[BioC] EdgeR GOF plots

Gordon K Smyth smyth at wehi.EDU.AU
Sun Feb 2 02:15:08 CET 2014


Dear Adriaan,

Having genewise deviances a little below the Q-Q line at the low end is 
not a concern.  It is deviations above the line at the high end that would 
be a concern.

Note that you are interpreting this phenomenon the wrong way around. 
Having deviances below the Q-Q line indicates in principle that variances 
are being over-estimated, not underestimated.  The plot is showing that 
the actual variability in the data is lower than that predicted by the 
model.  However the asymptotic chisquare approximation to the glm residual 
deviance is far from perfect, and imperfections in the plot at the low end 
are more likely to arise from this than from mis-estimation of the 
dispersions.

Best wishes
Gordon

> Date: Thu, 30 Jan 2014 17:56:26 +0000
> From: Adriaan Sticker <adriaan.sticker at gmail.com>
> To: bioconductor at r-project.org
> Subject: [BioC] EdgeR GOF plots
>
> Dear all,
>
> I'm doing an EdgeR analysis and made some GOF plots after estimating
> geneewise dispersion and fiting my data. I noticed that my values in the
> first half of my plot are consistently a bit lower then theorethically
> expected under the chisquared distribution. I'm trying to understand how
> this comes. I guess it has something to do with underestimating the
> variance because of the squeezing of the genewise dispersion towards the
> trended dispersion? But why underestimating and why at the lower variances?
> I don't see this in the gof plots in the authors paper ([McCarthy
> 2012][1]). And does this have some implications on my analysis?  I've put a
> url to the gof plot and the BCV plotbelow.
>
> Somebody cares to shed some light on this matter for this statistical
> genomics novice? :)
>
> Best regards
>
> GOF plot:
> http://s16.postimg.org/dtldp2o85/Rplot.png<http://s16.postimg.org/dtldp2o85/Rplot.png>
> BCV plot:
> http://s29.postimg.org/qnzg7e0w7/Rplot01.png
>
>  [1]:
> http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=3378882&tool=pmcentrez&rendertype=abstract

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



More information about the Bioconductor mailing list