Skip to main content
. 2010 Feb 3;107(8):3481–3486. doi: 10.1073/pnas.0914097107

Fig. 2.

Fig. 2.

The average AUC of ROC curves of identifying structural neighbors. The AUC measures how well a ranking of structures imitates the ranking according to a gold standard; larger values correspond to more successful imitators, ranging from 0.5 (a random ranker) to 1 (a perfect imitator). We consider three definitions of structural neighbors, using SAS thresholds of 2 Å, 3.5 Å, and 5 Å. The left panels show the performance of libraries with fragments of 6–12 residues, and different number of fragments (value along the x-axis), and using the cosine (plus sign), Euclidean (circles), and Histogram Intersection (diamonds) distances. On the right we compare the best FragBag result (400(11) library and the cosine distance) to other methods: Sequence-based similarity measure in finely dashed black, filter methods in dashed black, and structure alignment methods in solid black. We see that the FragBag performs similarly to CE and STRUCTAL—two computationally expensive and well-trusted structural aligners.