Computational Modeling and Design of Protein–Protein Interactions
Johns Hopkins University
Protein–protein interactions dictate biological functions, including ones essential to living organisms such as immune response or transcriptional regulation. To fundamentally understand these biological processes, we must understand the underlying interactions at the atomic scale. However interactions are overly abundant and traditional structure determination methods cannot manage a comprehensive study. Alternatively, computational methods can provide structural models with high-throughput overcoming the challenge provided by the sheer breadth of interactions, albeit at the cost of accuracy. Thus, it is necessary to improve modeling techniques if these approaches will be used to rigorously study protein–protein interactions. In this dissertation, I describe my advances to protein–protein interaction modeling (docking) methods in Rosetta. My advances are based on challenges encountered in a blind docking competition, including: modeling camelid antibodies, modeling flexible protein regions, and modeling solvated interfaces. First, I detail improvements to RosettaAntibody and Rosetta SnugDock, including making the underlying code more robust and easy to use, enabling new loop modeling methods, developing an automatically updating database, and implementing scientific benchmarks. These improvements permitted me to conduct the largest-to-date study of antibody CDR-H3 loop flexibility, which showed that traditional, small-scale studies missed emergent properties. Then, I pivot from antibodies to focus on the modeling of disordered protein regions. I contributed advances to the FloppyTail protocol, including enabling the modeling of multiple disordered regions within a single protein and pioneering an ensemble-based analysis of resultant models. I modeled Hfq proteins across six species of bacteria and demonstrated experimentally-validated prediction of interactions between disordered and ordered protein regions. My simulations provided a hypothetical mechanism for Hfq function. Finally, I designed crystallographic protein–protein interactions, with the goal of improving protein crystal resolution. To approach this exceptional challenge, I first demonstrated that, under homogenous conditions, Rosetta scores can correlate with crystal resolution. Next, I computationally designed and experimentally characterized sixteen variants of a model protein. Only five crystallized, with one providing an improvement in resolution, showing that improvement through computational design is challenging, but possible. In sum, my work advanced our understanding and our ability to model and design several challenging protein–protein interactions.
protein–protein interfaces, computational protein structure prediction and design