"Random variance t-test" is a pretty misleading name, since it's actually not related to things like var.test. The paper describes something quite similar to what limma does:
"...it is assumed that the variance of the residuals change from gene to gene, but represent random selections from a single distribution. ... By sharing the variance estimate across multiple genes, we can form a better estimate for the true residual variance of a given gene and effectively boost the residual degrees of freedom.
Basically, they fit an inverse gamma function to the residuals and get the variance used in the t-test from that. I'm sure that name has mislead more than a few people!
• 0 views
•
link
written
by
Devon Ryan
Well, I don't know if it is useful for you, but I use the package
limmathat does a moderated t-test, taking into account standard variations across the sample in order to overcome limited replication.Just to add to this, limma also does information sharing between genes/probes/whatever for variance calculation. Having not gone through the paper thoroughly, though, I can't say how similar its method is to that used by limma.
I use limma too. But the t-test in the package lets me choose either variance which is equal or unequal, not random. I was looking for a package which can do that for me. I know BRBtools does it. So I wanted to know if there is an equivalent in R.
I think you might have misinterpreted the BRB Tools variance model. The random variance model that BRBtools uses is between genes, not between samples, and it is closely analogous to what limma does. Both tools allow variances to be different between genes but the same for samples within each gene. BRB Tools doesn't give any extra flexibility not available in limma.