Figure 1

Cleaner on 152 B-cell samples profiled on U95Av2 chips. (a) Probe consistency scores for four genes (distinguished by color) and their corresponding null density distributions. Dots represent probes and are plotted according to consistency scores (horizontal axis) and distances from the transcript end (vertical axis). Solid lines depict null density distributions and dotted vertical lines are drawn at their 99 percentile. Consistent probes are to the right of their respective dotted lines. (b) Unsupervised hierarchical clustering of probes mapping to MAX (all isoforms); naturally occurring probe clusters are highlight in red. (c) Relative distance between each consecutive cluster in the dendrogram in (b); the right-most point represents the distance between highlighted clusters in (b). (d) Three known isoforms for MAX and the mapping positions of probes belonging to the two Cleaner probe clusters for MAX; probes from probe cluster 4149.2 are mapped to the splice-variant fifth exon, and probes from 4149.1 measure a convolution of the two transcripts. (e) The mean of Pearson correlations between overlapping and neighboring probes depends on the distance between them, and it is closely modeled by an exponential function. (f) False discovery rate (FDR) as a function of the consistency score, as estimated by permutation testing; no probe-cluster with consistency scores higher than 4 were identified in permuted clusters. (g) Probe-cluster consistency scores are correlated to their MAS5-assigned intensity, as measured before pruning. However, the intersection between distributions for the 4,702 consistent probe clusters (red), 3,708 inconsistent probe clusters (blue), and all probe clusters (black line) suggests that probe-cluster intensity does not perfectly predict probe-cluster consistency.