[Bioc-sig-seq] edgeR Warning Messages from maximizeInterpolant called from estimateGLMTagwiseDisp

[Gordon K Smyth]

Hi Sean,

No, we do not provide a means for users to pass arguments to 
maximizeInterpolant() and, I have no plans to permit this.  My feeling is 
that the iteration should provide a reasonable value within ten 
iterations, or else there is something wrong and continuing the iteration 
will not improve the situation.  For example the requested accuracy of 8 
significant figures might not be obtainable in machine precision, in which 
case the final iteration is as good as can be achieved.  Your data appears 
to be underdispersed, so the estimation of tagwise dispersions is in a 
sense degenerate.  Since all the dispersions will be more or less 1e-8, it 
doesn't seem very relevant whether each value iteration has converged to 8 
significant places.  This numerical accuracy might not be obtainable in 
computer precision for such small dispersions.

Best wishes
Gordon

---------------------------------------------
Professor Gordon K Smyth,
Bioinformatics Division,
Walter and Eliza Hall Institute of Medical Research,
1G Royal Parade, Parkville, Vic 3052, Australia.
Tel: (03) 9345 2326, Fax (03) 9347 0852,
smyth at wehi.edu.au
http://www.wehi.edu.au
http://www.statsci.org/smyth

On Fri, 22 Jul 2011, Sean Ruddy wrote:

> Hi Gordon,
>
> In addition to Kasper's email, is it possible to pass arguments to
> maximizeInterpolant, for example to increase the max iterations? It appears
> you can send additional arguments to dispCoxReidInerpolateTagwise but not to
> maximizeInterpolant.
>
> Thanks,
> Sean
>
>
> On Fri, Jul 22, 2011 at 6:39 AM, Kasper Daniel Hansen <
> kasperdanielhansen at gmail.com> wrote:
>
>> On Fri, Jul 22, 2011 at 9:31 AM, Sean Ruddy <sruddy17 at gmail.com> wrote:
>>> Hi Gordon,
>>>
>>> Thanks for the reply. That's good to know I shouldn't be worried about 
>>> the results, but then again I'm not really sure, as you say, if I 
>>> should even go down this road. Using the standard edgeR normalization 
>>> would also be meaningless in my case unfortunately. I have two data 
>>> sets of counts each belonging to different organisms. One data set 
>>> runs smoothly and the other has all these problems yet the experiments 
>>> done on both were exactly the same. Kind of puzzling and unfortunate 
>>> but I will investigate more into what you're saying about the offsets 
>>> and see if I can't adjust my strategy. Thanks for the insights!
>>
>> While I agree with Gordon that having warnings in a few genes are
>> hardly problematic, it would perhaps make sense to make it easy for
>> the user to identify exactly which genes (rows) have warnings
>> associated with them,  Having been bitten by convergence problems in
>> the past, I believe it is always prudent to check the data. Just a
>> suggestion, and I apoligize if this is already (easily) possible.
>>
>> Kasper
>>
>

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