Ranking Beta Sheet Topologies with Applications to Protein Structure Prediction

摘要

One reason why ab initio protein structure predictors do not perform very well is their inability to reliably identify long-range interactions between amino acids. To achieve reliable long-range interactions, all potential pairings of β-strands (β-topologies) of a given protein are enumerated, including the native β-topology. Two very different β-topology scoring methods from the literature are then used to rank all potential β-topologies. This has not previously been attempted for any scoring method. The main result of this paper is a justification that one of the scoring methods, in particular, consistently top-ranks native β-topologies. Since the number of potential β-topologies grows exponentially with the number of β-strands, it is unrealistic to expect that all potential β-topologies can be enumerated for large proteins. The second result of this paper is an enumeration scheme of a subset of β-topologies. It is shown that native-consistent β-topologies often are among the top-ranked β-topologies of this subset. The presence of the native or native-consistent β-topologies in the subset of enumerated potential β-topologies relies heavily on the correct identification of β-strands. The third contribution of this paper is a method to deal with the inaccuracies of secondary structure predictors when enumerating potential β-topologies. The results reported in this paper are highly relevant for ab initio protein structure prediction methods based on decoy generation. They indicate that decoy generation can be heavily constrained using top-ranked β-topologies as they are very likely to contain native or native-consistent β-topologies.