Fig. 3

ROC curves showing predictive performance of the comparative methods on the test set of simulated short genomes. Comparison was carried out between DotcodeR and RNApdist. Here we used the window size of 120 nt and the step size of 30 nt. Note that accuracy was calculated by averaging over all results of the families in the dataset. The terms ‘gene-shuffled’ and ‘genome-shuffled’ refer to how the negative datasets were generated