There was substantial agreement among the filtering methods in which probe sets were retained when the thresholds were chosen to retain comparable numbers of probe sets (Table 1). Between 85 and 89% of the probe sets retained with the fraction Present filter were also retained when filtering by minimum MAS5 signal. The agreement was less when filtering by RMA values, with only 77% of the probe sets in common. Similar findings are seen for the vitamin A data (Table 1). Very few probe sets with p ≤ 0.001 are lost by filtering and the overlap between signal and fraction Present filtering is better with MAS5 data than RMA data (Table 2). [Note that the signal and RMA values that remove similar numbers of probe sets are quite different for the vitamin A data than for the IFN data.] Increasing the stringency of filtering decreases the number of probe sets with nominally significant p-values, as expected, but the more significant probe sets (p ≤ 0.01 and p ≤ 0.001) are the least affected (Fig. 4, Table 2).
