The code from "Assoc_map_winbugs_1.txt" can be run as such but data needs
to be given first.
No separate data input files are needed!
DATA STRUCTURE:
Data can be given in, for example, with the list-command.
To specify 200 individuals,
5 markers (with 3,4,4,4, and 6 segregating alleles)
with intermarker distances (in Morgans) as : 0.01, 0.02, 0.001, and 0.01.
You can give this specification with one line:
list (n1=200, m=5, n2=c(3,4,4,4,6), n2max=7, d=c(0.01, 0.02, 0.001, 0.01))
To give genotype data, you can write for example:
list (y=structure(.Data=c(1,2,...,), .Dim=c(200,10)))
Here, number 200 corresponds to the number of individuals
and number 10 means 5 markers and two alleles in each.
Corresponding data is given as one vector so that first 10 numbers
are marker alleles for the first individual (2 alleles at 5 markers),
next 10 numbers for the second individuals and so on.
Allele codes: it is required that alleles are coded in consecutive numbers
starting from 1. The code for missing allele is NA.
To give phenotypes you can write for example:
list (z1=c(100,210,33,...))
Here, 200 consecutive numbers in data vector are the phenotypic
values of all 200 individuals.
To run the model, you can first "paint" the word
and select "check model" from the menu. It may be that
before model runs, WinBUGS sometimes needs good starting values for
some of the parameters.
INITIAL VALUES:
Initial values can be given in using similar structure / format than
data (with the list-command). It is OK with WinBUGS that you specify
initial values only for some of the parameters. Other parameters can
then be generated by using "GET INITS". However, you should not
use 'NA' as initial value. Initial values must also be within
prior limits (you should not give zero if it is not possible to get
from the prior).
Definition of convergence is that estimates are not depending on your
initial values. Therefore, INITIAL values should only influence on the
convergence speed but not to the estimates. If latter happens, it means
that there are problems in convergence.
OUTPUT:
The variable 'beta' includes original allelic coefficients. 'bo' is allelic
coefficients weighted with genetic variance. The latter is needed if the
indicators and effects are confounded by strong LD in the region.
In other cases, you can inspect 'beta' directly. However, remember that
this is overparameterized model where number of estimable contrasts are
one less than number of alleles. Because of this, coefficients
themselves can have very arbitraty scale/level. Therefore, you should
centralize the coefficients after estimation. For example,
see Table II in Kilpikari and Sillanpaa (2003).
The variable 'X2' includes information for posterior distribution of
QTLs (markers having simultaneously indicator value one).
It has (number of markers+1) elements so that
X2[1] contains information for P(N_qtl=0 | data).
X2[2] contains information for P(N_qtl=1 | data).
x2[3] contains information for P(N_qtl=2 | data).
...
..
Good luck!