How  can i apply moderated t-test in Anova as ebayes function used in limma?

Please Help

Thanks

Following up on James' comment, the best example is in Section 9.3 of the user's guide. This does a moderated F-test for the ANOVA-style comparison, rather than a t-test. Generally speaking, we need to:

1. Use makeContrasts to define all of the comparisons you're interested in. For an ANOVA-style test for equality across multiple groups, we can define all pairwise comparisons between groups (note that we don't really need all pairwise comparisons, but this approach gives us all of the relevant log-fold changes in the downstream results, so we might as well do it).
2. Use contrasts.fit to re-fit the linear model according to the contrast matrix from step 1.
3. Use eBayes to perform empirical Bayes shrinkage on the refitted model from step 2.
4. Use topTable without any coef specified (i.e., set to NULL). This will do an ANOVA-style test for all of the comparisons at once.

If you need more advice, you'll have to give specific details about your samples, experimental design and comparisons of interest.

limma easily does anova-style F-tests, but with empirical Bayes moderation. You can use in limma on any model formula that you would use for the anova() function (except for those with random factors).

Suppose you have a matrix of expression values y and an experimental factor group. To do a oneway anova for the first gene, you might use:

out <- anova(y[1,] ~ group)
summary(out)

In limma you do this:

design <- model.matrix(~group)
fit <- lmFit(y, design)
fit <- eBayes(fit)
topTable(fit)

This does a moderated oneway anova for every gene in your dataset, not just one.

There is absolutely no way to use anova() to do an empirical Bayes analysis like limma does. anova() does a univariate analysis for one gene only, and empirical Bayes analysis is not possible for just one gene.

Hi,

Sorry if I am confusing your question but I think you're asking how to calculate a moderated t-statistic without using limma (for high-dimensional data)? I think you have to write your own function or just simply use limma.

If, for whatever reason, you do not want to use limma:

Here's how I would approach this but there may be a more efficient way someone else can point out:

(0) How many samples do you have? If you have a lot then you can just do a t-test.

(1) First take a look at Gorfon Smyth's paper on the moderated t statistic (formula on page 8): http://www.statsci.org/smyth/pubs/ebayes.pdf

(2) Using the notation in the paper, there are two quantities needed for the moderated t-test: d_{0} and s_{0}, the prior d.o.f and standard deviation. So if you download the limma package, there is a function called "squeezeVar()" (I think there's a file in the R directory called "squeezeVar.R") that returns both d_{0} and s_{0}.

So I would create a new function, fit your linear model,  use squeezeVar() to calculate s_{0} and d_{0} and then calculate the moderated t-statistic.

As u know limma and Anova are two different approaches for gene analysis. I want to ask that in limma we can perform moderated t-test using ebayes function as it take parameter of model created by lm.fit but we use lm function for creating model in anova and the model created by annova does not work in ebayes. So can we apply moderated t-test in annova or what is the test name for anova similiar to ebays in limma or what is the best test for annova type model.

Thanks for your answers please guide in this.