A MULTISTAGE APPROACH TO PROTEIN-PROTEIN DOCKING

Boston University, Boston MA

Investigators

Linked publications & trials

Abstract

Fourier correlation techniques are very efficient for docking bound (i.e., co-crystallized) protein conformations using measures of surface complementarity as the target function. However, when docking unbound (i.e., independently crystallized) conformations, the method yields an enormous number of false positives (i.e., conformations with good score but large RMSD). The major goal of this proposal is to add post-processing steps to the Fourier docking algorithm in order to remove all false positives. Post-processing will include a rigid-body filter to reduce the number of candidate structures, a flexible filter that eliminates all far-from-native conformations (with RMSDs over 10 A), and a flexible docking algorithm to refine the remaining few. The filtering and refinement steps utilize the changing contributions that electrostatics, desolvation, and molecular mechanics exhibit at the different stages of both protein-protein association and docking. The rigid body analysis is based on the mapping of electrostatic and desolvation interactions between proteins in encounter complexes; i.e., before extensive surface contacts are established. These interactions are much less sensitive to structural perturbations than the measures of surface complementarity, and provide a useful scoring function. However, due to the differences in side chain conformations between bound and unbound states, within the framework of the rigid body analysis, the discrimination is not perfect, and cannot eliminate all false positives. Inclusion of molecular mechanics and accounting for flexibility yield dramatic improvement. In particular, after extensive molecular mechanics minimization of the structures, the sum of van der Waals, electrostatic, and solvation/entropic energy terms, not only discriminates the near-native conformations from decoys in all systems studied, but also provides a relatively good ranking of near-native structures. These preliminary results support the central hypothesis that all false positives can be removed while retaining and improving the good docked conformations. The efficiency of the entire docking algorithm is further increased by pre-processing steps that either attempt to improve the conformations of surface side chains, or introduce a nonuniform Gaussian blurring in order to avoid spurious overlaps. An automatic procedure combining Fourier docking programs with the post- and pre-processing steps will provide a powerful research tool in applications that currently cannot be addressed by computational methods.