[GenABEL-dev] deflated chi2 from ProbABEL output

Yurii Aulchenko yurii.aulchenko at gmail.com
Thu Jan 6 22:48:06 CET 2011 

Oh mine. You used wrong option - we have had multiple problems using
the procedure you've outlined. Sorry it was not clear it is bad one...

The interaction covariate you need to include at step 2 (probabel),
not to polygenic. never use --interaction_only (I actually thought we
have killed this option, but apparently not - something to do). I will
send you some slides about that known problem + the latest protocol we
used for this type of analysis; if you do not get that within a day,
please remind me.

Yurii

On 1/6/11, Zhai, Guangju <guangju.zhai at kcl.ac.uk> wrote:
> Hi, Yurii
>
> Happy New Year!
>
> I managed to complete my interaction analyses but noticed that there is a
> huge deflated chi2 for the interaction effect, the lambda ranged from 0.175
> to 0.678 in my four phenotype analyses. There is no problem for the SNP main
> effect, the lambdas are close to 1 as expected. Can you help explain why?
>
> What I did was I first run polygenic model in GenABEL with all covariates
> included, then run ProbABEL with interaction_only and mmscore function. The
> following is one of my command line used:
>
> ProbABEL/bin/palinear -p z_fev1_interaction_smk_pheno.PHE -i
> proab_CHR22.machinfo -d 22.gen.dose.fvi --map proab_CHR22.machlegend
> --mmscore mmscore_z_fev1_InvSigma_aj_smk_ever_smk_currrent_packyr.dat
> --interaction_only=1  -c 22 --out z_fev1_interaction_smk_chr22
>
> Thanks,
>
> Guangju
>
> -----Original Message-----
> From: Yurii Aulchenko [mailto:yurii.aulchenko at gmail.com]
> Sent: 21 December 2010 17:09
> To: Zhai, Guangju
> Subject: Re: time to run ProbABEL
>
> This is indeed big problem - the running time with mmscore. We work
> hard on new method (grammar+) which will help a lot. We expect it to
> be out for use in next month or two.
>
> Yurii
>
> On 12/21/10, Zhai, Guangju <guangju.zhai at kcl.ac.uk> wrote:
>> Thanks, Yurii
>>
>> In my situation, it looks take much more than 7 days to complete. Given I
>> run 4 phenotypes, it will take couple of months to complete.
>>
>> I don't know what command Dana used, but she has unrelated samples, so it
>> wouldn't be "mmscore", but with three covariates plus one interaction term
>> in linear regression model.
>>
>> Kind regards
>> Guangju
>>
>> From: yurii.aulchenko at gmail.com [mailto:yurii.aulchenko at gmail.com]
>> Sent: 21 December 2010 08:48
>> To: Zhai, Guangju
>> Subject: Re: time to run ProbABEL
>>
>> It is expected that 'mmscore' analysis is taking a while (say, 3000 people
>> is about 3-7 days to complete). Also, analysis with 3 covariates + SNPxE
>> interaction is expected to take twice longer then main effects only with 1
>> covariate.
>>
>> Do you know what command/model did Dana use to run analysis?
>>
>> Yurii
>>
>> On Dec 20, 2010 5:05pm, "Zhai, Guangju" <guangju.zhai at kcl.ac.uk> wrote:
>>> Hi, Yurii
>>>
>>>
>>>
>>> I am running my GWAS interaction analysis using ProbABEL, the program
>>> seems running fine but taking such a substantial time. I have just over
>>> 3000 related samples, after running polygenic model in GenABEL, I set up
>>> ProbABEL to run mmscore test last Wed, it has just completed 2 analyses
>>> on
>>> chr 1 (I have 4 analyses on each of the 22 chromosomes) by yesterday. It
>>> looks it will take ages to complete the analyses.
>>>
>>>
>>>
>>> Today I have a communication with Dana from ARIC study, she used ProbABEL
>>> to run the same analyses as I am doing in 9000 unrelated individuals, but
>>> only took her less than a day to complete.
>>>
>>>
>>>
>>> I  have checked my output so far, it seems fine, I didn't find any
>>> strange
>>> thing. So wonder if that is kind of what expected or there is something
>>> that I can do to speed up the analyses? I am using the ProbABEL vesion
>>> 0.1-9c.
>>>
>>>
>>>
>>> Thanks,
>>>
>>>
>>>
>>> Guangju
>>>
>>>
>>>
>>> -----Original Message-----
>>>
>>> From: Yurii Aulchenko [mailto:yurii.aulchenko at gmail.com]
>>>
>>> Sent: 16 December 2010 16:25
>>>
>>> To: Zhai, Guangju; l.karssen
>>>
>>> Subject: Re: impute2mach
>>>
>>>
>>>
>>> You use version of ProbA which is very old and does not support DatA
>>> format
>>>
>>>
>>>
>>> Upgrade to the latest (0.1-9d or so)!
>>>
>>>
>>>
>>> Yurii
>>>
>>>
>>>
>>> On 12/16/10, Zhai, Guangju guangju.zhai at kcl.ac.uk> wrote:
>>>
>>> > Hi, Yurii
>>>
>>> >
>>>
>>> > After converting all the files, I tried to run the ProABEL, but got a
>>>
>>> > problem which said "segmentation fault", the below is the log file. I
>>>
>>> > thought it was a RAM issue, but we got 250GB RAM here, the ProABEL
>>> > manual
>>>
>>> > said it only requires a small RAM. Could you please advise how to sort
>>> > this
>>>
>>> > problem out?
>>>
>>> >
>>>
>>> > Many thanks,
>>>
>>> >
>>>
>>> > Guangju
>>>
>>> >
>>>
>>> > [guangju at dtrcluster interaction-lung]$ ProbABEL/bin/palinear -p
>>>
>>> > z_fev1_pheno.PHE -i chr22.machinfo -d 22.gen.dose.fvi --map
>>> > chr22.machlegend
>>>
>>> > --mmscore mmscore_z_fev1_InvSigma_aj_smk_ever_smk_currrent_packyr.dat
>>>
>>> > --interaction_only=1 --out z_fev1_interaction_chr22
>>>
>>> > ProbABEL v. 0.1-3 (Nov 11, 2009) (C) Yurii Aulchenko, Maksim
>>> > Struchalin,
>>>
>>> > EMCR
>>>
>>> >
>>>
>>> > Options in effect:
>>>
>>> >          --pheno   = z_fev1_pheno.PHE
>>>
>>> >          --info    = chr22.machinfo
>>>
>>> >          --dose    = 22.gen.dose.fvi
>>>
>>> >          --ntraits = 1
>>>
>>> >          --ngpreds = 1
>>>
>>> >          --interaction = 0
>>>
>>> >          --interaction_only = 1
>>>
>>> >          --mmscore =
>>>
>>> > mmscore_z_fev1_InvSigma_aj_smk_ever_smk_currrent_packyr.dat
>>>
>>> >          --map     = chr22.machlegend
>>>
>>> >          --nids    = estimated from data
>>>
>>> >          --chrom   = not in output
>>>
>>> >          --out     = z_fev1_interaction_chr22
>>>
>>> >          --skipd   = 2
>>>
>>> >          --separat = " "
>>>
>>> >          --score   = OFF
>>>
>>> >          --nohead  = OFF
>>>
>>> >          --allcov  = OFF
>>>
>>> >          --robust  = OFF
>>>
>>> > Number of SNPs = 40322
>>>
>>> > Actual number of people in phenofile = 5654; using all of these
>>>
>>> > model is running without smk_ever, term
>>>
>>> > Linear model: ( z_fev1 ) ~ mu + smk_ever + smk_current + packyr +
>>> > SNP_A1
>>> > +
>>>
>>> > smk_ever*SNP_A1
>>>
>>> > you are runing mmscore...
>>>
>>> > Reading data ...Segmentation fault
>>>
>>> >
>>>
>>> > -----Original Message-----
>>>
>>> > From: Yurii Aulchenko [mailto:yurii.aulchenko at gmail.com]
>>>
>>> > Sent: 18 November 2010 18:25
>>>
>>> > To: Zhai, Guangju; l.karssen
>>>
>>> > Subject: Re: impute2mach
>>>
>>> >
>>>
>>> >> By the way, how the generated .fvd and .fvi be used in the ProABEL?
>>>
>>> >
>>>
>>> > the same way you do that with regulat files -- if the program sees
>>>
>>> > that you specify *fvi as geno/dose file, it will know this DA data.
>>>
>>> > Should be described in PA manual
>>>
>>> >
>>>
>>> > Yurii
>>>
>>> >
>>>
>>> >> -----Original Message-----
>>>
>>> >> From: Yurii Aulchenko [mailto:yurii.aulchenko at gmail.com]
>>>
>>> >> Sent: 18 November 2010 17:25
>>>
>>> >> To: Zhai, Guangju; Stepan Yakovenko; Message Logger
>>>
>>> >> Subject: Re: impute2mach
>>>
>>> >>
>>>
>>> >> Ok, great news! Stepan, thanks for doing this work!
>>>
>>> >>
>>>
>>> >> Zhai, I think by now you are a kind of expert in issues which may go
>>>
>>> >> wrong with impute2* procedures :) In the future, if we get people with
>>>
>>> >> similar problems arising, would that be ok to involve you into
>>>
>>> >> dicussion so you can provide an advise based on your experience?
>>>
>>> >>
>>>
>>> >> Best wishes,
>>>
>>> >> Yurii
>>>
>>> >>
>>>
>>> >> On 11/18/10, Zhai, Guangju guangju.zhai at kcl.ac.uk> wrote:
>>>
>>> >>> It looks yes! At least on chr 22 data.
>>>
>>> >>>
>>>
>>> >>> I will keep you updated for other chrs.
>>>
>>> >>>
>>>
>>> >>> Many thanks for both you and Stepan!!
>>>
>>> >>>
>>>
>>> >>>
>>>
>>> >>> Guangju
>>>
>>> >>>
>>>
>>> >>> -----Original Message-----
>>>
>>> >>> From: Yurii Aulchenko [mailto:yurii.aulchenko at gmail.com]
>>>
>>> >>> Sent: 18 November 2010 12:45
>>>
>>> >>> To: Zhai, Guangju; Stepan Yakovenko; Message Logger
>>>
>>> >>> Subject: Re: impute2mach
>>>
>>> >>>
>>>
>>> >>> Loss of precision -- no problem with that, expected and appreciated
>>>
>>> >>> behaviour\
>>>
>>> >>>
>>>
>>> >>> So, is the problem fixed at the end?!
>>>
>>> >>>
>>>
>>> >>> On Thu, Nov 18, 2010 at 1:41 PM, Zhai, Guangju
>>> >>> guangju.zhai at kcl.ac.uk>
>>>
>>> >>> wrote:
>>>
>>> >>>> Hi, Stepan
>>>
>>> >>>>
>>>
>>> >>>>
>>>
>>> >>>>
>>>
>>> >>>> I just tested after adding two missing columns in the first two
>>> >>>> lines
>>> >>>> in
>>>
>>> >>>> the
>>>
>>> >>>> sample file and adding the headers in the info file, it seems fine,
>>> >>>> no
>>>
>>> >>>> warning appeared! But I still have the following statement: not sure
>>>
>>> >>>> what
>>>
>>> >>>> that mean and if it has impact on the generated files?
>>>
>>> >>>>
>>>
>>> >>>>
>>>
>>> >>>>
>>>
>>> >>>> Guangju
>>>
>>> >>>>
>>>
>>> >>>>
>>>
>>> >>>>
>>>
>>> >>>> Loss of precision / loss of data during conversion from DOUBLE to
>>> >>>> FLOAT.
>>>
>>> >>>>
>>>
>>> >>>> Futher conversion warnings omitted.
>>>
>>> >>>
>>>
>>> >>
>>>
>>> >> --
>>>
>>> >> Sent from my mobile device
>>>
>>> >>
>>>
>>> >
>>>
>>>
>>>
>>> --
>>>
>>> Sent from my mobile device
>>>
>>
>
> --
> Sent from my mobile device
>

-- 
Sent from my mobile device

More information about the genabel-devel mailing list