Dear Gordon / list,

I would like to compare the correlation between 3 treatments that are compared to the same reference group using the `genas()`

function present in limma. However, I also used the function `arrayweights()`

to account for the "consistency" of each array within each treatment group.

In the help page for `genas()`

it is stated that "it [`genas()`

] only works with fit objects that were created without observation weights or missing values."

Although `arrayweights()`

is array-based and not observation-based, I would like to be sure cq. have it confirmed it is OK to combine `arrayweights()`

and `genas()`

. If so, would it then be best to use "predFC" to select the genes used for the correlation analysis, or does the gene selection criterion not really matter ?

Thanks,

Guido

I was looking to comb through the link provided for Belinda Phipson's thesis for a more detailed explanation of the GENAS procedure, but seems like this is not accessible?

Are there tidbits of it we can find online (perhaps it was presented somehwere(?)) where we can learn more, or?

No, we haven't made any write-up of the method public yet. We need to do that. If it helps, the method is based on a multivariate generalization of the limma eBayes variance estimation algorithm.

Any updates on a genas version compatible with voom?

No, genas() is still not available with voom(). It has turned out to be a bigger job than we expected.

In the meantime, we are switching to limma-trend instead of voom() when we want to use genas().