From Protein Structure to Function with Bioinformatics.pdf

58 A. Fiserwith more than 50% sequence identity, transfer of annotation between homologues leadsto an erroneous attribution with a totally dissimilar function in 6% of cases.Functional characterization of a protein is often facilitated by its three-dimensional(3D) structure. The insight that one may gain from a 3D model ranges fromsuch low level functional descriptions as confirming the fold (G. Wu et al. 2000)and inferring a general functional role, to such high resolution descriptions asunderstanding ligand specificities and designing inhibitors in the context of structurebased drug discovery (Becker et al. 2006; Evers et al. 2003).3.1.2 Sequences, Structures, Structural GenomicsGenome scale sequencing projects have already produced around six millionunique sequences to date (Apweiler et al. 2004; C. H. Wu et al. 2006), a numberthat would double if metagenomic data from Craig Venter’s Global Ocean Surveywas added to public databases (Rusch et al. 2007; Venter et al. 2004; Yooseph et al.2007). Meanwhile only ∼50,000 of these proteins have their three-dimensionalstructures solved experimentally using X-ray crystallography or Nuclear MagneticResonance (NMR) spectroscopy (Berman et al. 2007). Because of the inherentlytime consuming and complicated nature of structure determination techniques, thefraction of experimentally known 3D models is expected to further shrink from thecurrent level of less than 1%. Computational approaches need to be employed tobridge the gap between the number of known sequences and that of 3D models.Structural genomics projects were launched worldwide around the year 2000.One key aim is the experimental solution of the three dimensional structures of acarefully selected few thousand target sequences of structurally uncharacterizedproteins. These newly solved structures could be used as templates for computationalmodelling of about 100 times more proteins whose sequences are related(Burley et al. 1999). These worldwide structural genomics efforts are becoming thedominant source of experimentally solved protein structures. In recent years 75%of new folds deposited to the PDB emerged from these efforts (Burley et al. 2008).Meanwhile these efforts further underline the importance of theoretical approachesto structure modelling, since more than 99% of all three dimensional models willbe obtained computationally (Manjasetty et al. 2007).3.1.3 Approaches to Protein Structure PredictionThe study of principles that dictate the three-dimensional structure of natural proteinscan be approached either through the laws of physics or the theory of evolution.Each of these approaches provides foundation for a class of protein structureprediction methods (Fiser et al. 2002).The first approach, ab initio or template-free modelling methods, discussed inChapter 1, predicts the structure from sequence alone (Bonneau and Baker 2001;