In contrast to removing probe sets that represent genes not expressed, setting a fold change limit by itself does not appear to increase the likelihood that a change is in fact real. Absent probe sets with low signals have an increased probability of generating spurious large fold changes, especially in smaller experiments (3–5 samples). For these small experiments, the probe sets remaining after 50% Present filter generate a modest number of fold changes ≥2 of which 71–84% are called significant in the full 10-sample experiment at p ≤ 0.01. On the other hand those probe sets removed by the 50% filter generate about 4 times as many fold changes ≥2, of which only 6 to 10% are called significant in the 10-sample experiment at p ≤ 0.01.
