View article

L lNTRODUCTlON ln previous work [1-4], we have investigated the ability of simple potential functions, derived from statistics in the Protein Data Bank (PDB [5, 6]}, to generate correct predictions of protein tertiary structure given the native secondary structure as input. Most recently [2], we studied an unbiased sample of 95 proteins in the size range of 30-160 residues, and we were able to locate nativelike low energy structures in a significant number of cases. However. there were also many examples of unsatisfactory performance; furthermore, the utilization of native secondary structure derived from PDB coordinates is an obvious limitation in terms of the utility of the method for protein structure prediction. Thus, a significant improvement in the potential function. along with tests under more realistic conditions, were required before one could consider applying the methodology to problems of practical interest. A principal reason for carrying out the studies described above was to generate a large database of plausibly misfolded structures in the hope of elucidating systematic flaws in the database potential function that we employed, a principal component of which is the pairwise potential of mean force developed by Sippl and co-workers [7]. We have recently uncovered one systematic error in the Sippl formulation of the statistical pair potential, and we remedied this deficiency in a straightforward fashion: The potential function. at least as applied to the problems discussed here, should be dependent upon the size of the protein. a feature that has also been uncovered in other. more theoretical work [8]. To this end, we developed a statistical potential …