From Protein Structure to Function with Bioinformatics.pdf

8 3D Motifs 205folds; they shared a 3D motif of atoms from a four-residue backbone segment andthree sequentially separated residues (Kobayashi and Go 1997). A similar approachwas used to generate consensus binding-site motifs (Nebel et al. 2007). By the timeof this study, many more structures had become available, and adenine mono-, di-,and tri-phosphate complexes were handled separately. Pairwise similarities wereevaluated as the fraction of atoms near ligand that could be assigned to correspondingpairs. Structures were clustered using these similarity values, and outliers wereremoved. Within each cluster, only atoms common in all pairwise comparisonswere retained, and their positions were averaged to generate a 3D motif. Finally,highly similar motifs were merged. The resulting 13 motifs are derived from 3 to20 structures, contain 6 to 71 atoms, and correspond in most cases to some knownclassification of structure or function. The motif coordinates are available as supplementaryinformation to the publication (Nebel et al. 2007).Consensus templates were developed for porphyrin-binding sites usingNestor3D (Nebel 2006). Templates generated by this program can include atoms,pseudoatoms representing functional groups, and “solvent” (actually points on agrid to represent the cavity volume). Nestor3D includes a graphical interface andis available for download (Table 8.2). Users must specify a list of PDB files andwhich atoms to fit to superimpose the structures; several other parameters areoptionally adjustable.An all-by-all comparison of 3,737 phosphate environments from protein-nucleotidecomplexes allowed classification into 476 tight clusters and ten broader groupings(Brakoulias and Jackson 2004). The clusters generally agreed with classificationsbased on global structure or function. An efficient clique detection method wasused to find corresponding sets of atoms, so it was not necessary to use ligandatoms to superimpose the structures.SOIPPA (Sequence Order-Independent Profile-Profile Alignment) finds sharedpatterns of local structure in pairwise comparisons (Xie and Bourne 2008). A proteinstructure is reduced to its alpha-carbons, each associated with a geometricpotential value and a profile of allowed substitutions obtained from an automaticallyconstructed sequence alignment. The geometric potential of an alpha-carbonis calculated from its distance to the surface and the arrangement of neighbouringalpha-carbons (Xie and Bourne 2007). A possible match between two structuresstarts with a pair of points with similar geometric potentials; neighbouring pairs canbe added if their distances and surface normal angles are consistent. Each alphacarbonpair is weighted by the similarity of their substitution profiles, and a maximumweightcommon subgraph is found. The alignment score after superposition is asum over pairs incorporating pair weight, coincidence in space, and angle betweensurface normals. Statistical significance is estimated with a nonparametric model ofthe distribution of scores when the pattern is compared to a representative set ofstructures. SOIPPA was used to compare diverse adenine-binding structures and tosearch a representative set of structures for matches to known functional sites; itwas better able to align binding sites and identify local similarities than were globalsequence or structure comparisons. This work focused more on identifying relationshipsthan on 3D motif discovery.