Figure 3

ROC analysis for comparing the ability of upstream or downstream ChIP sites to predict induced genes. In each experiment, the induced genes (positives) are taken as the 1% highest ranking transcripts. The remaining 99% are taken as the negatives. Note that for discrete data such as the number of sites in a specific genomic window, ROC analysis consists in a set of operative points (cf. Methods). The number of sites downstream of the TSS (light curves) shows the best performance among all the definitions tested (high stringency and low stringency ChIP-chip, ChIP-pet). Shown are the cancer expression compendium [12] and the study on primary estrogen receptor targets [20]. Further expression sets are shown in [see Additional file 3].