We carried out a second set of permutations this time randomly selecting equal numbers of samples from each treatment group to create two new groups expected to produce no true positives. For these tests of 4, 6 and 8 samples in each treatment group, a disproportionate number of the false positives were found in the probe sets that are Absent for all arrays (Table 6). the number found significant was less than expected for a normal distribution, results that are similar to the results of permutation tests on the full data set.
