Performance of our algorithm on publicly available Chip2 data  is evaluated using the Yeast Proteome Database collection of transcription factor targets [28,29] and compared with another popular means of computing p values . (a) Receiver-operating characteristic curves for our method (black, 'Chipper') and that of Lee et al.  (green, 'Lee') using three replicate experiments for the transcription factor Ino4, made publicly available by Lee et al. (b) Sensitivity as a function of significance threshold. The broken line represents the performance of choosing potential targets at random. (c,d) Analogous curves for the transcription factor Sko1. FP, false positive; TP, true positive.