[BioC] limma roast syntax for overall anova

[Gordon K Smyth]

Dear Juliet,

On Thu Jul 3 21:09:46 CEST 2014, Juliet Hannah wrote:

> All,
>
> I am testing an overall anova (4 groups; is there any difference across
> groups) using the following design matrix in limma
>
>  (Intercept) B C D
> 1           1 0 0 0
> 2           1 0 0 0
> 3           1 0 0 0
> 4           1 0 0 0
> 5           1 0 0 0
> 6           1 0 0 0

Mmm, the columns seem to be all zero except for the first.  I assume that 
you are actually just giving the first 6 rows of a larger matrix.

> I have used ROAST (in limma) in the past with a specific contrast, such as
>
> roast(expression_matrix,index,design_matrix,contrast=10,nrot=3000).
>
> In the current case, I didn't need any contrasts. In limma, I had used
>
> design <- model.matrix(~key$stage)
> colnames(design)[2:4] = c("B","C","D")
> fit <- lmFit(e, design)
> fit <- eBayes(fit)
> topTable(fit,adjust.method="BH")
>
> to obtain the overall F-test

To get the overall anova F-test you need:

   topTable(fit[,-1])

or alternatively

   topTable(fit, coef=2:4)

which gives the same result.  Note that the intercept needs to be removed 
from the test, as it always is in classical anova, and that's what the -1 
does.

> (I understand this test is not that interesting, but it is requested of 
> me).
>
> Using this setup, how can I specify this overall test within ROAST.

That's a good question.  We haven't implemented F-tests for ROAST.  If we 
get enough requests, we will consider doing so, but we haven't so far.

If the meantime, you can input the F-statistics to the geneSetTest() 
function:

   fit2 <- fit[,-1]
   geneSetTest(index, fit2$F)

etc.

Best wishes
Gordon


> Thanks for your help.
>
> Juliet
>

______________________________________________________________________
The information in this email is confidential and intend...{{dropped:4}}