|advertisement: compare things at compare-stuff.com!|
In the previous sections, the discussion of the relative performance of various prediction methods has been kept to a minimum. Whilst many secondary structure prediction methods have been tested on the same or similar sets of proteins, there is no substitute for predictions of structures not yet determined experimentally (blind predictions). A number of these have taken place over the years through collaboration between experimentalists and predictors[Russell & Barton, 1993, this paper gives a summary of pre-1993 blind predictions]. Recently, however, extensive blind trials of comparative modelling, fold recognition, ab initio prediction and docking have been undertaken at two international meetings under the title of Critical Assessment of Structure Prediction at Asilomar in California, USA. The organisers of the meetings requested sequences of proteins which were under study by X-ray crystallography or NMR spectroscopy. Blind predictions were then gathered until the expiry dates set for each target sequence (for example before structures were made public). The clear benefit of this coordinated approach is the evaluation of prediction results on the same targets using the same criteria. The meetings also provide a focus for predictors and delineate between those who are prepared to see their work tested in public and those who are not. Their main drawback is the relatively small and/or variable number of targets in some categories (docking and fold recognition in particular). The results and evaluation of the first meeting (CASP1, December 1994) are available in a special edition of Proteins: Structure, Function and Genetics[Lattman, 1995]. Summaries of CASP2 (December 1996) have already been published[Dunbrack et al., 1997,Marchler-Bauer & Bryant, 1997], although a forthcoming edition of Proteins: Structure, Function and Genetics will give the definitive analysis. Some conclusions from CASP2 will also be discussed in this work.