From Protein Structure to Function with Bioinformatics.pdf

3 Comparative Protein Structure Modelling 59Pillardy et al. 2001). The ab initio methods assume that the native structure correspondsto the global free energy minimum accessible during the lifespan of theprotein, and attempt to find this minimum by an exploration of many conceivableprotein conformations (Dill and Chan 1997; Sali et al. 1994).The second class of methods, called template-based modelling, includes boththose threading techniques that return a full three dimensional description for thetarget (J. Xu et al. 2007) – see also Chapter 2 – and comparative modelling (Fiser2004). This class relies on detectable similarity spanning most of the modelledsequence and at least one known structure. Comparative modelling refers to thosetemplate-based modelling cases when not only the fold is determined from a possibleset of available templates, but a full atom model is built (Marti-Renom et al.2000). When the structure of at least one protein in the family has been determinedby experimentation, the other members of the family can be modelled based ontheir alignment to the known structure. Comparative modelling approach to proteinstructure prediction is possible because a small change in the protein sequence usuallyresults in a small change in its 3D structure (Chothia and Lesk 1986). It is alsofacilitated by the fact that 3D structure of proteins from the same family is moreconserved than their amino-acid sequences (Lesk and Chothia 1980). Therefore, ifsimilarity between two proteins is detectable at the sequence level, structural similaritycan usually be assumed. The increasing applicability of comparative or template-basedmodelling is due to the observation that the number of different foldsthat proteins adopt is rather limited (Andreeva et al. 2008; Chothia et al. 2003;Greene et al. 2007).Both of these approaches to structure prediction have their advantages and limitations.In principle, ab initio approach can be applied to model any sequence.However, due to the complexity and our limited understanding of the protein foldingproblem, ab initio methods usually result in relatively low resolution models.Despite significant progress in ab initio protein structure prediction (R. Das et al.2007), it remains applicable to a limited number of sequences of approximately 100residues. Benchmarks indicate that ab initio techniques still cannot get the overallfold correct for the majority of targets (Jauch et al. 2007). Our increasing understandingabout the accuracy and performance of currently available forcefields andsampling techniques should be acknowledged as being due, in substantial part, tothe stunning improvement in computational capacity. To further exploit thisresource several “largest ever” studies took off recently that expected to providefurther critical insights into the folding process. These involve among others theRosetta@home (http://boinc.bakerlab.org/rosetta/), Folding@home (http://folding.stanford.edu/) and the IBM supported Blue Gene projects. In the Rosetta@homeand Folding@home projects the process of protein folding or modelling is studiedby running simulations on voluntarily contributing private computers, connectingup to a million CPUs worldwide. IBM aims at a similar scientific target by buildingBlue Gene, a computer farm of processors with an estimated 596 teraflops peakperformance. Currently various flavours of Blue Gene computers occupy a total offour of the top ten positions in the TOP500 supercomputer list announced inNovember 2007 (http://www.research.ibm.com/bluegene/).