Beta-Sheet Structure Prediction Methods

The amino acid sequence rules that correspond to beta-sheet structures in proteins are still not well understood. Current protein structure prediction methods are more accurate for alpha-helical structures than for beta-sheet structures. One reason that beta-strand structure prediction is more difficult is because of its high prevalence of nonlocal interactions between regions of the protein chain that are not necessarily consecutive in the amino acid sequence. These long range interactions make it difficult to approximate structures from amino acid sequence information. However, several methods and approaches have been developed for predicting beta-sheet structures. Typical protein structure prediction methods, including Hidden Markov models, sequence profile searches, and protein threading methods, assign structures to protein sequence using folds of known structures as templates [1]. These methods are limited by the number of known structures which can serve as templates, thus causing strong biases in the predicted results because there are relatively fewer known beta-sheet structures from higher eukaryotes than for prokaryotes. These methods do not work for many folds that have very low sequence homology with each other. For example, beta-helix is characterized by very little regular repeats in its sequence, thus preventing the use of structure templates for structure prediction. Ab initio methods use physical and knowledge-based information to predict protein structures. Rosetta is a common ab initio protein structure prediction algorithm that is based on protein folding in which local sequence segments sample between different possible local structures, and folding occurs when the conformations and relative orientations of the segments satisfy burial of hydrophobic residues and pairing of beta-strands without steric clashes [2]. The distribution of the sampled structures are approximated by the distribution of the conformations adopted by the sequence segment and related sequence segments in the protein structure database. However, recent experimental studies have contributed to the improvement of ab initio structure prediction, which is one of the most reliable method for predicting protein structure in the absence of homologue. Experimental studies showed that protein folding rates are correlated with the relative contact order (CO) of the native structure, which is the average sequence separation of residues that form contacts in the 3D structure divided by the length of the protein [3]. Proteins with more local contacts fold more rapidly than proteins with nonlocal contacts. This correlation