Figure 3

Performance of the evaluated methods according to the ROC curves. A match is defined as a true positive if a corresponding structural model gets TM-score ≥ 0.4, and a false positive if TM-score ≤ 0.17. HHsearch was tested both with and without (w/o ss) consideration of secondary structure information. Numbers in parenthesis indicate relative area under the corresponding ROC curve. (A) Global mode analysis, in which alignments are evaluated in respect to the entire structural domains. Empty squares mark approximate transition points, where false positives start accumulating more rapidly. Corresponding E-values for COMA, COMPASS, PSI-BLAST and HHsearch w/o ss are 0.01, 0.03, 0.0007 and 3e-6 respectively. HHsearch probability is 97.6% (B) Local mode analysis; evaluation is done within the boundaries of aligned segments; E-values at transition points for COMA, COMPASS, PSI-BLAST and HHsearch w/o ss are 8e-6, 1e-4, 1e-5, and 0.65 respectively. HHsearch probability is 65.3%. Correlation between the estimated and observed statistical significance values for each method from the purely statistical perspective without differentiating global and local modes is shown in Additional file 1, Figure S3.