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

Subject: search.filters.subject.computational biology

Search Results

Now showing 1 - 3 of 3
  • Thumbnail Image
    Item
    ANALYTICAL APPROACHES FOR COMPLEX MULTI-BATCH -OMICS DATASETS AND THEIR APPLICATION TO NEURONAL DEVELOPMENT
    (2023) Alexander, Theresa Ann; Speer, Colenso M; El-Sayed, Najib M; Biology; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
    High-throughput sequencing methods are extremely powerful tools to quantify gene expression in bulk tissue and individual cells. Experimental designs often aim to quantify expression shifts to characterize developmental trajectories, disease states, or cellular drug responses. Experimental and genetic methods are also rapidly evolving to capture specific aspects of gene expression such as in targeting individual cell types, regulatory stages, and spatially resolved cell subcompartments. These studies frequently involve a variety of experimental conditions that require many samples to guarantee sufficient statistical power for subsequent analyses. These studies are frequently processed in multiple batches due to limitations on the number of samples that can be collected, processed, and sequenced at once. To eliminate erroneous results in subsequent analyses, it is necessary to deconvolve non-biological variation (batch effect) from biological signal. Here, we explored variational contributions in multi-batch high throughput sequencing experiments by developing new methods, evaluating heterogeneity-contributors in an axon-TRAP-RiboTag protocol case-study, and highlighting biological results from this protocol. First, we describe iDA, a novel dimensionality reduction method for high-throughput sequencing data. High-dimensional data in complex, multi-batch experiments often result in discrete clustering of samples or cells. Existing unsupervised linear dimensionality reduction methods like PCA often do not resolve discreteness simply with projections of maximum variance. We show that iDA can produce better projections for separating discrete clustering that correlates with known experimental biological and batch factors. Second, we provide a case study of special considerations for a complex, multi-batch high throughput experiment. We investigated the multi-faceted heterogenic contributions of a study using the axon-TRAP-RiboTag translatomic isolation protocol in a neuronal cell type. We show that popular batch-correction methods may reduce signal due to true biological heterogeneity in addition to technical noise. We offer metrics to help identify biological signal-driven batch-differences. Lastly, we employ our understanding of variational contributions in the intrinsically photosensitive retinal ganglion cell (ipRGC) -omics case study to explore the biological transcriptomic and translatomic coordination. Our analysis revealed ipRGCs participate in subcompartment-specific local protein translation. Genetic perturbations of photopigment-driven neuronal activity led to global tissue transcriptomic shifts in both the retina and brain targets, but the ipRGC axonal-specific translatome was unaltered.
  • Thumbnail Image
    Item
    STRATEGIES AND RESOURCES FOR RATIONAL VACCINE DESIGN AND ANTIBODY-ANTIGEN DOCKING AND AFFINITY PREDICTION
    (2022) Guest, Johnathan Daniel; Pierce, Brian G.; Cell Biology & Molecular Genetics; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
    Antibody recognition of antigens is a unique class of protein-protein interactions, and increased knowledge regarding the determinants of these interactions has advanced fields such as computational vaccine design and protein docking. However, the diversity and flexibility of antibodies and antigens can hinder generation of potent vaccine immunogens or prediction of correct antibody-antigen interfaces, slowing progress in the design of vaccines and antibody therapeutics. In this thesis, we present strategies to design vaccine candidates for a difficult viral target and describe expanded resources for benchmarking and training antibody-antigen docking and affinity prediction algorithms.We utilized rational design to develop candidate immunogens for a vaccine against hepatitis C virus (HCV), which represents a global disease burden despite recent advances in antiviral treatments. This design strategy produced a soluble and secreted E1E2 glycoprotein heterodimer with native-like antigenicity and immunogenicity by fusing ectodomains with a leucine zipper scaffold and a furin cleavage site. We developed additional constructs that incorporated synthetic or non-eukaryotic scaffolds or alternative ectodomains that included consensus sequences designed using a large reference database. Finally, we utilized previously published data on HCV antibody neutralization and E1E2 mutagenesis to predict residues that impact antibody neutralization and E1E2 heterodimerization, offering potential insights that can aid vaccine design. To improve our knowledge of and accuracy in modeling antibody-antigen recognition, we assembled a set of antibody-antigen complex structures from the Protein Data Bank (PDB) that expanded Docking Benchmark 5, a widely used benchmark for protein docking. These complexes more than doubled the number of antibody-antigen structures in the benchmark and, based on tests of current algorithms, highlight significant challenges for docking and affinity prediction. Building on this resource, we assembled and curated a dataset of ~400 antibody-antigen affinities and corresponding structures, forming an expanded and updated benchmark to guide ΔG prediction of antibody-antigen interactions. Using this dataset, we retrained combinations of terms from existing scoring functions and potentials, demonstrating that this resource can be used to improve antibody-antigen ΔG prediction. Overall, these findings can advance HCV vaccine design and antibody-antigen docking and affinity prediction, helping to better elucidate the determinants of antibody-antigen interactions and to better display vaccine immunogens for induction of neutralizing antibodies.
  • Thumbnail Image
    Item
    Computational methods for the identification of mutation signatures and intracellular microbes in cancer
    (2021) Robinson, Wells; Leiserson, Mark D.M.; Ruppin, Eytan; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
    Cancer is the second leading cause of death in the United States behind heart disease, killing ~600,000 Americans per year. Technological advances have lowered the cost of sequencing a tumor genome even faster than would have been predicted by Moore’s law. However, specialized computational techniques are required to effectively analyze this genomic data. In this dissertation, we present two such computational approaches to address key challenges in the field of computational cancer biology. Given the importance of reproducibility in biomedical research, we provide publicly available workflows for reproducing the results from our computational approaches. In the first part of this thesis, we present a novel approach for the extraction of mutation signatures from matrices of somatic mutations. One computational challenge for extracting mutation signatures is the relatively small number of mutations in each tumor compared to the relatively large number of distinct signatures, which can be mathematically similar to each other. To help address this computational challenge, we apply ideas from the field of topic modeling to develop the first mutation signature model, the Tumor Covariate Signature Model (TCSM), that can incorporate known tumor covariates. We focus on two mathematically similar signatures associated with distinct covariates to evaluate TCSM and show that by leveraging these covariates, TCSM can more accurately distinguish between mutations attributed to these two signatures than existing NMF-based approaches. The second part focuses on the microbes in the tumor microenvironment. It is not currently known whether microbial reads identified from tumor sequencing datasets result from contamination or represent either extracellular or intracellular microbial residents of the tumor microenvironment. We develop a computational approach named CSI-Microbes (computational identification of Cell type Specific Intracellular Microbes) that mines single-cell RNA sequencing (scRNA-seq) datasets to distinguish cell-type specific intracellular microbes from other microbes. We show that CSI-Microbes can identify previously reported intracellular microbes from both human-designed and cancer scRNA-seq datasets. Finally, we apply CSI-Microbes to a large scRNA-seq lung cancer dataset and identify microbial taxa in tumor cells with a transcriptomic signature of decreased immune activity.