INFLUENCE OF PROTEIN-STRUCTURE DATABASES ON THE PREDICTIVE POWER OF STATISTICAL PAIR POTENTIALS

A long standing goal in protein structure studies is the development o f reliable energy functions that can be used both to verify protein mo dels derived from experimental constraints as well as for theoretical protein folding and inverse folding computer experiments. In that resp ect, knowledge-based statistical pair potentials have attracted consid erable interests recently mainly because they include the essential fe atures of protein structures as well as solvent effects at a low compu ting cost. However, the basis on which statistical potentials are deri ved have been questioned, In this paper, we investigate statistical pa ir potentials derived from protein three-dimensional structures, addre ssing in particular questions related to the form of these potentials, as well as to the content of the database from which they are derived . We have shown that statistical pair potentials depend on the size of the proteins included in the database, and that this dependence can b e reduced by considering only pairs of residue close in space (i.e., w ith a cutoff of 8 Angstrom). We have shown also that statistical poten tials carry a memory of the quality of the database in terms of the am ount and diversity of secondary structure it contains. We find, for ex ample, that potentials derived from a database containing alpha-protei ns will only perform best on alpha-proteins in fold recognition comput er experiments. We believe that this is an overall weakness of these p otentials, which must be kept in mind when constructing a database. (C ) 1998 Wiley-Liss,Inc.