Figure 3
Work flow of target gene prediction. Three programs (miRanda, RNAhybrid, SeedMatch) were used separately to predict target sites. The energy score and p-values of RNAhybrid were supplemented by an alignment score calculated from the additionally provided alignment string. The overall six scores were treated as different predictors, motivated by a poor Pearson correlation (R2 below 0.5) of the scores within each program. Within each input, the predicted target sites were filtered ("Classic", "Vari", "3000" filter) and combined to a final prediction of microRNA target sequences. For the resulting sets, the overlap of the different inputs was evaluated with an unweighted voting process. The parameter n of the voting process was set to four with an additional conservation filter, otherwise, the filter was set to five. A detailed description of target prediction is provided in Additional file 5.