Figure 3

Performance of PROmiRNA based on a comparison with PolII ChIP-seq data from the ENCODE project. (a) ROC computed by varying the posterior probability cutoff for four promoter classes and two sets of random intergenic and intronic regions. The '+' symbols correspond to the recall values and false positive rates at the decision cutoff of our binary classifier, p(prom) = 0.5 for the four promoter classes. (b) Precision-recall curve computed by varying the posterior probability cutoff c for four promoter classes and two sets of random intergenic and intronic regions. The horizontal lines correspond to the random case. The '+' symbols correspond to the recall and precision values calculated at a cutoff c = 0.5, which represents the minimum required posterior probability p(prom) from our classifier to classify candidate regions as belonging to the promoter class. The precision of the method in identifying promoters at a cutoff of p(prom) = c = 0.5 is equal to 83%, 92%, 78% and 75% for all, host gene, intergenic and intronic promoters, respectively.

ROC, receiver operating.