Figure 7

Comparison of prediction abilities of miRPara and other miRNA prediction software. The predictive ability of MiRPara was compared with four other current miRNA prediction software packages (BayesMiRNAFind, MiRAlign, MiREval and MiRPara). Software was tested with three data sets to test (i) the ability to correctly predict experimentally verified known miRNA (Known) (ii) make predictions for new miRNAs from a set of sequences that were new submissions in version 14.0 of miRBase and which had not been previously presented to any of the software packages (New) (iii) reject negative sequences containing no miRNA (Negative) Full details of how the datasets were generated are given in Materials and Methods. miRPara outperformed all other packages in all three categories with the exception of the positive prediction rate of miREval, but this package automatically searches miRBase for homology to every test sequence and thus matches every sequence. When presented with the new positive sequences, the performance is significantly worse.