Are the logFC confidence intervals returned by topTable adjusted for multiple comparisons?
1
0
Entering edit mode
@calin-jageman-robert-6431
Last seen 7.8 years ago

I'm using limma to analyze both microarray and qPCR data. For logFC values, I'd like to report CIs. I found that this feature was added to topTable around 2011 (https://stat.ethz.ch/pipermail/bioconductor/2011-January/037579.html)

I'm wondering, though: are the CIs produced by topTable expanded for the multiple comparison method selected? Or are these uncorrected/standard CIs for each transcript? For reference, I used:

fit <- lmFit(qpcr_1hr)
fit2 <- eBayes(fit)
regulated <- topTable(fit2, adjust="BH", confint=TRUE,number=nrow(fit2))

So I was expected the CIs to also be adjusted based on Benjamini- Hochberg (as discussed in Jung, Friede, and Biebarth, 2011 - http://www.biomedcentral.com/1471-2105/12/288 ). But looking at the CIs I'm getting, they seem more likely to be uncorrected.

Does anyone know for sure?

Thanks,
Bob

-------------------
Bob Calin-Jageman
Neuroscience Program Director
Associate Professor
Department of Psychology
210 Parmer Hall
Dominican University
7900 West Division
River Forest, IL 60305
708.524.6581
rcalinjageman at dom.edu

Microarray qPCR limma • 1.9k views
ADD COMMENT
1
Entering edit mode
@gordon-smyth
Last seen 3 minutes ago
WEHI, Melbourne, Australia

limma reports ordinary confidence intervals: they are the same as what would be returned by any linear regression software except they take advantage of limma's empirical Bayes squeezing of the standard errors.

This seems to me to be what most users will want and expect. The adjustment method used by topTable is documented to affect the p-values rather than the CIs. The CIs are in a sense already adjusted for multiple testing, in that the specified proportion of CIs will cover the true logFC values across all genes and contrasts. No special adjustment is required.

The paper by Jung et al (2011) proposes a special adjustment that applies to the CIs of significantly differentially expressed (DE) genes only. This is not the traditional purpose of CIs, which are supposed to apply to all effects whether significantly nonzero or not. In limma, the CIs are intended to apply to all genes, whether DE or not. So the adjustment method of Jung et al (2011) doesn't seem what is required.

ADD COMMENT

Login before adding your answer.

Traffic: 620 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6