Re: G*E in animal model

Rob and Alastair

There is an easier way to do the same thing using the same data file you set
up for the univariate run. You need to sort on environments prior to
analysis then code the run as follows:

CE_PN ~ mu ENV !r ENV.Anim
tab CE_PN ~ ENV !count #do this first to get counts in .tab file

2 1 1
XXX #put number of records in ENV1 here from .tab file
YYY # records in ENV2

ENV.Anim 2
2 0 US !GP
1
0.1 1 #or better starting values

It's a good idea to get the program to count the records for you in a
preliminary run. This is simply done by putting a zero on the line after
the tab statement. You can then look up the number of records in the .tab
file to insert in the R section (where the XXXs and YYYs are). Remove the
zero and run again. The job will provide two residuals, one for each ENV,
plus the 3 animal components as described by Rob Brooks (2 variances and
covariance). The .pin file is pretty straightforward. As pointed out by
Rob you could equally replace the unstructured G with a correlation
structure, eg:

ENV.Anim 2
2 0 CORUH 0.9 !GP #guess correlation
2*0. #tells program to use its own initiative for starting values

This provides a direct estimate of the rg in the .asr file. It can easily
be extended to more envs and more complex correlation structures.

Good luck!

Regards

rw

-----Original Message-----
From: ASReml users discussion group [mailto:ASREML-L_at_AGRIC.NSW.GOV.AU] On
Behalf Of Rob Brooks
Sent: Thursday, 21 July 2005 11:11 AM
To: ASREML-L_at_AGRIC.NSW.GOV.AU
Subject: Re: G*E in animal model

Thanks Alastair,

I had a similar idea, but didn't know if it would be an acceptable test of
G*E. Your suggestion has encouraged me to plough ahead with this approach.

Allt he best,

Rob

At 10:04 AM 20/07/2005 +0100, you wrote:
>Rob,
>
>Since it looks like you have 2 environments another way to do it would be
to
>set it up as a two trait model with CE_PN_env1 and CE_PN_env2 as separate
>traits. Adjust the data file accordingly and then the rest could look
>something like...
>
>CE_PN_env1 CE_PN_env2 ~ Trait !r Trait.Anim
>
>1 2 1
>
>0
>Trait 0 US
>0.1
>0.01 0.1 #or better starting values
>
>Trait.Anim 2
>Trait 0 US
>1
>0.1 1 #or better starting values
>Anim
>
>
>
>This would give you a residual and additive variance for the trait in each
>environmnet, plus the covariances across environments. Scaling the COVa to
a
>genetic correlation would let you see if rG was significantly les sthan 1
>(one way to define GxE). If each animal is only measured in one
environment
>or the other then the residual covariance will be 0 but you should still
>estimate the additive covariance.
>
>Cheers,
>
>Alastair
>
>Alastair J Wilson PhD
>Postdoctoral Research Fellow
>
>
>Alastair.Wilson_at_ed.ac.uk
>
>tel. 0131 6507287
>
>Institute of Evolutionary Biology
>School of Biological Sciences
>The University of Edinburgh
>Ashworth Laboratories
>The King's Buildings
>West Mains Road
>Edinburgh EH9 3JT, UK
>
>
>
>
>----- Original Message -----
>From: "Rob Brooks" <rob.brooks_at_UNSW.EDU.AU>
>To: <ASREML-L_at_AGRIC.NSW.GOV.AU>
>Sent: Wednesday, July 20, 2005 2:05 AM
>Subject: G*E in animal model
>
>
>>Dear ASREMLers
>>
>>I have a general and a specific query. Generally, I have enormous trouble
>>designing the G and R structure lines of my ASREML jobs. Unfortunately I
>>am
>>just not well enough schooled in thinking about these problems, but I find
>>I have to really bang my head agains the wall to figure out how to design
>>these bits of my jobs. Apart from teh AREML help, Asreml cookbook and
>>various web nortes from workshops, does anyone know of a good starting
>>place to learn about how to set these things up properly, from first
>>principles.
>>
>>More specifically the problem I have been grappling with for the last
>>while
>>is how to test for G*E interactions in a univariate animal model. I have
>>the animal model going fine, as per the below .as file, but I cannot fit
>>the Anim.ENV term. Does one specify Anim.ENV as a random effect, or is
>>there another way of doing it. Irrespective, if anyone has any insights on
>>how to specify the G and R structure lines in the job file, they would be
>>much appreciated.
>>
>>Many thnaks,
>>
>>Rob
>>
>>
>>
>>Single Trait Animal Model
>> Anim * !P
>> Sire * !P
>> Dam * !P
>> SEX !A 2
>> GENERAT !A
>> ENV 2
>> CE_PN
>>am_tfd.txt !SKIP1 !ALPHA !MAKE #pedigree file
>>am_tfd.txt !SKIP1
>>CE_PN ~ mu ENV !r Anim
>>0 0 1
>>Anim 1
>>0 0 AINV
>>..........................................................................
..
>>............................
>>School of Biological, Earth and Environmental Sciences
>>The University of New South Wales
>>Kensington, Sydney 2052
>>NSW, Australia
>>PH: +61-2-9385-2587 FAUX: +61-2-9385-1558
>>http://www.bees.unsw.edu.au/research/groups/brookslab/brookslab.html

............................................................................
............................
School of Biological, Earth and Environmental Sciences
The University of New South Wales
Kensington, Sydney 2052
NSW, Australia
PH: +61-2-9385-2587 FAUX: +61-2-9385-1558
http://www.bees.unsw.edu.au/research/groups/brookslab/brookslab.html
Received on Wed Jul 21 2005 - 12:34:58 EST