From Protein Structure to Function with Bioinformatics.pdf

70 A. Fiseralignments are built and assessed by a variety of criteria, partly depending on anatomic statistical potential. In another approach, a genetic algorithm was applied toautomatically combine templates and alignments. A relatively simple structuredependent scoring function was used to evaluate the sampled combinations. Despitesome limitations, the procedure is shown to be robust to alignment errors, whilesimplifying the task of selecting templates (Contreras-Moreira et al. 2003).Other attempts to optimize target-template alignments include the Robettaserver, where alignments are generated by dynamic programming using a scoringfunction that combines information on many protein features, including a novelmeasure of how obligate a sequence region is to the protein fold. By systematicallyvarying the weights on the different features that contribute to the alignment score,very large ensembles of diverse alignments are generated. A variety of approachesto select the best models from the ensemble, including consensus of the alignments,a hydrophobic burial measure, low- and high-resolution energy functions, and combinationsof these evaluation methods were explored (Chivian and Baker 2006).Those meta-server approaches that do not simply score and rank alternativemodels obtained from a variety of methods but further combine them could also beperceived as approaches that explore the alignment and conformational space for agiven target sequence (Kolinski and Bujnicki 2005).Another alternative for combined servers is provided by M4T. The M4T programautomatically identifies the best templates and explores and optimallysplices alternative alignments according to its internal scoring function thatfocuses on the features of the structural environment of each template (Fernandez-Fuentes et al. 2007b).Meta-ServersRecently Meta-server approaches have been developed to take advantage of thevariety of other existing programs. Meta-servers collect models from alternativemethods and either use them for inputs to make new models or look for consensussolutions within them. For instance FAMS-ACE (Terashi et al. 2007) takes inputsfrom other servers as starting points for refinement and remodelling after whichVerify3D (Eisenberg et al. 1997) is used to select the most accurate solution. Otherconsensus approaches include PCONS, a neural network approach that identifies aconsensus model by combining information on reliability scores and structuralsimilarity of models obtained from other techniques (Wallner et al. 2007). 3D-JURY operates along the same idea, its selection is mainly based on the consensusof model structure similarity (Ginalski et al. 2003).3.2.4.2 Template Independent Modelling: Modelling Loops, InsertionsIn comparative modelling, target sequences often have inserted residues relative to thetemplate structures or have regions that are structurally different from the correspond-