Comparison of two optimization methods to derive energy parameters for protein folding: Perceptron and Z score

Two methods were proposed recently to derive energy parameters from known n ative protein conformations and corresponding sets of decoys. One is based on finding, by means of a perceptron learning scheme, energy parameters suc h that the native conformations have lower energies than the decoys. The se cond method maximizes the difference between the native energy and the aver age energy of the decoys, measured in terms of the width of the decoys' ene rgy distribution (Z-score). Whereas the perceptron method is sensitive main ly to "outlier" (i.e., extremal) decoys, the Z-score optimization is govern ed by the high density regions in decoy-space. We compare the two methods b y deriving contact energies for two very different sets of decoys: the firs t obtained for model lattice proteins and the second by threading. We find that the potentials derived by the two methods are of similar quality and f airly closely related. This finding indicates that standard, naturally occu rring sets of decoys are distributed in a way that yields robust energy par ameters (that are quite insensitive to the particular method used to derive them). The main practical implication of this finding is that it is not ne cessary to fine-tune the potential search method to the particular set of d ecoys used. Proteins 2000;41:192-201, (C) 2000 Wiley-Liss, Inc.