Repeated Measures Anova with Limma, directionality of covariates.
1
0
Entering edit mode
kodream ▴ 20
@kodream-6952
Last seen 7.0 years ago
United States

Are repeated measures anova with limma(IE using duplicate correlation and a blocking variable and a measurement covariate) directional?  Will it automatically test for hypothesis where it is expected that the measurement variables are negatively correlated?  Or does it only test the hypothesis that means of subsequent measurements are not the same?  Is there anyway to do the former without repeating the analysis with the signs flipped on one of the measurements(with two measurements).

limma • 1.9k views
ADD COMMENT
0
Entering edit mode
@gordon-smyth
Last seen 55 minutes ago
WEHI, Melbourne, Australia

I don't understand your question. But limma always allows both positive and negative correlations. There is no situation where you would need to repeat an analysis with signs flipped.

ADD COMMENT
0
Entering edit mode

Confusion is contagious :)

I've been thinking about it some, and maybe this will get a better answer.

Suppose instead of a design matrix like such:

~Group+Measures

I did 

~Group:Measures

Then I could test various contrasts between the combinations of measures and groups

such as

makeContrasts(GroupA:MeasureOne+GroupA:MeasureTwo+GroupB:MeasureOne-GroupB:MeasureTwo,design)

Now what I wonder is does duplicateCorrelation use these designs of the ~Group:Measure variety?

Also, what is my null hypothesis when my alternate hypothesis is

GroupA:MeasureOne is not correlated or positively correlated with GroupA:MesureTwo but GroupB:MeasureOne is negative correlated with GroupB:MeasureTwo?

ADD REPLY
0
Entering edit mode

duplicateCorrelation will use whatever design matrix you supply to it. You must decide whether the interaction model (i.e., ~Group*Measures, or something equivalent with the : operator) is more appropriate than the additive model (i.e., ~Group+Measures) for your circumstances. If so, then you should use the former throughout your entire analysis. In fact, you could make things a lot simpler by using a one-way layout with the combinations:

Combination <- factor(paste(Group, ".", Measures))
design <- model.matrix(~0 + Combination)

As for your second question, I'm not sure what you're trying to do. Why do you want to deal with correlations between coefficients? For each gene, you've only got one estimate of each coefficient, so it'd be like trying to figure out the correlation from a single point. Perhaps you could give this a bit more biological context, so we know the relevant question that you're trying to answer.

ADD REPLY

Login before adding your answer.

Traffic: 656 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