From Protein Structure to Function with Bioinformatics.pdf

8 3D Motifs 191To improve the signal for detecting functionally relevant motifs, residue conservationin sequence alignments and spatial clustering of the residues in a given motifare often considered as well.8.2.2 Motif Description and MatchingPoints in a 3D motif are either atoms or pseudoatoms derived directly from theatom positions of a structure. A side chain centroid, for example, is simply a pseudoatomat the average position of the atoms in the side chain. Up to a few pointsare used per residue in the motif, and the points are labelled with additional informationsuch as atom type, residue type, or physicochemical characteristics.When a structure is searched for matches to a 3D motif, qualitative rules governwhich points in the structure are allowed to pair with which points in the motif, andgeometric cutoffs determine which sets of points are sufficiently spatially similar tobe considered a match, or hit. Match stringency also depends on the numbers of residuesand points in the motif. There is a tradeoff between match stringency and tolerance:it may be desirable to allow for residue substitutions, conformational flexibility,and low structural resolution, but doing so will increase the number of biologicallymeaningless hits along with the hard-to-find meaningful ones. Including specificatom positions in a 3D motif emphasizes localized interactions such as hydrogenbonding, whereas using centroids of functional groups or side chains is more accommodatingof flexibility and type substitutions (Fig. 8.1). Representing side chaingroups that are symmetrical, e.g., the aromatic ring in Phe, as a single point alsoprecludes having to compare them in multiple ways (Oldfield 2002).Searching can be computationally intensive, especially considering that thousandsof structures may be compared to thousands of motifs. Three-dimensionalmotif searching has relied on the development of efficient algorithms, often involvingone or more of the following:●Geometric hashing. Hashing is a somewhat broad term for reducing complexdata to a simpler form that can be compared more rapidly. Multiple values suchas distances, angles, and residue types can be collapsed with a function intofewer numbers or even a single number. Other sets of values that reduce to thesame result correspond to potentially matching substructures. Geometric hashingencodes spatial relationships among points (Fischer et al. 1994), but otherkinds of information such as physicochemical descriptors can be included(Shulman-Peleg et al. 2004). Individual substructure matches that imply similartransformations (translations/rotations to superimpose the paired points) can becollated into larger groups before the more computationally intensive steps oftransformation and scoring are performed (Pennec and Ayache 1998). Hashingor preprocessing the data takes time, but only needs to be done once per structureand can greatly speed up comparisons.