Learned global models are better breakpoint detectors than default models. Chromosomes with detailed annotations for neuroblastoma copy number profile i=207 are plotted in panels from left to right, and each row shows the breakpoints detected by a model as vertical black lines. The bottom 3 rows show default models trained using no annotations, and the top 2 rows show global models trained using annotations of other profiles. The equivalent cghseg.k and pelt.n global models show only 2 annotation errors compared to 3 and 11 in the corresponding default models. Furthermore, the equivalent cghseg.k and pelt.n global models perform better than dnacopy.sd, which shows 4 annotation errors.