Prediction of Residue−Residue Pair Frequencies in Proteins

Knowledge of how frequently different types of residues are found near each other in protein structures has been widely used in threading and in simulating protein folding. In this paper we show that the residue−residue pair frequencies can be reproduced by a simple, physical model. The central component is the nonpolar in−charge out character. This character was captured by obtaining for each type of residue the relative density at a given distance from the protein's center of geometry. These densities were conveniently fitted to exponential or linear functions of the radial distance and used to generate atomic positions. To account for chain connectivity, distances between residue pairs were constrained by independent Gaussian functions, which have increasing means and deviations for increasing sequence separations. Interactions between nonpolar residues and between charged residues were found to extend up to a distance of ∼7.5 Å and the interaction potentials extracted appear to be an intrinsic property. This radial-distance based model, constructed and tested on a set of 243 nonhomologous proteins, has a clear physical basis and may hold important clues for structure prediction.