Figure 3

Detecting structures from the same class, architecture and topology according to CATH classification. To obtain an idea about the scope and resolution of the prefiltering we are proposing (red), the results are shown on the same graph with representative full resolutions methods. It should be understood here that the prefiltering step we are proposing is some 40 to 1000 times faster than the full resolution methods. (The purpose of full resolution methods, of course, is achieving the high quality of the alignment, rather than the speed database scanning. The quality of pairwise alignment is not tested in this type of experiment.) The results are presented in terms of a ROC curve: for a sliding threshold in the quality score, the number of true positives (TP) above the threshold (y-axis) is shown as function of the fraction of false positives (FP) falling above the threshold (x-axis). Red line: the ROC curve using the total aligned score (Eq. 6) to rank the quality of the match, with δ = 0.5 (full red line) and with δ = 0.3 (dashed red line) and gap opening penalty of -1 in the alignment step. Gray: various high resolution methods (CE, [21]; STRUCTAL, [16]; LSQMAN, [58]; DALI, [17]; SSM, [12]; SSAP, [15]) scored using using SAS score [16]. For the original context, timings, and discussion see Kolodny et al. [38]. Green line: "generation 2000" high resolution methods, in the order of decreasing area under the ROC curve (and, roughly, the time taken for the task): 3Dhit [9](80 CPU hrs), TMalign [13](80 CPU hrs), SABERTOOTH [14](60 CPU hrs), MAMMOTH [8](40 CPU hrs). Green, dash-dotted: SGM [24] (several CPU minutes). Inset: comparison of the method discussed in the text with the pre-filters used in VAST and SSM, on a smaller data set, acceptable to all three methods [Additional File 1]. Red line (full): the ROC curve using the total aligned score (Eq. 6); blue line (full) pre-filter used in VAST; orange line (full): SSM pre-filter optimized for performance on this type of a test; orange line (dashed): native SSM pre-filter.