Global models are better breakpoint detectors than local models. Leave-one-out cross-validation over the annotions on each profile was used to compare breakpoint detection error of global and local models. The two panels show the two annotation data sets, and each row shows the performance of one of the models described in the Algorithms: copy number profile smoothing models section. After selecting the smoothness parameter λ by minimizing either the global or local annotation error, we plot the mean and standard deviation over 10 test sets. Each default model does not have a smoothness parameter, and shows equivalent local and global model error. Squares show the same colors as the other figures and tables, and are absent for default models that have no smoothness parameters.