Evaluation using differential PAR-CLIP. KSHV miRNA target sites should only have reads in KSHV-infected cell lines (a-c), and EBV miRNA target sites should be exclusive to EBV infected cell lines (d-f). PARma-assigned KSHV miRNA target sites have a higher fraction of exclusive sites than any other method (a, d) (see main text for a description of the other methods) and when PARma was run without being constrained to known miRNA seeds, it yielded a higher fraction of exclusive sites than PARma using seeds as priors. (b) and (e) show the log fold changes (control/infected) of PAR-CLIP read counts for clusters assigned to KSHV and EBV miRNAs, respectively. The log fold change of exclusive clusters (that is, clusters that have no reads in one of the experiments) has been set to -10 or 10. PARma not only has the largest fraction of exclusive clusters in both datasets (compare the left ends of (b) and (e) to (a) and (d), respectively) but it also has the smallest fraction of KSHV or EBV clusters that have more reads in the KSHV or EBV negative cell line. The dependency of scores on the accuracy is shown in (c) and (f). In both datasets and for both scores, accuracy increases as low scoring clusters are removed. As a reference, the accuracies of the other assignment methods are indicated with the same colors as in (b) and (e). EBV: Epstein-Barr virus; KSHV: Kaposi's sarcoma-associated herpesvirus.
Erhard et al. Genome Biology 2013 14:R79 doi:10.1186/gb-2013-14-7-r79