Permutation Test output

If you chose to do a permutation test [Churchill and DoergeChurchill and Doerge1994] for the purpose of estimating experiment specific threshold values, Zmapqtl will create two auxiliary files to store interim comparisonwise and experimentwise test statistics. If the filename stem is ``qtlcart'' and the model for analysis is ``6'', then these files will be qtlcart.z6c and qtlcart.z6e. The former file should look something like this:

#Row Chrom Mark Position  Original    P-Val     Count    -perm 899
   -start
    1   1   1  0.00010    0.00000   0.982202        883
    2   1   1  0.02010    0.00000   0.976641        878
.
.
.

whose columns are

1. Integer indicating the row.
2. Chromosome of test position.
3. Left flanking marker of test position.
4. Absolute position of test from left telomere, in Morgans.
5. Likelihood ratio test statistic of actual data. For backcrosses, this is $\frac{H_1}{H_0}$, while for $F_2$'s, it is $\frac{H_3}{H_0}$.
6. Proportion of permuted data sets with an LR greater than or equal to the observed LR.
7. Actual count of the number of permuted data sets with an LR greater than or equal to the observed LR.

In each step of the permutation test, this file is rewritten and the number following the ``-perm'' token incremented. This way, if the computer crashes during a run, Zmapqtl can be restarted from where it left off. If you were running Zmapqtl with 1,000 permutations, and the process stopped at 899 as above, then restarting Zmapqtl with 1,000 permutations will begin with permutation 900 and continue to 1,000.

The second file, qtlcart.z6e, will contain two columns of numbers: the permutation and the maximal likelihood ratio over the genome in that permutation. Each permutation will add a line to the output. When enough permutations have been done, Eqtl can be run to summarize the experimentwise levels. A small table will be written to the log file that looks like:

-start
 Performed 899 permutations of the phenotypes and genotypes
Here are the Experimentwise significance levels for different alpha
Permutation significance level for alpha = 0.1   : 11.6858
Permutation significance level for alpha = 0.05  : 13.3108
Permutation significance level for alpha = 0.025 : 14.6669
Permutation significance level for alpha = 0.01  : 16.8008
-end of shuffling results

For each shuffle, the largest likelihood ratio test statistic over all test positions is saved in the file. At the end of the shuffling, these maximum values are sorted, and the $(1-\alpha) \times 899$ th largest is the experimentwise significance level for a test of size $\alpha$. The number of permutations can be changed from 899 to any integer from 0 to 10,000. This upper bound could be made higher by changing the appropriate definition in the Main.h source file and recompiling. In general, we find that 1000 permutations is a sufficient number. In a test, values of 1000 and 17,000 were used with little difference in the ultimate comparisonwise and experimentwise values.