As another estimate of the number of false positives, we carried out balanced permutations designed such that there should be no true positives. For the IFN data, arrays were randomly selected such that 5 from each actual treatment group of 10 samples were selected for one group and the remaining 10 samples (5 from each group) were used as the second treatment group. For the smoking data, 10 arrays each were randomly selected from the non-smokers and smokers groups for the first treatment group and the remainder used for the second treatment group. Thus each permuted group had equal numbers of arrays from both groups, and the comparison is expected to show no difference. Permutation were carried out 1000 times. For each probe set, totals were kept for the number of times that a probe set met a selected combination of p-value and fraction Present threshold (p-values: 0.05, 0.02, 0.01 and 0.001; fraction Present: no filter, >0%, 25% 50%, 75% and 100%). Experiment-wide totals (total number of probe sets meeting the criteria) are an average of the 1000 permutations.
