Re: Multivariate analysis

From: Arthur Gilmour <arthur.gilmour_at_CARGOVALE.COM.AU>
Date: Mon, 2 Nov 2009 17:55:31 +1100

Dear Hong-Sheng,

1) The first thing I note is that there are two AI singularities in the
you did.
a) The first is because you specified the error variance structure as
CORGV but ASReml has not recognised the !S2==1
statement because it needs to be on the same line as CORUV
Note the message saying it was ignored!

b) The second is that the explicit error term you specified as
Trait.PID 1
2000 0 IDV 0.5

is completely confounded with the error structure CORUV
so you need to drop this term from the model.

2) I ran an analysis using US structures for R ang G and obtained.
 Covariance/Variance/Correlation Matrix UnStructured Residual
  0.6349 0.2902 0.2250 0.1377 0.1757
  0.2002 0.7496 0.2262 0.1897 0.1501
  0.1461 0.1596 0.6640 0.1997 0.1743
  0.9135E-01 0.1368 0.1355 0.6931 0.1536
  0.1177 0.1092 0.1193 0.1074 0.7059
 Covariance/Variance/Correlation Matrix UnStructured Trait.PID
  0.1124 1.339 0.4900 0.6756 0.8059
  0.7580E-01 0.2851E-01 0.5560 1.226 1.521
  0.6242E-01 0.3567E-01 0.1444 0.6694 0.6920
  0.8530E-01 0.7798E-01 0.9579E-01 0.1418 0.7283
  0.8631E-01 0.8204E-01 0.8400E-01 0.8762E-01 0.1021

The latter matrix is not positive definite. If you just have four
families, that could be the reason.

The model failed to converge and I think a simpler model is in order.

I also ran using uniform matrices ar both R and G level and got LogL of
-891 (the US+US model was getting -822 so the uniform structures are
too simple.) I leave you to try other combinations.

In any case, the fitted geno regressions are
  Trait.geno 1.001 0.2297 0.6133E-01
  Trait.geno 2.001 0.1519E-01 0.6342E-01
  Trait.geno 3.001 0.5924E-02 0.6361E-01
  Trait.geno 4.001 0.3112E-01 0.6473E-01
  Trait.geno 5.001 0.1467E-01 0.6392E-01
  Trait 1 0.5264E-02 0.4482E-01
  Trait 2 -0.5961E-02 0.4546E-01
  Trait 3 0.6066E-01 0.4673E-01
  Trait 4 0.5838E-01 0.4750E-01
  Trait 5 0.5538E-01 0.4650E-01

I note you left the intercept term out (presumably because you did not
simulate it in the data which is more or less centred around zero).

4) I think the answers to your questions are included above.


On Sun, 2009-11-01 at 22:52 -0500, Wu, Hong-Sheng wrote:
> We tried to use ASREML to model a multivariate outcome. We simulated
> five correlated traits on four members of independent nuclear
> families. Each family has four individuals, two parents and two
> children. We simulated the genotype of a SNP, and tried to model the
> relationship between the SNP and the multivariate phenotype (P1 – P5).

Best Wishes,
Arthur Gilmour
Adjunct Professor
School of Computing and Mathematics
Charles Sturt University
Jesus went to the synagogue in every town in Galilee to preach and drive
out demons.
A leper came, knelt down and said, "If you will, you can make me clean".
Jesus was moved. He reached out and touched him and said 
"I am willing.  Be clean." He was cleansed immediately.
Mobile Number +61 427 227 468
Home phone +61 2 6364 3288  Skype: Arthur.Gilmour
Tumut  5-6 Nov (RT & BC)
Brisbane Probe 29Nov to 6 Dec
Bangladesh Proshikkon 31Dec - 6 Feb
Received on Tue Nov 02 2009 - 17:55:31 EST

This webpage is part of the ASReml-l discussion list archives 2004-2010. More information on ASReml can be found at the VSN website. This discussion list is now deprecated - please use the VSN forum for discussion on ASReml. (These online archives were generated using the hypermail package.)