From Protein Structure to Function with Bioinformatics.pdf

11 Function Predictions of Structural Genomics Results 27511.2 Large Scale Function Prediction Case StudiesConsidering the fact that structural genomics initiatives have been producing vastnumbers of structures over a number of years now, there have been surprisingly fewstudies into structure-based function prediction. Here we review various attemptsto address the effectiveness of structure-based function prediction using structuralgenomics targets. A summary of these examples and the methods key to their successis provided in Table 11.1.An early review of 15 hypothetical proteins of known structure and their functionalassignment (Teichmann et al. 2001) provided some glimpses of the qualityof functional assignments that can be made from structure. The structures in conjunctionwith alignments of homologous sequences were used to find surface cavitiesand grooves in which conserved residues indicated an active site. Using acombination of any bound co-factors in the structures and available experimentaldata for the protein in question or related sequences, assessments were made as tothe depth of functional information that could be obtained. For the 15 proteinsexamined, detailed functional information was obtained for a quarter of them, somefunctional information could be obtained for half of them, and no functional informationcould be obtained for the remaining quarter.In 2003, Kim et al. published an analysis of eight structures, some solved at theBerkeley Center for Structural Genomics, others with collaborators, where the proteinstructure provided functional or evolutionary insights. The examples demonstratedin the paper were classed by the authors into one of five categories:1. Remote homologues. Here function is inferred from structural similarity thatcould not be observed through sequence. In this example they use the case ofMJ0882 which was initially suggested to be a putative methyltransferase throughfold similarity and later experimentally verified (Huang et al. 2002).2. Proteins with unexpected bound ligands. Here function can be inferred bythe chance binding of a substrate or cofactor. The first example, MJ0577from Methanococcus jannaschii, involved the identification of a bound ATPin the structure, thus suggesting an ATP hydrolysis function. More detailedanalysis of the ATP-binding pocket of MJ0577 indicated the presence ofsome common motifs found in nucleotide-binding proteins, but the sequentialarrangement of them differed from existing examples, so that motifbasedmethods were unable to detect them. Subsequent experimental assaysconfirmed an ATP hydrolysis function but only in the presence of cellextract, thus suggesting the protein is a molecular switch that requires one ormore partner proteins.The second example, TM841 from Thermotoga maritima, is a member of thelarge DegV Pfam family and is a member of the COG1307 group which hasunknown function. Solution of the structure of TM841 showed the presence ofa bound palmitate molecule, thus demonstrating fatty acid binding. Comparisonof other members of the DegV and COG1307 families indicated a greater degreeof conservation in the region binding the head group of the palmitate and more