Re: Genetic grops

Thanks, Arthur!

I interpret your answer as regards computation of relative BVs (that is, ratios with a common denominator; of course differences are invariant) in this way (let MBV be the mean of BVs and Mblock be the mean of block effects):

relative BVi = 100(BVi + Mblock)/(MBV + Mblock)%

in order to have a relative value that is not biased by what happens to become the first factor-indexed block in the ASReml run.

Tore Ericsson
Skogforsk, Sävar

-----Original Message-----
From: ASReml users discussion group [mailto:ASREML-L_at_AGRIC.NSW.GOV.AU] On Behalf Of arthur.gilmour_at_DPI.NSW.GOV.AU
Sent: 22 september 2006 08:48
To: ASREML-L_at_AGRIC.NSW.GOV.AU
Subject: Re: Genetic grops

Dear Tore,
re:

Regarding computation of "real" breeding values when using an individual
(tree) model with genetic groups

Model: y ~ block !r genid

In the sln file, I identify

Block effects: 0, b2, b3, b4, b5 (blue for five blocks, a singularity
due to groups)

Group effects: g1, g2, g3, g4 (four genetic groups)

Individual breeding values (BVs): One for each tree in the data file (or
in the pedigree file)

ARG: I like to check this out on a simple example but my interpretation is
that the overall mean is included in the group effects and therefore in
the 'BV's. i.e. these values are relative to Block_1

So a 'least squares mean' overall would be the average of the block and
group effects.

You would get relative BV's centred on average of Groups by subtracting
the average of the 4 group means.

Problem: How should I compare and/or combine the mean of (1) BVs, (2)
group effects, and (3) block effects?

So, you could predict group effects with
predict genid 1:4 !ave block

You can predict block means with
predict block !ave genid { 4*1 ???*0}/4

I assume that the group effects are included in the BVs, and that the
group effects represent the group means in some way.

YES: for BLOCK_1

The averages of BVs and group effects, respectively, do not agree very
well for me. Also, adjusting either of them with tha mean of the block
effects (zero included) makes it still worse.

Only in a very simple model would the average of tree effects be same as
group means. They are regressed to that mean.

Question: How should I best combine these effects to get values that are
adequate for computation of relative breeding values?

The breeding values should be directly comparable as they are
(differencing will cancel any offset added).
Presumably you are not wanting to subtract out the 'group' effects on the
assumption these are genetic.

Trust this helps.

Grace, mercy and peace to you from God our Saviour Jesus Christ

Arthur Gilmour PhD
Received on Wed Sep 22 2006 - 10:54:32 EST