From Protein Structure to Function with Bioinformatics.pdf

60 A. FiserIn contrast to ab initio techniques comparative protein structure modellingusually provides models that are comparable to low resolution X-ray crystallographyor medium resolution NMR solution structures. However, its applicabilityis limited to those sequences that can be confidently mapped to known structures.Currently, the probability of finding related proteins of known structurefor a sequence picked randomly from a genome ranges approximately from 30%to 80%, depending on the genome. Approximately 70% of all known sequenceshave at least one domain that is detectably related to at least one protein ofknown structure (Pieper et al. 2006). This fraction is more than an order of magnitudelarger than the number of experimentally determined protein structuresdeposited in the Protein Data Bank (Berman et al. 2007). The applicability ofcomparative modelling is steadily increasing because the increasing number ofexperimentally determined novel structures. This trend is accentuated by theProtein Structure Initiative (PSI) project that aims to determine at least onestructure for each protein families (Burley et al. 2008; Vitkup et al. 2001). Aftera five year period of feasibility tests and technology build-up (PSI-1, years2000–2005), these structural genomics efforts are now in “production phase”(PSI-2, years 2005–2010) and it is conceivable that their aim will be substantiallyachieved in less than 10 years, making comparative modelling applicableto most protein sequences.As we will see, in practice, template based modelling always includes informationthat is independent from the template, in form of various force restraints fromgeneral statistical observations or molecular mechanical force fields. As a consequenceof improving forcefields and search algorithms the most successfulapproaches are more and more often explore template independent conformationalspace (R. Das et al. 2007; Y. Zhang 2007). Similarly, the most successful ab initioapproaches, in fact, are using fragments of known structures to build up models(Bystroff and Baker 1998; Zhou et al. 2007). While it makes sense to discuss thetwo fundamental principles behind the techniques employed in structure modellingseparately, the current trends are pointing to approaches that extensively combineboth. While truly ab initio approaches can shed light on the dynamics of the actualfolding process, in practice, effective structure modelling almost always involves acertain flavour of template-based modelling.3.2 Steps in Comparative Protein Structure ModellingComparative or homology (template-based) protein structure modelling builds athree-dimensional model for a protein of unknown structure (the target) based onone or more related proteins of known structure (the templates) (Blundell et al.1987; Fiser 2004; Ginalski 2006; Greer 1981; Marti-Renom et al. 2000; Petrey andHonig 2005). The necessary conditions for getting a useful model are (i) detectablesimilarity between the target sequence and the sequence of the template structureand (ii) availability of a correct alignment between them.