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

[Gordon K Smyth]

Dear Sean,

The fact that you get a warning message does not mean that the function 
hasn't worked satisfactorily.  It is mainly to alert you to an unusual 
situation.  The warnings affect only 15 tags out of thousands, so are 
unlikely to have any influence on your results.

The main issue is that there is no biological dispersion in your data. 
My guess is that you have computed offsets in such as way that they soak 
up all the biological variation in your data, and there is no 
Poisson-overdispersion left to estimate.  Your data could even now be 
underdispersed relative to Poisson, once the offsets are accounted for. 
This suggests to me that your offsets are over-fitted in some sense, and 
that statistical analysis using these offsets might not be meaningful.

Have you tried a more standard analysis of your data, using the built-in 
normalization procedures in edgeR?

Best wishes
Gordon

> Date: Wed, 20 Jul 2011 18:17:42 -0700
> From: Sean Ruddy <sruddy17 at gmail.com>
> To: bioc-sig-sequencing at r-project.org
> Subject: [Bioc-sig-seq] edgeR Warning Messages from
> 	maximizeInterpolant	called from estimateGLMTagwiseDisp
>
> Hi,
>
> I'm trying to get tagwise estimates from estimateGLMTagwiseDsip with the
> argument "method = 'common' " and 15 warning messages each of which are one
> of 2 flavors:
>
> 9: In maximizeInterpolant(spline.pts, apl.smooth[j, ]) : Divergence
> 10: In maximizeInterpolant(spline.pts, apl.smooth[j, ]) : max iterations
> exceeded
>
> After looking at the functions "dispCoxReidInterpolateTagwise" and
> "maximizeInterpolant", I believe that each of these warnings correspond to a
> specific tag; correct me if I'm wrong. I'd like to find out which tags are
> creating this warning so I can get an idea of what type of data is causing
> this. I've tried debug() but that information isn't passed to the functions.
> In particular, the part of the "dispCoxReid..." function below
>
> for (j in 1:ntags) d[j] <- maximizeInterpolant(spline.pts, apl.smooth[j, ])
>
> I tried changing the functions internally to pass that information and
> output it but it doesn't seem to work. The changes don't have any effect. Is
> there an feasible way of figuring out the tags that cause the warnings? I'm
> out of ideas...
>
> Also, the common dispersion estimate is very low -- 1e-08 neighborhood --
> which I'm guessing is one main reason why maximizeInterpolant is having
> trouble. On that note, I have no idea why the common dispersion value is so
> low. I do not have technical replicates and the replicates I do have are
> hardly biological. I would expect to get a higher dispersion estimate, but
> it's somewhat difficult to tell since I also have an offset value for each
> tag and sample and thousands of tags.
>
> Any help is appreciated!
>
> Thanks!
> Sean

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