From Protein Structure to Function with Bioinformatics.pdf

206 E.C. Meng et al.The Common Structural Cliques method was proposed for identifying functionallyrelevant atoms in structures that share a function but are evolutionarily unrelatedor distantly related (Milik et al. 2003). Each protein is reduced to a graph thatincludes only representative atoms from each side chain. Sets of four atoms areextracted and compared to identify common structural cliques, or sets with equivalentinteratomic distances and atom types in both structures. These are merged intolarger sets of corresponding atoms. Notably, the resulting 3D motifs can includedifferent weights for different interatomic distances, even weights of zero. Thisallows matching even when certain distances vary significantly due to flexibility.For example, a motif could include atoms in domains on either side of a hinge. Lowor zero weights for interdomain distances allow a motif to recognize structures withdifferent hinge conformations, while weights for intradomain distances can be kepthigh to specify precise geometric relationships within each domain. A limitation isthat results from pairwise comparisons are not combined automatically.DRESPAT (Detection of REcurring Sidechain PATterns) extracts a shared motiffrom a set of positive example structures (Wangikar et al. 2003). Each protein isreduced to a graph of one functional atom per residue, excluding residues withnonpolar side chains and disulphide-bonded cysteines. Patterns of three or moreresidues are extracted and compared to patterns with the same residue types fromthe other structures, considering alpha- and beta-carbons in addition to the functionalatoms and discarding matches with distance deviations and/or RMSD valuesgreater than specified cutoffs. Other adjustable parameters are the pattern size(default three to six residues) and how many of the input structures must containthe pattern. Based on pattern occurrence in randomly chosen sets of structures,empirical relationships were derived for calculating the statistical significance of adiscovered pattern from its size, the total number of input structures, and thenumber of input structures required to contain the pattern. Results were presentedfor nonredundant sets from 17 SCOP superfamilies. It was found that motifs of atleast four residues derived from sets of five or more structures typically correspondedto functional sites. Too many additional patterns were obtained from onlypairwise comparisons of evolutionarily related structures. DRESPAT is availablefrom the authors as C++ code (Wangikar et al. 2003).The funClust server (Ausiello et al. 2008) (Table 8.1) identifies 3D motifs sharedby multiple input structures. Up to 20 structures can be used. The structures are filteredby sequence identity and then compared all-by-all with Query3D (Ausiello etal. 2005a). Query3D uses two points per residue, alpha-carbon and side chain centroid.Besides the maximum sequence identity, users can specify whether cutoffs inRMSD and side chain proximity should be low, medium, or high; whether buriedor hydrophobic residues should be excluded; and whether similar residue typesshould be allowed to pair in addition to identical ones. The server reports motifs ofthree or more residues found in three or more of the input structures.The PAR-3D (Protein Active site Residues using 3-Dimensional structuralmotifs) server (Goyal et al. 2007) (Table 8.1) compares an uploaded structure tomotifs for six classes of proteases, ten glycolytic pathway enzymes, and metal-bindingsites of three or four residues (Goyal and Mande 2008). The motifs, each generated