Figure 3

Performance as measured by AUC-ROC. For each SCOP family, we run SVM and DT models T times (see Methods) over the same positive set, but over several different negative sets. We calculated the AUC-ROC for each run and the AUC-ROC of SCOP family by averaging the AUC-ROC on all runs. We plot the AUC-ROC versus the number of families that achieved a given AUC-ROC score or better for the database S _full (A) and for the database S ₃₀ (B). We shows only ILP models that achieved best performance. We carried out rank-sum test [38] to compare the curves, see Table 4. For both databases ILP-SVM-Seq-Aln _cons -Aln _ps outperformed all methods, except SVM-LA, and ILP-DT-Seq-Aln _cons -Aln _ps outperformed only HMMer-3.0.