[BioC] GLMs in DESeq

[Simon Anders]

Hi Seanna

On 03/08/2011 05:13 PM, Seanna McTaggart wrote:
> I am using DESeq to determine DE of an RNA-Seq project that has
> multiple genotypes exposed to different treatments. I would like to
> use the GLM functionality to partition the variance in the count data
> between genotype and treatment. However, when I follow the
> suggestions of Simon (Bioconductor Digest, Vol 96, Issue 9), neither
> model (fit0 or fit1) is reaching convergence, and I was wondering if
> it was possible for me to increase the number of iterations to see if
> this would help out. Otherwise I would appreciate any advice on how
> to proceed.

I've just added a new argument to the function 'nbinomTestGLM', called 
'glmControl', which is a list of GLM control parameters as described in 
the 'glm.control' help page.

So, you could try something like

fit0 <- nbinomTestGLM( cds, count ~ whatever,
    glmControl = list( maxit=75 ) )

This should increase the maximum number of iterations from 25 to 75.

However, it is well possible that this does not help much. If your model 
fails to converge for any gene, it might be a deeper problem, and if it 
fails to converge only for a few genes, these might be somehow 
"stubborn" and won't converge at all, no matter how many iterations you 
wait. If necessary, ask again with more details.

   Simon