[BioC] Questions about limma

Sean Davis sdavis2 at mail.nih.gov
Wed Apr 6 21:24:46 CEST 2005

On Apr 6, 2005, at 1:44 PM, He, Yiwen (NIH/CIT) wrote:

> Hi Sean,
> Thanks for your help!
>  lmFit didn't work on my log ratio data because the data was not 
> numeric, so I got all NAs for the result. It's now working after I 
> converted it to numeric.

Glad to hear it.

> I did read the User's guide and saw those case studies. However, I was 
> not able to find any test data that I can load into R to run the test. 
> Can you point me to the right place?

You are right.  I spoke too quickly--it doesn't look like the example 
files are included--sorry about that!  It looks like they are available 


> I have a general question: How is limma compared with other 
> statistical analysis like SAM? I ran both procedures on the same 
> dataset (one class) and got similar results (28 out of the 30 top 
> genes overlap, orders differ slightly). But the significant levels 
> (p/B for limma and q for SAM) differ. For small sample size, SAM's q 
> is bigger than limma's p values, on the other hand, when sample size 
> is large, q is much smaller. I understand limma uses empirical bayes 
> mothods so that it works on small number of replicates. What are the 
> other advantages of limma?

As for p/q values, they will be different--are you using fdr multiple 
comparisons correction?  I don't generally use siggenes much, though, 
so I don't know how they compare in practice.  As for other advantages 
of limma, it does linear modeling of experiments, so many complex 
experimental designs can be handled in the same framework.


More information about the Bioconductor mailing list