Figure 1.

Performance of different classifiers. Performance of different classifier variants under cross-validation. Left panel: boxplots of maximum F1 measure. Left to right: 8 original kernels from [1], extended kernel set, combined subcellular location and functional prediction, pairwise learning of the kernel weights (in the combined model). Right panel: precision, recall and F1 for the pairwise classifier used as "building block" in the HMM under variation of the threshold τ which controls the label multiplicity.