Figure 1
From: Identification of novel DNA repair proteins via primary sequence, secondary structure, and homology
GO-PDB identification experiments at 50% sequence similarity. In each line at the bottom of the graph, the computational method (see Table 1 for abbreviations) used to identify DNA repair proteins is given. Methods other than BLAST used SVMs, and the value(s) of the primary sequence and (where applicable) secondary structure spectrum kernels, as well as experimentally optimal SVM parameters are listed after the transformation method (e.g., 11R PSH-3–8 means that method PSH used a primary sequence 3-spectrum kernel, a secondary structure 8-spectrum kernel, and homology with 11R cross-validation; see Methods for the 11R method). To the right of each method, AUC values and true positive rates when allowing maximums of no false positives (FPs), 1% FPs, and 5% FPs, respectively, are shown.