Degree

Doctor of Philosophy (PhD)

Department

Biological Sciences

Document Type

Dissertation

Abstract

Proteins commonly convey their functions in coordination with other proteins, small molecules and/or other biological assemblies such as sugars, lipids, DNA or RNA. Understanding the nature of these interactions is therefore central to improving our knowledge of biological systems. This body of work is a consolidation of three different computational approaches to study bio-molecular interactions: large-scale protein-ligand modeling, fragment-based cheminformatics, and computational analysis of interactions in single proteins.

The first issue that is addressed in this study is the scarcity of atomic crystal structures of protein-drug complexes. In general, a drug molecule’s affinity for multiple protein targets may causes unsolicited side effects and therefore is considered a burden; nonetheless, drug promiscuity can be leveraged to repurpose known drugs to treat multiple diseases. We offer large-scale modeling of drug-protein complexes based on data curated in DrugBank, Binding Database and Protein Data Bank. eModelBDB, a library of 200k high-quality drug-protein models as well as a database of protein models bound to FDA-approved drugs explored for rare disease drug repurposing are presented.

Two fragment-based drug design software tools are built to generate targeted virtual screening libraries. These tools are based on minimalistic combinatorial chemistry implemented in a robust graph-based algorithm in C++ and python. eMolFrag deconstructs chemical compounds into small fragments, while eSynth can generate libraries of new compounds by conjoining those fragments. However, eSynth does not require any complicated chemistry rules and is independent of the structures of the protein targets. Computational performance and accuracy of these methods are discussed.

To analyze specific protein interactions, modeling of Herpes Simplex Virus type-1 (HSV-1) proteins, including a difficult case of a transmembrane glycoprotein, are tackled. These atomic models were used to complement and to guide experimental efforts answering important questions about virus protein functions as well as identifying new druggable targets. Identifying a domain in glycoprotein-K in HSV-1 and it’s role in neuronal entry, recognizing a disulfide bond formation that is essential for viral reproduction in vitro, characterizing a protein-protein interaction site in UL37, and discovery of a novel allosteric anti-HSV druggable site in the DNA-packaging motor are examined computationally and validated experimentally.

Date

4-6-2018

Committee Chair

Brylinski, Michal

Available for download on Monday, April 01, 2019

Share

COinS