[BioC] RRBS question
gilgi.friedlander at weizmann.ac.il
Wed Jan 22 21:01:24 CET 2014
Thank you so much for all the help and important information!
From: Alex Gutteridge [mailto:alexg at ruggedtextile.com]
Sent: Wednesday, January 22, 2014 7:06 PM
To: Kasper Daniel Hansen
Cc: Gilgi Friedlander; bioconductor at r-project.org
Subject: Re: [BioC] RRBS question
Yes, this is exactly what I found. I ended up tweaking maxGap to 10kbp which seemed to give a reasonable result over most regions.
On 22.01.2014 16:29, Kasper Daniel Hansen wrote:
> Because RRBS data is non-contiguous, you basically have to use the
> 'maxGap' argument to do the smoothing on each of the contiguous
> groups. You probably also have to use 'local.correct=FALSE' as the
> algorithm is currently written. I have not yet experience with
> applying the algorithm to RRBS data myself.
> On Wed, Jan 22, 2014 at 5:15 AM, Alex Gutteridge wrote:
>> On 21.01.2014 20:48, Gilgi Friedlander wrote:
>>> Hi Alex,
>>> Sorry to bother you, but I have a question regarding RRBS analysis.
>>> I saw your post from last year, and also was wondering if bsmooth
>>> can work well for RRBS data, as going to work on such data.
>>> I wanted to reply to the post on the mailing list, but didn't see
>>> such an option.
>>> If you already have results, and can share your experience if one
>>> can use bsmooth for RRBS, it will be great. And if not, if you have
>>> recommendations for other tools.
>>> Thanks a lot,
>> cc'ing the list for future reference and in case anyone else has
>> additional insights on RRBS analysis with Bioconductor:
>> Hi Gilgi,
>> I had mixed results with bsmooth and rrbs data when I tried it.
>> Certainly the various smoothing parameters required tweaking before
>> it would run successfully. I can let you know what worked for me, but
>> I'd be lying if I said I arrived at them through careful
>> experimentation. One also has to be cautious I think about the
>> overall approach of whether smoothing makes sense when the data
>> rrbs is by definition non-contiguous. That said, bsmooth did run
>> did detect differential methylated regions for us which looked
>> correct on deeper inspection. What bsmooth (by design I guess) does
>> not detect are the single CpGs that seem to change between
>> experimental groups, but sit within larger regions that clearly do
>> not change. It's still an open question (in my mind at least) whether
>> such sites are likely to be biologically significant or
>> Alex Gutteridge
>> Bioconductor mailing list
>> Bioconductor at r-project.org 
>> https://stat.ethz.ch/mailman/listinfo/bioconductor  Search the
>  mailto:Bioconductor at r-project.org
>  https://stat.ethz.ch/mailman/listinfo/bioconductor
>  http://news.gmane.org/gmane.science.biology.informatics.conductor
>  mailto:alexg at ruggedtextile.com
More information about the Bioconductor