Figure 2

Consistency across technical replicate experiments. Density distributions for the Spearman correlation coefficient across technical replicates considering: (a) all the probe clusters generated by each method (size in parenthesis); (b) only the 1,000 probe clusters with the highest sample variation for each method; and (c) only probe clusters corresponding to genes that were detected by all methods. Cleaner probe clusters show dramatically better agreement across technical replicates even after pruning out low variability probe sets (b), and when restricting the comparison to genes that are present in all annotation strategies (c). Remap-only probe sets were statistically indistinguishable from AffyProbeMiner probe sets; both probe sets showed significantly better agreement across technical replicates than Affymetrix probe sets in (a, c), but the three were statistically indistinguishable after pruning out low sample variation probe sets (b).