Files in this directory are:
 README		This file
 diff.dat	data for qtldiff.mx and qtldiffpi.mx
 q5v1850.dat    data for qtl.mx and qtlpi.mx
 q5v185275.dat  data for twoloc.mx
 qtl.mx         Mixture model approach, raw scores of twin 1 and twin 2
 qtldiff.mx     Mixture model approach, twin 1 - twin 2 difference scores
 qtldiffpi.mx   Pi-hat approach, twin 1 - twin 2 difference scores
 qtlpi.mx	Pi-hat approach, raw scores of twin 1 and twin 2
 twoloc.mx      Two locus mixture model approach (9 models for IBD status 
		    at two unlinked loci)

These scripts correspond to methods described in 

Eaves, Neale & Maes (1996) Multivariate Multipoint Linkage Analysis of
Quantitative Trait Loci.  Behavior Genetics 26:519-525

and in 

D.W. Fulker and S.S. Cherny (1996) An improved multipoint sib-pair 
analysis of quantitative traits. Behavior Genetics 26:527-532

Scripts qtl.mx, qtldiff.mx, qtldiffpi.mx and qtlpi.mx correspond to the 
four methods described in Fulker & Cherny.

Script twoloc.mx does a multivariate multipoint sib pair analysis, with
five traits and two loci. The script should recover similar parameter
estimates to those reported in the paper, after quite a bit of computer
time.  CPU time is high because there are nine different models (IBD 0/1/2
at locus 1 times IBD 0/1/2 at locus 2 = 9 combinations).  The likelihood
of these models is weighted by the outer product of the vectors of IBD
probability for each sib pair.  The model also estimates the residual 
(non-qtl) genetic and environmental factor loadings, making a total of 35 
free parameters.  With 1000 sib pairs this takes about 4 hours on a DEC 
alpha 3000/700. 

Note that these models need to be re-run for successive loci.  We usually 
do this by saving the parameter estimates from a run at one locus, and 
reading them in as starting values for the next locus.  This method saves 
about 50% or more computing time, depending on the number of traits 
involved.

Also note that there are two "null models" with no Q effects.  First 
possibility is to fix the IBD probabilities for each sib pair to .25, .5 
and .25 but to keep the Q parameter(s) free.  This is the lod-score 
approach.  An alternative is to use the log-likelihood ratio for Q fixed 
to zero (for which it makes no difference to have estimated IBD 
probabilities).

Data files contain the IBD probabilities pre-computed by other software, 
such as Mapmaker/sibs (Kruglyak & Lander, AJHG 1995).
see:
ftp://ftp-genome.wi.mit.edu/distribution/software/sibs 
and 
http://linkage.rockefeller.edu/soft/mapmaker/sibs20.html

We used a FORTRAN program written by Dr. Lindon Eaves (eaves@gems.vcu.edu)
to compute IBD probabilities. 

Mike Neale
December 2 1996