Predicting protein folding cores based on complex network and phylogenetic analyses

A crucial event of protein folding is the formation of a folding core, which is the subset of the protein´s structure that is the first to form during folding and the last to break during denaturation. Both experimental and simulation techniques to capture folding core membership have found it challenging. In this paper, the six different computational prediction methods were proposed to predict folding core residues from a protein structure. The protein structure is modeled as an undirected network with the amino acids the vertexes and the contacts between them the edges. We find that the core residues have mainly high degree and low clustering coefficient compared with the surface residues. The closeness values of core residues are similar with the surface residues. The degree distribution of core and surface residues all are Poisson distribution. This suggests that the same small-world organization prevails throughout the protein, despite the heterogeneous density distribution. The six different centrality measurements have been proposed to predict folding core residues from a set of 14 well-characterized proteins. We show that the six measurements (degree, clustering coefficient, closeness centrality, betweenness, conservation score and solvent accessible surface area) all accurately detect the folding core residues. The best methods are based on the solvent accessible surface area and the local centrality measurement closeness. This indicates that the folding core residues have high hydrophobicity and tightness packed.