| The most complete structural characterization of a protein-protein interaction is given by the three dimensional coordinates of two proteins in complex. Experimentally, X-ray crystallography has been able to offer insight into such interactions, but remains a costly and inefficient method for elucidating these atomic details. Hence, in order to better understand molecular interactions, the computational prediction of protein-protein complexes has been of scientific interest for nearly 30 years. With recent advances in both methodology and computational power, such docking methods have become relevant to solving this problem.; The first goal of this research was to develop a rapid, fully-automated web server for protein-protein docking. The protocol run by the server consists of the following steps: a complete sampling of the six dimensional rotational and translational space between two proteins using the Fast Fourier Transform (FFT)-based program DOT, a filtering of the putative complexes using distance-dependent electrostatics and a knowledge-based scoring function, and a clustering algorithm using pairwise binding site Root Mean Squared Deviation (RMSD) as the distance metric for ranking final docked complexes. The server has been validated in the Critical Assessment of PRediction of Interactions (CAPRI) blind assessment of docking protocols, generating near-native models for several targets.; To complement the server, an algorithm for assembling oligomers with N-fold symmetry, where N ranges from two to seven, has been developed. The protocol calls for the pairwise docking between two identical monomers followed by successive geometric transformations to the docked poses in order to find conformations that satisfy symmetry constraints. The method has been validated on 109 oligomeric complexes, predicting the native conformation of the biologically active oligomer in all but four cases.; The final goal of this research was the generation of scoring functions for protein-protein complexes using a novel reference state derived from contacts found within false-positive docking decoys. This is in contrast to the traditional mole fraction reference states used by other commonly used scoring functions. Using only shape complementarity in the docking stage, this new potential leads to significant increases in the number of high-scoring near-native structures generated. |
