Performance of trained predictors.

<p>(A) Cross-validation (CV) performance of models trained on all available native IRES sequences shown for different combinations of <i>k</i>-mer lengths, and <i>k</i>-mer count (solid lines) or presence (dashed lines) features (left), with the selected combination marked with a circle. Scatter plot of predicted and true IRES activities for the selected model (middle) coloured according to the local density (blue to red as low to high density). The Receiver Operating Characteristic (ROC) curve and the area under the curve (AUC) for the selected combination. (B) CV performance of models trained for different groups of sequences. Only results for groups with models achieving sufficiently high performance are shown. (C) Training and test performance of the feature and <i>k</i>-mer length combination selected for the group of all native IRESs evaluated using several metrics.