Theses and Dissertations from UMD
Permanent URI for this communityhttp://hdl.handle.net/1903/2
New submissions to the thesis/dissertation collections are added automatically as they are received from the Graduate School. Currently, the Graduate School deposits all theses and dissertations from a given semester after the official graduation date. This means that there may be up to a 4 month delay in the appearance of a give thesis/dissertation in DRUM
More information is available at Theses and Dissertations at University of Maryland Libraries.
Browse
2 results
Search Results
Item Using Many-Core Computing to Speed Up De Novo Transcriptome Assembly(2016) O'Brien, Sean; Vishkin, Uzi; Electrical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)The central dogma of molecular biology implies that DNA holds the blueprint which determines an organism's structure and functioning. However, this blueprint can be read in different ways to accommodate various needs, depending on a cell's location in the body, its environment, or other external factors. This is accomplished by first transcribing DNA into messenger RNA (mRNA), and then translating mRNA into proteins. The cell regulates how much each gene is transcribed into mRNA, and even which parts of each gene is transcribed. A single gene may be transcribed in different ways by splicing out different parts of the sequence. Thus, one gene may be transcribed into many different mRNA sequences, and eventually into different proteins. The set of mRNA sequences found in a cell is known as its transcriptome, and it differs between tissues and with time. The transcriptome gives a biologist a snapshot of the cell's state, and can help them track the progression of disease, etc. Some modern methods of transcriptome sequencing give only short reads of the mRNA, up to 100 nucleotides. In order to reconstruct the mRNA sequences, one must use an assembly algorithm to stitch these short reads back into full length transcripts. De novo transcriptome assemblers are an important family of transcriptome assemblers. Such assemblers reconstruct the transcriptome without using a reference genome to align to and are, therefore, computationally intensive. We present here a de novo transcriptome assembler designed for a parallel computer architecture, the XMT architecture. With this assembler we produce speedups over existing de novo transcriptome assemblers without sacrificing performance on traditional quality metrics.Item Highly Parallel Geometric Characterization and Visualization of Volumetric Data Sets(2012) Juba, Derek Christopher; Varshney, Amitabh; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Volumetric 3D data sets are being generated in many different application areas. Some examples are CAT scans and MRI data, 3D models of protein molecules represented by implicit surfaces, multi-dimensional numeric simulations of plasma turbulence, and stacks of confocal microscopy images of cells. The size of these data sets has been increasing, requiring the speed of analysis and visualization techniques to also increase to keep up. Recent advances in processor technology have stopped increasing clock speed and instead begun increasing parallelism, resulting in multi-core CPUS and many-core GPUs. To take advantage of these new parallel architectures, algorithms must be explicitly written to exploit parallelism. In this thesis we describe several algorithms and techniques for volumetric data set analysis and visualization that are amenable to these modern parallel architectures. We first discuss modeling volumetric data with Gaussian Radial Basis Functions (RBFs). RBF representation of a data set has several advantages, including lossy compression, analytic differentiability, and analytic application of Gaussian blur. We also describe a parallel volume rendering algorithm that can create images of the data directly from the RBF representation. Next we discuss a parallel, stochastic algorithm for measuring the surface area of volumetric representations of molecules. The algorithm is suitable for implementation on a GPU and is also progressive, allowing it to return a rough answer almost immediately and refine the answer over time to the desired level of accuracy. After this we discuss the concept of Confluent Visualization, which allows the visualization of the interaction between a pair of volumetric data sets. The interaction is visualized through volume rendering, which is well suited to implementation on parallel architectures. Finally we discuss a parallel, stochastic algorithm for classifying stem cells as having been grown on a surface that induces differentiation or on a surface that does not induce differentiation. The algorithm takes as input 3D volumetric models of the cells generated from confocal microscopy. This algorithm builds on our algorithm for surface area measurement and, like that algorithm, this algorithm is also suitable for implementation on a GPU and is progressive.