Executables:

Windows executable:  k.exe
Solaris executable:  gcontrol.exe


Introduction:

GControl is a computer program for Bayesian analysis of case-control
data that controls for population stratification and cryptic
relatedness.  GControl performs these analyses using Markov chain
Monte Carlo algorithms.  Below we outline how to implement this
analysis.  See Devlin and Roeder (1999) for analytical details.  A
frequentist version of GControl and a program for computing power is
forthcoming and will be posted to this web site.

Written by Bobby Jones in collaboration with Kathryn Roeder, 1999.
This program is written in C.

Definitions and input parameters (*):

* n = number of loci under consideration

   lambdahat = median{of the n trend statistics}/0.46
     (This value is computed internally)
   
   Prior for noncentrality parameter A_i is 
   Normal(sqrt(lambdahat)kappa,lambdahat*tau2)

* kappa = multiplier in prior above, by default set at 1.6 * sqrt{log(n)}  
* tau2 = multiplier in prior above, by default set at 1
* epsilon = prior probability a marker is associated, by default set at 10/n
* ngib = number of cycles for the gibs sampler after burn in, by default 500
* burn = number of cycles for the gibs sampler to burn in, by default 50

The program computes the test statistic, which is Armitage's trend
test and the posterior probability each marker is associated
with the disorder.  This is not a p-value.  Any value greater
than 0.5 suggests association. All values greater than pout will be 
printed in the output file.  

* pout =  minimum probability of association to include in the results,
        by default 0.5

The default values have been chosen to give a modest number of
Type I errors, see Devlin and Roeder (1999) for details.

Instructions:

When you run the program, it will prompt you for a value of n.
This must be entered first.  Then you will have the opportunity 
to do one of the following at the gcontrol prompt:
 - Change one of the following variables by entering the name of 
   the variable:
     n, kappa, tau2, epsilon, ngib, burn, pout
     Note that default values for kappa and epsilon are 
     calculated using the INITIAL value of n.
 - Include all results by typing "all"
 - Change the name of the output file by typing "output"
 - Save current variable settings by typing "save"
 - Load previously saved variable settings by typing "load"
 - Run the analysis by typing "run"
 - Quit by typing "q"
 - For a list of commands, options, and current variable values, 
   type "help"


Input:

Data should be put into a file called kdata.txt.
Column 1 should consist of names of loci
Columns 2 through 7 should consist of tables of A1 alleles 
(1 row of 6 values per table)
  cases followed by controls
  r0 r1 r2 s0 s1 s2 
File should be space delimited.
File should have no blank lines.  
The last line of data should be the last line of the file.


Output:

File of loci names, probability observation is an outlier, 
and chi squared value.  
Includes all observations with probability > pout.



Reference:

Devlin, B. and Roeder, K (1999) Genomic Control for Association 
Studies. Biometrics 55, 997-1004.