Text

theses

This dissertation describes the development, benchmarking, validation, and application of computational methods, mostly written within the Rosetta suite of macromolecular modeling software, for the design and/or analysis of proteins.

The introductory chapter brings the reader into the historical context of computational protein design, while detailing important concepts referenced in subsequent chapters.

The second chapter is comprised of a benchmarking study of “LooDo”, a computational algorithm for the design of novel nested-domain proteins, which are proteins where one domain is inserted into another. This algorithm, named after its primary sampling method of loop-directed domain placement, was shown to be able to recapitulate native domain orientations for a benchmark set of five nested-domain proteins, as well as recapitulate domain-domain interface sequences and rank native versus nonnative domain combinations highly.

In the next chapter, to improve the therapeutic ratio of the yeast cytosine deaminase (yCD)/5-fluorocytosine (5FC) directed-enzyme prodrug therapy system for targeted chemotherapy, we hypothesized that light-induced structural changes in yCD via bifunctional azobenzene derivative cross-linking would allow for the design of a photoswitchable yCD enzyme. Using generalizable computational design methods and experimental validation, we present one such design that allowed for a roughly 2-fold increase in activity towards cytosine under UV versus blue light stimuli.

The fourth chapter presents a study in which the Rosetta and Amber energy functions are systematically compared by their performance in two structural evaluation tests and afterwards combined to increase the overall performance over both individually. The minimum-sum-of-ranks method employed in this chapter reduces the RMSD of the selected decoy by 1Å in 14 cases for the ff14SBonlySC energy function in Amber and 13 cases for the current Rosetta energy function, REF2015, in a large decoy discrimination benchmark test.

The final chapter investigates bioisosteric alternatives to Axitinib in order to reduce the metabolic vulnerability of the heteroaryl thioether group. Using QM calculations, Rosetta docking protocols, and Amber molecular dynamics simulations, this study computationally evaluates four proposed structures by their predicted behaviors within the VEGFR2 kinase and ABL1 T315I gatekeeper mutant kinase binding pockets.

Together, the work herein represents a collection of developments in the fields of computational biology and protein design.

ETD_9516

Ph.D.

Includes bibliographical references

by Kristin Blacklock

doi:10.7282/t3-kapg-5y95

ETD doctoral

The author owns the copyright to this work.