Re: [Useq-users] artefactual peaks

[David A. N. <dav...@gm...>]

Yes this is a common problem and why chIPSeq detectors always need an input control.  Better yet, compare two chIPSeq samples.  The problem is with over dispersion.  With high read counts, minor changes in the count #'s between t and c will return a very significant p-value with a binomial test.  Thus the popularity of a log2ratio threshold as well as a p-value/ FDR threshold.  Better to use a negative binomial test with a robust estimation of the over dispersion.  This necessitates replicas though, at least three for the chIP and input samples.  That said my incorporation of the DESeq package into the USeq MultipleReplicaScanSeqs app didn't really solve the problem.  ChIPSeq samples have such high variability that DESeq returns very few, if any significantly enriched regions.  So my current recommendations are to run the standard ScanSeqs and MultipleReplicaScanSeqs as well as the Liu lab's MACS package.  The latter does some clever things to estimate the expect/ lambda at each location instead of using a genome average.

-cheers, D


On Dec 16, 2013, at 5:25 PM, Noboru Jo Sakabe <ns...@uc...> wrote:

>    Hi David, analyzing a sample from my lab that seems to be a failed IP, I found a number of regions in the genome that have many reads, but are not real peaks. They are basically the same across samples, including inputs and failed IPs. One example region is mm9 chrX:166,393,669-166,477,668 (figure attached).
>    This is not a problem exclusive to USeq, but I was wondering if there's something that can be done to improve peak callers, since inputs are also "enriched" in these regions. I know that people have generated files containing regions that should be masked, but since input samples are also enriched, I wonder why they still come up as peaks.
>    Thank you.
> 
> <fig.png><nsakabe.vcf>------------------------------------------------------------------------------
> Rapidly troubleshoot problems before they affect your business. Most IT 
> organizations don't have a clear picture of how application performance 
> affects their revenue. With AppDynamics, you get 100% visibility into your 
> Java,.NET, & PHP application. Start your 15-day FREE TRIAL of AppDynamics Pro!
> http://pubads.g.doubleclick.net/gampad/clk?id=84349831&iu=/4140/ostg.clktrk_______________________________________________
> Useq-users mailing list
> Use...@li...
> https://lists.sourceforge.net/lists/listinfo/useq-users