[BioC] bumphunter DMRs difficulties

James W. MacDonald jmacdon at uw.edu
Mon Feb 24 15:37:20 CET 2014


Hi Giovanni,

Tim advised you to do the exact opposite. If you set B = 100, then the 
smallest p-value you can achieve is 0.01, which as he says, 'sucks'. If 
you use B = 1000, the smallest p-value will be 0.001 which is marginally 
better.

Best,

Jim


On 2/24/2014 5:46 AM, Giovanni Calice wrote:
> Hi Tim,
> I'm already in the order of ideas to set the n of resamples below 
> about 100 as you say.
> I await news about your comments for a better marginal test statistic and
> the reference discussion about this topic.
>
> For now I'll follow your advice and previous ones.
>
> Thank you very much.
>
> Best,
> Giovanni
>
>
> Laboratory of Preclinical and Translational Research
> IRCCS - CROB Oncology Referral Center of Basilicata
> Rionero in Vulture - Basilicata - Italy
>
>
> 2014-02-23 5:47 GMT+01:00 Tim Triche, Jr. <tim.triche at gmail.com 
> <mailto:tim.triche at gmail.com>>:
>
>     Don't do that.  B is the number of permutations.  Your
>     significance estimates will suck if you set it below about 100.
>      If you can wait (a day or so) for 1000 or more permutations, they
>     will be quite good.  I have some ideas for a better marginal test
>     statistic to capture the omnibus null of "no difference" against
>     any reasonable alternative, but that's a topic for another
>     discussion.  It won't change the fact that you need a lot of
>     permutations to get solid significance estimates from a biased
>     sample (450k chip).
>
>     Best,
>
>
>     Statistics is the grammar of science.
>     Karl Pearson <http://en.wikipedia.org/wiki/The_Grammar_of_Science>
>
>
>     On Sat, Feb 22, 2014 at 10:02 AM, Giovanni Calice
>     <giovcalice at gmail.com <mailto:giovcalice at gmail.com>> wrote:
>
>         Hi,
>
>         I missed to ask advice about B parameter.
>         In my first bumphunter runs, I set B=1 (default setting)
>         having thus
>         results within a reasonable time.
>
>         Now, as mentioned in the previous mail, I set B=1000 according
>         to the reference
>         manual and
>         I thought so in order to have a statistically significant results.
>
>         In my next bumphunter call, have I to change this setting
>         (B=1000)?
>
>         Thanks,
>         Giovanni
>
>
>         Laboratory of Preclinical and Translational Research
>         IRCCS - CROB Oncology Referral Center of Basilicata
>         Rionero in Vulture - Basilicata - Italy
>
>
>         2014-02-21 18:53 GMT+01:00 Giovanni Calice
>         <giovcalice at gmail.com <mailto:giovcalice at gmail.com>>:
>
>         > Hi Jim,
>         >
>         > Ok!
>         > My next bumphunter call will be on 1/3 of total ncores (I
>         share compute
>         > cycles with co-workers too)
>         > with cutoff and cutoffQ = 0.95.
>         >
>         > Many Thanks.
>         >
>         > Best,
>         > Giovanni
>         >
>         > Laboratory of Preclinical and Translational Research
>         > IRCCS - CROB Oncology Referral Center of Basilicata
>         > Rionero in Vulture - Basilicata - Italy
>         >
>         >
>         >
>         > 2014-02-21 18:29 GMT+01:00 James W. MacDonald
>         <jmacdon at uw.edu <mailto:jmacdon at uw.edu>>:
>         >
>         > Hi Giovanni,
>         >>
>         >> I don't know about the number of cores. I share compute
>         cycles with
>         >> co-workers, so can't just hog them all up. A recent
>         bumphunter run I did on
>         >> our shared server (a single multi-core machine with I
>         believe 32
>         >> processors) was:
>         >>
>         >> library(doParallel)
>         >> registerDoParallel(cores = 12)
>         >> bmps.main <- bumphunter(eset, design.cd <http://design.cd>,
>         cutoff = 0.99)
>         >>
>         >> Which resulted in something like 2500 bumps and finished
>         in, I don't
>         >> know, say 15-20 minutes?
>         >>
>         >> Best,
>         >>
>         >> Jim
>         >>
>         >>
>         >>
>         >> On 2/21/2014 12:21 PM, Giovanni Calice wrote:
>         >>
>         >>> Hi Jim,
>         >>>
>         >>> the call for bumphunter is:
>         >>>
>         >>> bumphunter(myGenomicRatioSet, design=designMatrix,
>         cutoff=0.05, B=1000,
>         >>> verbose=TRUE)
>         >>>
>         >>> I set cutoff to 0.05 not cutoffQ.
>         >>> Next bumphunter call I will follow your advice;
>         >>> regard to parallelize the process what is your suggestion?
>         >>>
>         >>> Would it be better to parallelize only on some of the ncores?
>         >>>
>         >>> Best,
>         >>> Giovanni
>         >>>
>         >>> Laboratory of Preclinical and Translational Research
>         >>> IRCCS - CROB Oncology Referral Center of Basilicata
>         >>> Rionero in Vulture - Basilicata - Italy
>         >>>
>         >>>
>         >>> 2014-02-21 17:52 GMT+01:00 James W. MacDonald
>         <jmacdon at uw.edu <mailto:jmacdon at uw.edu> <mailto:
>         >>> jmacdon at uw.edu <mailto:jmacdon at uw.edu>>>:
>         >>>
>         >>>
>         >>>     Hi Giovanni,
>         >>>
>         >>>     You should give the actual call you used for
>         bumphunter, rather
>         >>>     than simply saying that you used 'recommended
>         parameters'. It
>         >>>     looks like you set cutoffQ to 0.05, which is exactly
>         opposite to
>         >>>     what you should be doing. Instead you should be using
>         something
>         >>>     like cutoffQ = 0.95.
>         >>>
>         >>>     Best,
>         >>>
>         >>>     Jim
>         >>>
>         >>>
>         >>>
>         >>>
>         >>>     On Friday, February 21, 2014 11:36:32 AM, Giovanni
>         Calice wrote:
>         >>>
>         >>>         Hi Kasper,
>         >>>
>         >>>         Thank you very much for the immediate response!
>         >>>
>         >>>           So,
>         >>>           What do you suggest?
>         >>>
>         >>>         Giovanni
>         >>>
>         >>>         Laboratory of Preclinical and Translational Research
>         >>>         IRCCS - CROB Oncology Referral Center of Basilicata
>         >>>         Rionero in Vulture - Basilicata - Italy
>         >>>
>         >>>
>         >>>         2014-02-21 17:27 GMT+01:00 Kasper Daniel Hansen
>         >>>         <khansen at jhsph.edu <mailto:khansen at jhsph.edu>
>         <mailto:khansen at jhsph.edu <mailto:khansen at jhsph.edu>>>:
>         >>>
>         >>>
>         >>>
>         >>>             On Fri, Feb 21, 2014 at 11:25 AM, Giovanni Calice
>         >>>             <giovcalice at gmail.com
>         <mailto:giovcalice at gmail.com> <mailto:giovcalice at gmail.com
>         <mailto:giovcalice at gmail.com>>>wrote:
>         >>>
>         >>>
>         >>>                 273952
>         >>>
>         >>>
>         >>>
>         >>>             This is a crazy number of putative bumps
>         >>>
>         >>>             Kasper
>         >>>
>         >>>
>         >>>                 [[alternative HTML version deleted]]
>         >>>
>         >>> _______________________________________________
>         >>>         Bioconductor mailing list
>         >>> Bioconductor at r-project.org
>         <mailto:Bioconductor at r-project.org>
>         <mailto:Bioconductor at r-project.org
>         <mailto:Bioconductor at r-project.org>>
>         >>>
>         >>> https://stat.ethz.ch/mailman/listinfo/bioconductor
>         >>>         Search the archives:
>         >>> http://news.gmane.org/gmane.science.biology.informatics.
>         >>> conductor
>         >>>
>         >>>
>         >>>     --
>         >>>     James W. MacDonald, M.S.
>         >>>     Biostatistician
>         >>>     University of Washington
>         >>>     Environmental and Occupational Health Sciences
>         >>>     4225 Roosevelt Way NE, # 100
>         >>>     Seattle WA 98105-6099
>         >>>
>         >>>
>         >>>
>         >> --
>         >> James W. MacDonald, M.S.
>         >> Biostatistician
>         >> University of Washington
>         >> Environmental and Occupational Health Sciences
>         >> 4225 Roosevelt Way NE, # 100
>         >> Seattle WA 98105-6099
>         >>
>         >>
>         >
>
>                 [[alternative HTML version deleted]]
>
>         _______________________________________________
>         Bioconductor mailing list
>         Bioconductor at r-project.org <mailto:Bioconductor at r-project.org>
>         https://stat.ethz.ch/mailman/listinfo/bioconductor
>         Search the archives:
>         http://news.gmane.org/gmane.science.biology.informatics.conductor
>
>
>

-- 
James W. MacDonald, M.S.
Biostatistician
University of Washington
Environmental and Occupational Health Sciences
4225 Roosevelt Way NE, # 100
Seattle WA 98105-6099



More information about the Bioconductor mailing list