Building Computational Tools for Antibody Modeling and Protein–Protein Docking
Marze, Nicholas A
MetadataShow full item record
Protein-protein interactions underlie countless biological functions, the nature of which is determined by the structure of the protein complex. Computational modeling is and important resource for evaluating protein complexes, with tools like RosettaDock offering structural insights in a high-throughput and cost-efficient manner. Antibodies provide an interesting test case for computational protein-protein docking protocols; they are a highly homologous class of protein that naturally bind an enormous range of antigenic proteins. In this dissertation, I describe new computational methods I developed to model both antibodies and protein-protein complexes, as well as evaluations I made of their performance. I begin with my additions to the RosettaAntibody protocol, which were motivated by community-wide shortcomings in antibody homology modeling revealed by the Second Antibody Modeling Assessment (AMA-II). I first built the Light-Heavy Orientational Coordinates (LHOC) framework to unambiguously describe the poorly defined antibody VL-VH orientation; I then developed the multiple-template grafting protocol, which leverages the LHOC framework to correctly model the VL-VH orientation in a majority of antibody targets, tripling the accuracy of the previous RosettaAntibody version. Seeing the guidance the AMA-II provided toward improving RosettaAntibody, I participated in several rounds of the Critical Assessment of PRediction of Interactions (CAPRI) to better understand the extant deficiencies of the RosettaDock protocol. CAPRI revealed a number of weaknesses in the protocol, including an inability to fully sample anisotropic proteins. I corrected this shortcoming in my novel Ellipsoidal Dock method, with which I correctly modeleld two challenging CAPRI targets. More broadly, all protein-protein docking methods fared poorly on CAPRI targets with binding-induced conformational changes and/or large surface areas to search. Addressing these difficult docking problems requires significantly more extensive conformational sampling protocols. So that such protocols remain computationally feasible, I developed Motif Dock Score (MDS) to rapidly evaluate the expanded pools of candidate structures. With no additional runtime, MDS provides three times the near-native enrichment and nine times the near-native discrimination as the low-resolution RosettaDock mode it replaces. In summary, I built computational tools that improve the fidelity of antibody homology modeling and broaden the scope of protein-protein docking. Additionally, my contributions to the RosettaDock protocol set the stage for the next generation of computational docking protocols.