Date of Award
Open Access Dissertation
Doctor of Philosophy (PhD)
This thesis presents an amalgamation of high performance computing techniques to get better insight into protein molecular dynamics. Key aspects of protein function and dynamics can be learned from their conformational space. Datasets that represent the complex nuances of a protein molecule are high dimensional. Efficient dimensionality reduction becomes indispensable for the analysis of such exorbitant datasets. Dimensionality reduction forms a formidable portion of this work and its application has been explored for other datasets as well. It begins with the parallelization of a known non-liner feature reduction algorithm called Isomap. The code for the algorithm was re-written in C with portions of it parallelized using OpenMP. Next, a novel data instance reduction method was devised which evaluates the information content offered by each data point, which ultimately helps in truncation of the dataset with much fewer data points to evaluate. Once a framework has been established to reduce the number of variables representing a dataset, the work is extended to explore algebraic topology techniques to extract meaningful information from these datasets. This step is the one that helps in sampling the conformations of interest of a protein molecule. The method employs the notion of hierarchical clustering to identify classes within a molecule, thereafter, algebraic topology is used to analyze these classes. Finally, the work is concluded by presenting an approach to solve the open problem of protein folding. A Monte-Carlo based tree search algorithm is put forth to simulate the pathway that a certain protein conformation undertakes to reach another conformation.
The dissertation, in its entirety, offers solutions to a few problems that hinder the progress of solution for the vast problem of understanding protein dynamics. The motion of a protein molecule is guided by changes in its energy profile. In this course the molecule gradually slips from one energy class to another. Structurally, this switch is transient spanning over milliseconds or less and hence is difficult to be captured solely by the work in wet laboratories.
Joshi, Arpita, "High Performance Computing Techniques to Better Understand Protein Conformational Space" (2019). Graduate Doctoral Dissertations. 500.