[BioC] Missing Autocorrelation in Ringo

[Torsten Waldminghaus]

Hi Joern,

I did do some updating. This is how it looks like now:

> sessionInfo()
R version 2.9.1 (2009-06-26) 
i386-pc-mingw32 

locale:
LC_COLLATE=Norwegian (Bokmål)_Norway.1252;LC_CTYPE=Norwegian (Bokmål)_Norway.1252;LC_MONETARY=Norwegian (Bokmål)_Norway.1252;LC_NUMERIC=C;LC_TIME=Norwegian (Bokmål)_Norway.1252

attached base packages:
[1] grid      stats     graphics  grDevices datasets  utils     methods   base     

other attached packages:
[1] Ringo_1.9.5        Matrix_0.999375-29 lattice_0.17-25    limma_2.18.2       RColorBrewer_1.0-2
[6] Biobase_2.4.1     

loaded via a namespace (and not attached):
[1] annotate_1.22.0     AnnotationDbi_1.6.1 DBI_0.2-4           genefilter_1.24.2   RSQLite_0.7-1      
[6] splines_2.9.1       survival_2.35-4     xtable_1.5-5       


I coul now also find and use the extractProbeAnno function but get this:

> test<-extractProbeAnno(RG)
Creating probeAnno mapping for chromosome Error in split.default(probeindex, as.factor(hits[[chrNameColumn]])) : 
  Group length is 0 but data length > 0
In addition: Warning message:
In extractProbeAnno(RG) :
  Some reporters had no or an unrecognized genome position in RG$genes$SystematicName.


You said that the function works with an RGlist object if it" encodes the probe position in a colum called "SystematicName"". This is the case but it looks like this: 

> RG$genes$SystematicName[100]
[1] "CGH_U00096_943375_943434_a"

Is this what you ment?

I also tried the autocorraltion again as before but did get the same strange result. Could it be that it is a problem that the data in the RG file are not ordered? I had problems with that when I used some other peak detection because the program looked just at the next entry without considering the probe position. My DNA fragments where 1000bp and smaller with a peak at 500bp.

About the judging of quality of found peaks, could you explain what " resort to visualizations" means and how I would do this practically.

Many thanks,
Torsten

 




-----Original Message-----
From: Joern Toedling [mailto:Joern.Toedling at curie.fr] 
Sent: 30. juni 2009 16:40
To: Torsten Waldminghaus; bioconductor at stat.math.ethz.ch
Subject: Re: [BioC] Missing Autocorrelation in Ringo

Hi Torsten,

I suggest that you first update your R and Bioconductor installation as you
are using very out-dated versions of both (R is currently 2.9.0 and for Ringo
please use the current development version (1.9.5), which can be downloaded as
a binary from here:
http://www.bioconductor.org/packages/devel/bioc/html/Ringo.html
). In the current version of Ringo, you will also find the function
extractProbeAnno. And indeed such a low degree of autocorrelation is
unexpected. What was the length distribution of fragments after sonication?
Typically you would expect some auto-correlation at least up to that length.
This factor is also important for setting the sliding-window width for
smoothing the signal prior to peak detection. However, your data objects look
alright, so I am afraid that I do not immediately see the problem.

For judging the quality of the found "peaks", you can resort to visualizations
of a.) the highest-ranking peaks b.) the lowest-ranking peaks c.) positive
control regions where you would expect peaks d.) negative control-regions
where no enrichment is to be expected.
Usually the "null distribution" of smoothed probe levels under non-enrichment
can only roughly be estimated from the data based on certain simplifying
assumptions, so I would not lend too much confidence to p-values and FDRs from
such estimates.

Regards,
Joern


On Tue, 30 Jun 2009 14:04:31 +0200, Torsten Waldminghaus wrote
> Hi,
> 
> thanks for the answer and sorry for not making it more clear. I get 
> a plot but it suggests that there is no autocorrelation. This means 
> the column at point "0" is at 1.0 as usual but at "100", "200",... 
> there are only dots which vary a bit in their size. Now the results 
> you were interested in:
> 
> > str(exAc)
> Class 'autocor.result'  atomic [1:11]  1.00000 -0.00469 -0.00660 
>  0.00538  0.00652 ...  ..- attr(*, "chromosome")= chr "1"
> 
> > ls(annoObject)
> [1] "1.end"    "1.index"  "1.start"  "1.unique"
> 
> > head(annoObject["1.start"])
> [1]  68 154 189 294 365 440
> 
> I tried the function extractProbeAnno but it does actually not seem 
> to be there. R did not find it and I did also not find a help entry 
> for it:
> 
> > help(extractProbeAnno)
> No documentation for 'extractProbeAnno' in specified packages and libraries:
> you could try 'help.search("extractProbeAnno")'
> 
> However, I could use for example posToProbeAnno and find the 
> corresponding help page?!
> 
> Here is the session info:
> 
> > sessionInfo()
> R version 2.7.2 (2008-08-25) 
> i386-pc-mingw32
> 
> locale:
> LC_COLLATE=Norwegian (Bokmål)_Norway.1252;LC_CTYPE=Norwegian (Bokmål)
> _Norway.1252;LC_MONETARY=Norwegian (Bokmål)
> _Norway.1252;LC_NUMERIC=C;LC_TIME=Norwegian (Bokmål)_Norway.1252
> 
> attached base packages:
> [1] splines   tools     stats     graphics  grDevices utils    
>  datasets  methods   base
> 
> other attached packages:
>  [1] Ringo_1.4.0          SparseM_0.78         RColorBrewer_1.0-2  
>  vsn_3.6..0            affy_1.18.2          [6] preprocessCore_1.2.1 
> affyio_1.8.1         geneplotter_1.18.0   annotate_1.18.0      
> xtable_1.5-4        [11] AnnotationDbi_1.2.2  RSQLite_0.7-1        
> DBI_0.2-4            lattice_0.17-13      genefilter_1.20.1   [16] 
> survival_2.34-1      Biobase_2.0.1        limma_2.14.7
> 
> loaded via a namespace (and not attached):
> [1] grid_2.7.2         KernSmooth_2.22-22
> 
> Beyond the problems with getting the autocorrelation I was wondering 
> if there is a good way to judge the quality of Ringo peak detection 
> or the probability of getting wrong hits or loosing some?
> 
> Thanks for any help,
> Torsten