| Much research has focused on the development of free energy potentials for use in protein structure prediction and protein-protein docking. The composition of these potentials varies significantly within the protein community and no single function exists which can robustly select near-native protein structures and complexes from sets of incorrectly folded and docked conformations.; In this dissertation we employ sets of protein decoys, i.e., collections of correctly and incorrectly folded proteins, for developing and evaluating discriminatory functions. In particular, we apply potentials that combine molecular mechanics with empirical solvation and entropic terms to several sets of orthogonal decoys, generated by either discrete-state or ab initio methods. We quantify the discriminatory abilities of the individual free energy terms, and their combinations, specifically focusing on the roles of the internal, solvation, and van der Waals energies. In addition, we show the robust discriminatory ability of a potential comprised of solvation free energy and neutral electrostatics and compare it with that of a hierarchical clustering algorithm based upon a pairwise RMSD criterion. From these studies we observe that our results directly depend on the method used to generate the decoys, internal energy is not a robust discriminator, and that including the solvation free energy is required for successful discrimination.; We develop a protocol consisting of free energy filtering, minimization, and re-scoring for discriminating among putative protein complexes sampled by rigid-body docking algorithms. We apply this protocol to three distinct sets of protein complex decoys, observing discrimination of near-native complexes in all cases. Augmented by a pairwise RMSD clustering algorithm, this protocol is applied to ten sets of protein complexes derived by an FFT-based docking algorithm. In all cases we observe satisfactory discrimination of near-native complexes when re-scoring the filtered and minimized complexes with a combination of solvation and van der Waals energies. In addition, we observe complementarity between two distinct statistical functions, one derived from the interior of proteins and the other from protein-protein interfaces. We discuss the results of applying our protocol to the seven targets released as part of the first Critical Assessment of Prediction of Interactions (CAPRI) protein docking competition. |
