Naef affinities for empty probesets. The probe positions affinities using Naef's model [5] for adenine (red), thymine (blue), cytosine (black) and guanine (green) of empty PM and MM probes in the GoldenSpike dataset. The sum of the affinities as each position is zero, and the affinity of a probe, is based on the sum of the single nucleotide affinities at each position for that probe.

MAS 5.0 present/absent calls. (a) The average frequency of P values of empty probesets generated by the MAS 5.0 present/absent algorithm when τ = 0.015 (solid black line) and when τ = 0 (dotted line). The average was taken over the six samples. The percentage of central nucleotides in PM probes for empty probesets with P values < 0.06, for all empty probesets (similar percentages are present in all probesets), and for empty probesets with P values > 0.94 are shown. (b) P values generated with the Wilcoxon signed rank test for random empty probesets. The PM-MM probe-pairs from empty probesets with fewer than six alignment errors to any transcript in the GoldenSpike dataset were randomly re-assembled into probesets based on the central nucleotide (for example, only central T nucleotides in the PM probes). Symbols and lines are colored according to the central nucleotide.

AUC performance for present/absent calls. AUC scores for 301 methods to generate probeset expression values (see Materials and methods and [8] for more information) based on the mean log₂ value of each probeset for control (C) samples (rainbow colors as in legend). The performance of a method for spiked-in (S) samples (gray) is shown in the same column. True positives are probesets that can be aligned to the DGC clones that were used to create the GoldenSpike dataset. False positives are the remaining empty probesets. The horizontal lines indicate the AUC scores for the mean MAS 5.0 present/absent P value for C replicates (blue) and S replicates (gray).

Log₁₀ of mean MAS 5.0 present/absent P value versus mean log₂ expression value. (a) Plot of log₁₀ mean MAS 5.0 present/absent P value versus mean log₂ expression value for control (C) replicates in the GoldenSpike dataset. The default P values for absent (0.06) and marginal (0.04) are marked with vertical lines (blue), and suggested expression value cutoffs (99% probesets with P values > 0.06 are below the log₂ cutoff) are marked by a dotted horizontal line (green). (b) ROC curves for C replicates using the best performing expression value method (black) and MAS 5.0 P values (red). Vertical lines correspond to the false positive rate of cutoff values suggested in (a). (c) Plot of log₁₀ of mean MAS 5.0 present/absent P value versus mean log₂ expression value for experimental replicates in the Latin-Square dataset. There are 42 spiked-in transcripts that can be aligned to probesets across a range of 14 different concentrations. The number of probesets called absent (MAS 5.0 P value < 0.06) is marked in the legend. Lines are as in (a).

AUC performance of alternative methods. (a) AUC scores for the MAS 5.0 present/absent algorithm when raw (black) or GC-RMA transformed (red) PM and MM are used. Scores were generated with a range of τ values. (b) AUC scores when MM probe values are replaced by a GC-RMA transformed PM value. The threshold values are based on the mean value of PM probes from probesets that have absent target transcripts. Scores using different stringencies of finding absent transcripts are shown (P > 0.06, black; P > 0.25, red; P > 0.50, green; P > 0.75, blue). (c) ROC curves based on MAS 5.0 present/absent P values when using raw probe values (gray), GC-RMA transformed probe values (black) and threshold probe values (blue dotted line) are shown. Green lines indicate the true (horizontal) and false (vertical) positive rate for the default P value cutoff (0.06) to separate 'marginally present' and absent transcripts using raw PM and MM values. AUC and ROC curves are based on the mean P value of control replicates.

MAS5 present/absent calls for alternative methods. (a) The average frequency of P values of probesets with present target transcripts generated by the MAS 5.0 present/absent algorithm when using raw PM and MM values (gray), GC-RMA transformed PM and MM values (black) and GC-RMA PM threshold values (blue) are shown. (b) The average frequency of P values of probesets with absent target transcripts (that is, empty probesets) generated by the MAS 5.0 present/absent algorithm.