this a good way? Can somebody give me some
suggestions on this? Thanks.
> Luggage? GPS? Comic books?
> [[alternative HTML version deleted]]
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch
> Search the archives: