Institute of Enzymology, Biological Research Center, Hungarian Academy of Sciences, Budapest H-1518, PO Box 7, Hungary
The ultimate goal in protein structure prediction is to elucidate the three dimensional structure of any protein purely from its amino acid sequence alone. While secondary structure prediction methods perform above 70 %, thus providing a reliable topology of secondary structure elements, the prediction of a 3D fold is successful only if a highly homolog, already known structure is present as a template. The common problem in tertiary structure prediction is the luck of information on long-range interactions which occur among residues separated sequentially but situated in space proximity. These long-range interactions responsible for the stabilization of the adequate native structure. Even a limited knowledge of long-range interactions could greatly reduce the number of possible 3D folds for a given secondary structure topology.
In this paper we present an algorithm to identify strongly interacting residue clusters based on the proteins' distance map. We call these residues stabilization centers. The analysis of the amino acid composition of the environment of stabilization centers and the rest of the protein showed significant difference, which can be exploited in the prediction of these strongly interacting residues using only sequence information. We used a computer based simulated neural networks the predict these clusters. The achieved efficiency is 65 %, which is significantly higher than the average efficiency on the randomly generated reference subset, with similar population to the stabilization centers (55 %).
We also discuss several properties of the stabilization centers as their average accessibility, types of preferred interactions, typical secondary structures and the distributions of stabilization residues along the sequences. Using the FSSP database we analyzed the structural and sequential conservation of stabilization centers compared with residues involved only simple long-range interactions.