Status of empirical methods for the prediction of protein backbone topography

Abstract

An empirical prediction algorithm which uses information on the short-range (intraresidue) and medium-range (up to 4 neighbors on either side) interactions in 20 proteins to assign every residue in a protein to 1 of 5 conformational states is described. The conformational states are defined in terms of the backbone dihedral angles of the residue so that the prediction algorithm can be used to generate starting conformations for subsequent energy-minimization procedures, which would be necessary to predict the 3-dimensional structure of a protein. An estimate is made of the statistical error in the determination of the parameters describing the effects of short-range and medium-range interactions in proteins. This statistical error plays a large role in limiting the accuracy of all prediction methods which rely on data from proteins of known structure. Using the method described, 56% of the residues in 20 proteins were assigned correctly to 1 of 5 conformational states. It seems unlikely that any prediction method can significantly improve on this accuracy for assigning residues to specific backbone conformations unless the size of the data base is increased greatly.