Figure 2

CDFs of cross-platform distances. a) CDF of gene-level distances for the Affymetrix/Agilent combination. b) CDF for probe-level distances for the Affymetrix/Agilent combination. c) CDF for gene-level distances for the Affymetrix/Codelink combination. d) CDF for probe-level distances for the Affymetrix/Codelink combination. e) CDF for probe-level distances for the Agilent/Codelink combination (probe-level after cross-hybridizing probes were removed). Genes with equal transcripts show higher agreement than genes with disjoint transcripts in terms of distance between log2 fold change expression vectors across platforms, as shown in 2a and 2c. The equal CDF rises faster and is farther to the left because a higher percentage of distances are small. Disjoint transcripts are more agreeable than random genes for the same reason, indicating transcript forms are regulated together better than random. Individual probe distances, shown in 2b, 2d and 2e, show the same pattern as the gene-level CDFs with equal, disjoint and random following each other in decreasing agreement. Affymetrix disjoint probes are not much improved over random, unlike the gene-level CDFs, indicating Affymetrix's reliance on redundant probe sets to obtain accurate expression data