To determine if filtering would also improve other types of analyses, we analyzed the IFN MAS5 data either unfiltered or after filtering by fraction Present with thresholds of >0%, 25% and 50% using Significance Analysis of Microarrays (SAM) [19]; SAM uses permutations to calculate an FDR (q-value). Filtering by fraction Present increased the number of probe sets deemed significant at each FDR, demonstrating that filtering is beneficial for this alternate method of determining significant changes. Comparing fraction Present >0% to unfiltered data, there was a 35% improvement in the number of probe sets significant at 1% FDR. In this 10-sample experiment, more restrictive filtering (25% and 50%) did not result in appreciable improvement over >0% Present.
