A potential function for protein folding

Abstract

One of the most difficult problems in computational chemistry is the prediction of the three-dimensional structure of a protein molecule given only its amino acid sequence. Although there are several programs for calculating the empirical or quantum mechanical energies, and there are more programs for either minimizing the energy as a function of conformation or for simulating the dynamics of a system of molecules, these approaches generally fail either to locate the known native conformation of small proteins and/or show that the native structure is preferred over alternative conformations. In this paper, we present the latest extension of our ongoing effort to devise an empirical potential function that correctly discriminates between the native and essentially all other conformations for more than one protein. Furthermore, the potential incorporates such a simplified description of the polypeptide chain that there is hope for locating nearly the global minimum as a means of predicting globular protein conformation. The current potential function has been parameterized to agree with the crystal structures of crambin and avian pancreatic polypeptide and [lie parameters thus derived are able to correctly predict the native conformations of apamin and mellitin. The key to this accomplishment is a novel nondifferentiable optimization approach to solving the nonlinear program for determining the parameters.