Testing for logFC significantly larger than threshold; or: calculating confidence intervals in limma

Dear all,

normally, one tests whether a difference between the experimental
conditions (expressed as logFC) is significantly different from 0.
Thus, limma provides a p-value (or adjusted p-value) for the test
where the H_0 is that logFC == 0, and H_alt is that logFC =/= 0.

I would like, however, to test the hypothesis abs( logFC ) <= T and
the alternative abs( logFC ) > T, where T is an arbitrary threshold.
For example, T=1 for genes for which the change of expression is
significantly more than twofold, either way.

Note that this is not the same as choosing genes for which abs( lfc )
> 1 and adj.P.Val < 0.05. A gene might have an estimated logFC over 1,
and the logFC might be significantly different from 0, but it can have
a variance large enough for the logFC not be significantly higher than
1.

An alternative is to calculate the 0.05 confidence interval for each
estimated logFC, since that automatically gives the answer that I'm
looking for.

Clearly, I can do the analysis myself, for example calculating the
regular t-statistic, but I would like to take advantage of the
moderated t-statistic present in limma as well as all the facilities
for creating complex contrasts.

Kind regards,

January

--

-- 
-------- January Weiner --------------------------------------

_______________________________________________
Bioconductor mailing list
Bioconductor@...
https://stat.ethz.ch/mailman/listinfo/bioconductor
Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor