Computer Science Theses and Dissertations
Permanent URI for this collectionhttp://hdl.handle.net/1903/2756
Browse
6 results
Search Results
Item High Performance Computing for DNA Sequence Alignment and Assembly(2010) Schatz, Michael Christopher; Salzberg, Steven L; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Recent advances in DNA sequencing technology have dramatically increased the scale and scope of DNA sequencing. These data are used for a wide variety of important biological analyzes, including genome sequencing, comparative genomics, transcriptome analysis, and personalized medicine but are complicated by the volume and complexity of the data involved. Given the massive size of these datasets, computational biology must draw on the advances of high performance computing. Two fundamental computations in computational biology are read alignment and genome assembly. Read alignment maps short DNA sequences to a reference genome to discover conserved and polymorphic regions of the genome. Genome assembly computes the sequence of a genome from many short DNA sequences. Both computations benefit from recent advances in high performance computing to efficiently process the huge datasets involved, including using highly parallel graphics processing units (GPUs) as high performance desktop processors, and using the MapReduce framework coupled with cloud computing to parallelize computation to large compute grids. This dissertation demonstrates how these technologies can be used to accelerate these computations by orders of magnitude, and have the potential to make otherwise infeasible computations practical.Item Whole-genome sequence analysis for pathogen detection and diagnostics(2010) Phillippy, Adam Michael; Salzberg, Steven L; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)This dissertation focuses on computational methods for improving the accuracy of commonly used nucleic acid tests for pathogen detection and diagnostics. Three specific biomolecular techniques are addressed: polymerase chain reaction, microarray comparative genomic hybridization, and whole-genome sequencing. These methods are potentially the future of diagnostics, but each requires sophisticated computational design or analysis to operate effectively. This dissertation presents novel computational methods that unlock the potential of these diagnostics by efficiently analyzing whole-genome DNA sequences. Improvements in the accuracy and resolution of each of these diagnostic tests promises more effective diagnosis of illness and rapid detection of pathogens in the environment. For designing real-time detection assays, an efficient data structure and search algorithm are presented to identify the most distinguishing sequences of a pathogen that are absent from all other sequenced genomes. Results are presented that show these "signature" sequences can be used to detect pathogens in complex samples and differentiate them from their non-pathogenic, phylogenetic near neighbors. For microarray, novel pan-genomic design and analysis methods are presented for the characterization of unknown microbial isolates. To demonstrate the effectiveness of these methods, pan-genomic arrays are applied to the study of multiple strains of the foodborne pathogen, Listeria monocytogenes, revealing new insights into the diversity and evolution of the species. Finally, multiple methods are presented for the validation of whole-genome sequence assemblies, which are capable of identifying assembly errors in even finished genomes. These validated assemblies provide the ultimate nucleic acid diagnostic, revealing the entire sequence of a genome.Item Novel Methods for Metagenomic Analysis(2010) White, James Robert; Pop, Mihai; Applied Mathematics and Scientific Computation; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)By sampling the genetic content of microbes at the nucleotide level, metagenomics has rapidly established itself as the standard in characterizing the taxonomic diversity and functional capacity of microbial populations throughout nature. The decreasing cost of sequencing technologies and the simultaneous increase of throughput per run has given scientists the ability to deeply sample highly diverse communities on a reasonable budget. The Human Microbiome Project is representative of the flood of sequence data that will arrive in the coming years. Despite these advancements, there remains the significant challenge of analyzing massive metagenomic datasets to make appropriate biological conclusions. This dissertation is a collection of novel methods developed for improved analysis of metagenomic data: (1) We begin with Figaro, a statistical algorithm that quickly and accurately infers and trims vector sequence from large Sanger-based read sets without prior knowledge of the vector used in library construction. (2) Next, we perform a rigorous evaluation of methodologies used to cluster environmental 16S rRNA sequences into species-level operational taxonomic units, and discover that many published studies utilize highly stringent parameters, resulting in overestimation of microbial diversity. (3) To assist in comparative metagenomics studies, we have created Metastats, a robust statistical methodology for comparing large-scale clinical datasets with up to thousands of subjects. Given a collection of annotated metagenomic features (e.g. taxa, COGs, or pathways), Metastats determines which features are differentially abundant between two populations. (4) Finally, we report on a new methodology that employs the generalized Lotka-Volterra model to infer microbe-microbe interactions from longitudinal 16S rRNA data. It is our hope that these methods will enhance standard metagenomic analysis techniques to provide better insight into the human microbiome and microbial communities throughout our world. To assist metagenomics researchers and those developing methods, all software described in this thesis is open-source and available online.Item Anomaly Detection in Time Series: Theoretical and Practical Improvements for Disease Outbreak Detection(2009) Lotze, Thomas Harvey; Shmueli, Galit; Applied Mathematics and Scientific Computation; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)The automatic collection and increasing availability of health data provides a new opportunity for techniques to monitor this information. By monitoring pre-diagnostic data sources, such as over-the-counter cough medicine sales or emergency room chief complaints of cough, there exists the potential to detect disease outbreaks earlier than traditional laboratory disease confirmation results. This research is particularly important for a modern, highly-connected society, where the onset of disease outbreak can be swift and deadly, whether caused by a naturally occurring global pandemic such as swine flu or a targeted act of bioterrorism. In this dissertation, we first describe the problem and current state of research in disease outbreak detection, then provide four main additions to the field. First, we formalize a framework for analyzing health series data and detecting anomalies: using forecasting methods to predict the next day's value, subtracting the forecast to create residuals, and finally using detection algorithms on the residuals. The formalized framework indicates the link between the forecast accuracy of the forecast method and the performance of the detector, and can be used to quantify and analyze the performance of a variety of heuristic methods. Second, we describe improvements for the forecasting of health data series. The application of weather as a predictor, cross-series covariates, and ensemble forecasting each provide improvements to forecasting health data. Third, we describe improvements for detection. This includes the use of multivariate statistics for anomaly detection and additional day-of-week preprocessing to aid detection. Most significantly, we also provide a new method, based on the CuScore, for optimizing detection when the impact of the disease outbreak is known. This method can provide an optimal detector for rapid detection, or for probability of detection within a certain timeframe. Finally, we describe a method for improved comparison of detection methods. We provide tools to evaluate how well a simulated data set captures the characteristics of the authentic series and time-lag heatmaps, a new way of visualizing daily detection rates or displaying the comparison between two methods in a more informative way.Item Highly Scalable Short Read Alignment with the Burrows-Wheeler Transform and Cloud Computing(2009) Langmead, Benjamin Thomas; Salzberg, Steven L; Pop, Mihai; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Improvements in DNA sequencing have both broadened its utility and dramatically increased the size of sequencing datasets. Sequencing instruments are now used regularly as sources of high-resolution evidence for genotyping, methylation profiling, DNA-protein interaction mapping, and characterizing gene expression in the human genome and in other species. With existing methods, the computational cost of aligning short reads from the Illumina instrument to a mammalian genome can be very large: on the order of many CPU months for one human genotyping project. This thesis presents a novel application of the Burrows-Wheeler Transform that enables the alignment of short DNA sequences to mammalian genomes at a rate much faster than existing hashtable-based methods. The thesis also presents an extension of the technique that exploits the scalability of Cloud Computing to perform the equivalent of one human genotyping project in hours.Item A framework for discovering meaningful associations in the annotated life sciences Web(2009) Lee, Woei-jyh; Raschid, Louiqa; Tseng, Chau-Wen; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)During the last decade, life sciences researchers have gained access to the entire human genome, reliable high-throughput biotechnologies, affordable computational resources, and public network access. This has produced vast amounts of data and knowledge captured in the life sciences Web, and has created the need for new tools to analyze this knowledge and make discoveries. Consider a simplified Web of three publicly accessible data resources Entrez Gene, PubMed and OMIM. Data records in each resource are annotated with terms from multiple controlled vocabularies (CVs). The links between data records in two resources form a relationship between the two resources. Thus, a record in Entrez Gene, annotated with GO terms, can have links to multiple records in PubMed that are annotated with MeSH terms. Similarly, OMIM records annotated with terms from SNOMED CT may have links to records in Entrez Gene and PubMed. This forms a rich web of annotated data records. The objective of this research is to develop the Life Science Link (LSLink) methodology and tools to discover meaningful patterns across resources and CVs. In a first step, we execute a protocol to follow links, extract annotations, and generate datasets of termlinks, which consist of data records and CV terms. We then mine the termlinks of the datasets to find potentially meaningful associations between pairs of terms from two CVs. Biologically meaningful associations of pairs of CV terms may yield innovative nuggets of previously unknown knowledge. Moreover, the bridge of associations across CV terms will reflect the practice of how scientists annotate data across linked data repositories. Contributions include a methodology to create background datasets, metrics for mining patterns, applying semantic knowledge for generalization, tools for discovery, and validation with biological use cases. Inspired by research in association rule mining and linkage analysis, we develop two metrics to determine support and confidence scores in the associations of pairs of CV terms. Associations that have a statistically significant high score and are biologically meaningful may lead to new knowledge. To further validate the support and confidence metrics, we develop a secondary test for significance based on the hypergeometric distribution. We also exploit the semantics of the CVs. We aggregate termlinks over siblings of a common parent CV term and use them as additional evidence to boost the support and confidence scores in the associations of the parent CV term. We provide a simple discovery interface where biologists can review associations and their scores. Finally, a cancer informatics use case validates the discovery of associations between human genes and diseases.