%0 Journal Article
%T Using neural networks and evolutionary information in decoy discrimination for protein tertiary structure prediction
%A Ching-Wai Tan
%A David T Jones
%J BMC Bioinformatics
%D 2008
%I BioMed Central
%R 10.1186/1471-2105-9-94
%X Results have shown that the best performing neural network is the one that uses input information comprising of PSI-BLAST [1] profiles of residue pairs, pairwise distance and the relative solvent accessibilities of the residues. This neural network is the best among all methods tested in discriminating the native structure from a set of decoys for all decoy datasets tested.This method is demonstrated to be viable, and furthermore evolutionary information is successfully used in the neural networks to improve decoy discrimination.In recent years, a rise in the number of genome sequencing projects around the world has led to an increase in the number of protein sequences with unknown structures. Protein structure prediction aims to bridge the gap between the number of such sequences and the number of sequences with experimentally determined structures. One advantage of computational protein structure prediction is that accurate in silico protein modelling can help guide the more expensive experimental efforts in protein structure determination. Ultimately, the goal is to understand protein function through its 3D structure and sequence and to further increase our biological insights of the behaviour and interactions of these macromolecules in ways that would be beneficial to mankind.Since 1994, the CASP experiments [2] have provided a useful platform for structure prediction groups to apply their methods to a common set of target sequences, thereby providing the means of direct comparison between these methods. If a target sequence has templates in the structure databases, comparative modelling and fold recognition methods are used to select the templates. In the event of a sequence having an unknown fold, in the case of the Template Free category, fragment assembly methods such as FRAGFOLD [3] and Rosetta [4] are used to build plausible models. Typically, large numbers of candidate models, also known as decoys, are built in order to sample as large a 3D conformationa
%U http://www.biomedcentral.com/1471-2105/9/94