From Protein Structure to Function with Bioinformatics.pdf

28 L.A. Kelley2.1.1 The Importance of Blind Trials: The CASP CompetitionOver the past 30 years a bewildering variety of techniques have been developed toattack the problem of protein structure prediction in general and fold recognition inparticular. As in any scientific endeavour, it is critical that any new technique isfully tested “experimentally”. It is for this reason that the Critical Assessment ofStructure Prediction or “CASP” meeting was devised (http://predictioncenter.llnl.gov/; Moult et al. 2007). The purpose of the CASP meeting or competition (heldevery two years) is to mimic the real-world situation of being presented with anamino acid sequence for which we do not know the structure. However, there is acritical difference – the organisers of the meeting do know the structure. Theseproteins have had their structures newly solved by experimentalists, but this datahas not yet been released to the scientific community. As a result, the assessors ofthe CASP meeting are in the rather rare situation of knowing the 3-dimensionalstructures of a set of proteins unknown to the predictors.CASP acts as a true blind experimental assessment of the viability of techniquesfor structure prediction in the real world. Therefore, the CASP competition hasbeen my guide in deciding what methodologies to describe in this chapter. This isnot to say that other methodologies may not indeed be powerful predictors, whichfor whatever reason did not perform well at CASP. There are literally hundreds ofdifferent techniques that have been developed over the years, and to avoid burdeningthe reader, I have chosen to use the results of CASP as a filter. For a review ofthe most recent CASP7 meeting see the CASP7 supplement (Moult et al. 2007).2.1.2 Ab Initio Structure Prediction Versus Homology ModellingIf we are to have any hope of structurally characterising any significant fraction ofthe proteins in nature, barring the discovery of some revolutionary experimentaltechnique, then we will require a method to predict structure from sequence computationally.After Anfinsen showed in 1961 that ribonuclease could be refoldedafter denaturation while preserving enzyme activity, we have been beguiled by theidea that all the information required by a protein to adopt its final conformation isencoded in its sequence. As a result ‘pure’ methods using only the sequence itselfas input and the laws of physics (or approximations to them) have been pursued fordecades and are showing some progress. These are covered in Chapter 1 of thisbook. However, in general, these methods are either computationally intractable ordemonstrate poor performance on everything but the smallest proteins (