Posted from a different acct. after some problem in receiving the
original. -the mod
-------- Original Message --------
Subject: morphmet post
Date: Sun, 21 Feb 2010 10:13:46 +1100
From: Paul Sanfilippo <[email protected]>
To: [email protected]
Dear List,
I am hoping I can open some discussion about the application of
quantitative genetics and in particular the calculation of heritability
for shape data. I'm in the middle of a PhD looking at the heritability
of optic nerve shape in humans.
I am aware of the constructive debate about univariate vs multivariate
approaches and the papers that came out of this. What I'm having trouble
with is understanding some of the basic concepts, particularly in the
multivariate context.
I'm from an ophthalmological background and have been involved with a
couple of heritability studies that our group have put out in the past.
These have all been based on twin data by the way. So, I've got some
idea about what heritability is, covariance modelling and expectations
based on genetic similarity between twins of differing zygosity, etc.
With regards to the multivariate case, for some reason I can't get my
head around the idea of G and P matrices (additive genetic and
phenotypic covariance matrices). The P matrix is based on observable
trait values, so a covariance matrix of PC scores from a GPA for
example? What makes up the G matrix? and how might this apply in the
twin study context (ie for a univariate trait, the genetic covariance
for a non-identical twin pair being half that for an identical twin pair
(which is 1)). Multiplying G by the inverse of P gives an approximation
of narrow-sense heritability in a multivariate context? I understand
taking this resultant matrix and pulling a single number from it is
ultimately disregarding a lot of information, but am I correct in
thinking that either the maximum eigenvalue or trace (G)/ trace (P)
gives a univariate approximation?
In terms of software, I see that either VCE or Wombat are largely used
for these types of analyses in GMM. Is anyone familiar with Mx (or
Open-Mx (in R)) and can this also be used to calculate GP-1. It's also a
maximum likelihoods based package. I ask, because we have statistical
genetics people we normally use for this work and they use Mx. It would
save me a lot of time in starting from scratch to learn one of the other
programs.
Sorry for all the questions - I'd certainly appreciate any thoughts that
any of you may have.
Many thanks,
Paul
--
For now, new message AND replies should be sent to:
[email protected]
/* Replies will be sent to the list. */
For more information visit http://www.morphometrics.org
