Show simple item record

Development of Novel Methods for Computational Protein Design and Protein-Ligand Docking

dc.creatorDeLuca, Samuel Louis
dc.date.accessioned2020-08-22T17:16:43Z
dc.date.available2015-07-21
dc.date.issued2015-07-21
dc.identifier.urihttps://etd.library.vanderbilt.edu/etd-07022015-151149
dc.identifier.urihttp://hdl.handle.net/1803/12795
dc.description.abstractThe ability to make rapid predictions of macro-molecular structures will enable researchers to carry out effective protein design, virtual High Throughput Screening (vHTS) and rational drug design studies. Here, we describe the development of several new methods in the service of these aims. Specifically, a novel energy term for protein design was implemented and optimized as part of the Rosetta molecular modeling suite. This energy term is based on a statistical analysis of experimental protein structural data, and aims to produce more native-like protein designs. Additionally, a new sampling algorithm and scoring function for the RosettaLigand protein-ligand docking tool was implemented. The new sampling algorithm results in a 30-fold increase in docking speed, and a 15% improvement in success rate across a wide range of protein systems. The new scoring function consists of a set of cartesian grids representing shape complementarity, hydrogen bonding, and ligand microenvironment environment properties. RosettaLigand docking predictions and scores were then combined with chemical descriptor information and used as the input to train a set of neural networks predicting ligand binding affinity. While the resulting networks did exhibit the ability to make predictions more effectively than RosettaLigand alone in a cross-validation study, the models were insufficiently general to make effective predictions when applied to a challenging, independent benchmarking set. In addition to these new methods, appendix chapters detail the results of a ligand docking study into potential drug candidates targeting both RPA70 and MgluR5, an analysis of the limitations of the Rosetta atom typing system with respect to ligand parameterization, descriptions of a number of new software tools for the analysis of large protein datasets, and documentation of the protocols necessary to recapitulate the scientific studies reported here. In total, this work represents a substantial improvement in the overall performance of RosettaLigand, and provides numerous avenues for further research and development.
dc.format.mimetypeapplication/pdf
dc.subjectRosettaScripts
dc.subjectRosettaDesign
dc.subjectprotein design
dc.subjectmachine learning
dc.subjectDocking
dc.subjectProtein-Ligand Docking
dc.subjectRosetta
dc.titleDevelopment of Novel Methods for Computational Protein Design and Protein-Ligand Docking
dc.typedissertation
dc.contributor.committeeMemberStephen Fesik
dc.contributor.committeeMemberHeidi Hamm
dc.contributor.committeeMemberBrian Bachmann
dc.type.materialtext
thesis.degree.namePHD
thesis.degree.leveldissertation
thesis.degree.disciplineChemical and Physical Biology
thesis.degree.grantorVanderbilt University
local.embargo.terms2015-07-21
local.embargo.lift2015-07-21
dc.contributor.committeeChairTerry Lybrand


Files in this item

Icon

This item appears in the following Collection(s)

Show simple item record