NSiteMatch: Prediction of Binding Sites of Nucleotides by Identifying the Structure Similarity of Local Surface Patches

Figure 3

The success rates (y-axis) of the NSiteMatch and the three competing methods (Findsite, MetaPocket, and Q-SiteFinder) measured using (the minimal distance from the center of the predicted site to the center of the ligand) on the benchmark datasets. A given binding site is regarded as correctly predicted if the minimal distance between this site and the top predictions is below the cutoff distance D (x-axis), where is the number of binding sites of the protein that includes the evaluated binding site. All methods are evaluated at 4 filter levels, the 40% sequence similarity level ((a), (e), and (i)), family level ((b), (f), and (j)), superfamily level ((c), (g), and (k)), and fold level ((d), (h), and (l)). (a), (b), (c), and (d) show results for the ADP. (e), (f), (g), and (h) show results for the ATP and (i), (j), (k), and (l) show results for the AMP. The 40% sequence similarity level indicates that all chains in the template library that were used for the prediction share less than 40% sequence similarity to the test protein. The family, superfamily, and fold levels indicate that all chains in the template library that were used for the prediction are classified as belonging to a different family, superfamily, and fold (annotated using the SCOP database), respectively, when compared with the annotation of the test protein.

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

(i)

(j)

(k)

(l)