22691 research outputs found
Sort by
Iterative Decision Making in Structured Prediction
Full text linkThis thesis investigates structured prediction problems in natural language processing through iterative decision-making, where a model (or agent) makes a sequence of decisions to build one or multiple structured outputs. In Part I, we introduce decision-making in various structured prediction tasks, the mathematical frameworks used to understand these problems, and the machine learning techniques for training such models. We also discuss evaluation metrics, presenting a unified view of various metrics used in prior work and a generalized framework to understand and interpret them, which provides a basis for evaluations in the subsequent parts of the thesis.
Part II focuses on the empirical development of models that adopt the decision-making approach in structured prediction. We investigate a range of semantic structured prediction tasks, from information extraction to semantic parsing. In Chapter 4, we explore a decision strategy that mimics humans’ reading process for event extraction, where an agent iteratively reads text and annotation guidelines to extract elements and build an event structure.
Chapter 5 examines structuring the action space for argument linking, expanding the expressiveness of single actions to predict multiple structural elements simultaneously. This approach is further extended and discussed in Chapter 6 to develop an approach that iteratively predicts multiple structures in a sequence of actions. Chapter 7 shifts the focus from developing specialized models to leveraging large language models for in-context semantic parsing tasks while retaining an iterative decision-making perspective
Computational Systems Pharmacology of Antibody-Drug Conjugates
Full text linkDesigned as targeted cancer therapeutics, antibody-drug conjugates (ADCs) comprise a monoclonal antibody base attached to one or more cytotoxic agents via chemical linkers. This strategy enables specific targeting of antigens expressed on the cancer cell surface, resulting in receptor-mediated endocytosis and delivery of toxic payload to the cells of interest while sparing healthy tissues. Despite these promising mechanisms of action, ADCs may still exhibit low efficacy and considerable toxicity, as preclinical efficacy and safety do not always translate to clinical settings. Thus, developing a quantitative understanding of ADC mechanisms and pharmacokinetics/ pharmacodynamics (PK/PD) is important to design safe and effective therapies. Quantitative systems pharmacology (QSP) modeling combines mechanistic knowledge with preclinical and clinical data to generate computational simulations, enabling predictions of efficacy and toxicity.
In this thesis, I detail the development of multiscale, computational systems pharmacology models of antibody-drug conjugates, specifically those carrying pyrrolobenzodiazepine (PBD) payloads. First, I built a mechanistic computational model platform reflecting the cellular mechanisms of PBD ADCs, using ordinary differential equations to track changes in concentration over time in the system. I parameterized the model using in vitro experimental data for PBD ADCs targeting B cell maturation antigen (BCMA), intended to treat multiple myeloma, and PBD ADCs targeting human epidermal growth factor receptor 2 (HER2), intended to treat HER2-positive solid tumors. Using the model, I conducted simulations of cancer cell culture experiments, seeking the factors most important to cell killing and exploring how changes in ADC design and systemic parameters impacted the predicted efficacy and potential toxicity.
Next, I extended the mechanistic model into a compartmental model to represent tumor xenografts in mice by adding mouse PK/PD. By designing and incorporating a novel tracking module into the models, I was able to identify the recent location history of the cytotoxic payload, which facilitated estimates of both the on-target and off-target (bystander) potential for cell killing. Using this multiscale, mechanistic, computational model platform, I can generate insights for optimization of ADC design and determine which factors are most critical to efficacy and toxicity, leading to more informed and rational development of cancer therapies to ultimately improve patient lives
DESIGN AND OPTIMIZATION OF ENHANCED MICROELECTRODES TOWARDS ENGINEERING PRIMARY CELLS
No full textThe advent of recombinant gene expression into mammalian cells has revolutionized biomedical research, advancing our understanding of gene functions and advanced disease mechanisms to foster the development of targeted therapeutic strategies. Primary cell gene transfection is critical to advance our understanding of gene functions and advanced disease mechanisms to foster the development of targeted therapeutic strategies. Streamlined workflows for analyzing primary cells are critically needed for the advancement of personalized medicine given the complexity in processing biofluid samples, which involve intricate pre-purification steps and manual transfers between purification and analysis stages. Thus, the focus of this dissertation is to present a microfluidic chip designed to integrate cell purification and biomolecule delivery to realize the efficiency and practicality advantages of a seamless workflow for primary cell assays. Our approach utilizes a microscale electrode layer in conjunction with vortex cell purification technology for sequential electroporation-mediated transfection of cells trapped in vortex. An automative solution exchange system enables multiplexing a variety of biomolecule cargo delivery for a versatile range of downstream cell assays.
We first demonstrate the feasibility of the workflow for clinical applications through pilot-scale combinatorial drug testing onto drug-resistant cancer cells isolated from blood samples. We then expanded the microelectrode circuitry to match the electroporation capacity for an adapted version of the ultra-high throughput vortex cell purification technology, and validated patient-derived cell purification from metastatic breast cancer liquid biopsy samples and transfection through membrane-impermeable molecule delivery. Moreover, we extended our chip applicability in complex gene expression assays by delivering genetic materials such as deoxyribonucleic acid (DNA) and messenger ribonucleic acid (mRNA) into primary cells. By enhancing diagnostic precision and assisting in therapeutic development, our integrated microfluidic tool holds significant promise for advancing personalized medicine
Insights into the diversity of human gene regulation and functional genetic variation
Full text linkGene regulatory networks and functional genetic variation that confers differences in regulation between individuals are major contributors to the development of human traits and disease. In this dissertation, I describe several new resources to facilitate a deeper understanding of the diversity of human functional genetic variation and of the molecular mechanisms by which such variation drives human complex traits. I first describe the development of a large globally diverse human RNA-sequencing resource, which I used to uncover expression- and splicing-associated genetic variation, and to explore the prevalence and nature of functional variation private to populations historically underrepresented in human genetics research. Next, I describe a database of candidate cis-regulatory element (cCRE) annotations derived from a model of epigenetic signatures in the hematopoietic lineage, and verify the enrichment of these cCREs in heritability of human blood-related traits. Finally, I describe the development of the first truly complete human genome by the Telomere-to-Telomere consortium, and I benchmark the utility of this assembly as a reference genome in characterizing human functional genetic variation, particularly in regions of the genome that were previously unresolved. Together, this work provides deeper understanding of the sources of variation in genome function across diverse human populations, while providing valuable resources to facilitate exploration of the molecular mechanisms underlying human traits
IMPROVING EMPLOYEE RETENTION: CHANGING ORGANIZATIONAL HIRING PROCESSES
Full text linkThe growing rate of voluntary employee turnover is having devastating effects on organizational success. By using relevant organizational models and theories, this paper is an extensive literature review supplemented with secondary data analysis of qualitative textual data. Its purpose is to identify the causes of voluntary employee turnover and evaluate recommended methods for employee retention. Across all literature and data, the main themes that support employee retention are possessing a positive organizational culture, providing comprehensive training and professional growth opportunities, and promoting a healthy work-life balance. This information was reframed and placed into a usable plan focusing on a structural change to organizational hiring processes to improve employee retention. This plan includes screening for applicants who are naturally less inclined towards turnover, assessing which applicants are the best organizational fit based on organizational culture and personal attitude, and providing a comprehensive onboarding process that includes training and outlining career growth opportunities
Methods and applications for large-scale pangenomic analysis
Full text linkRecent technological and algorithmic advances have propelled genome sequencing from a multi-billion dollar, decades long endeavor to common and available research practice. Thanks in large part to long-read, ultra-long-read, and high fidelity sequencing, assembling complex genomic regions from non-model organisms is now possible. While telomere-to-telomere assemblies are not yet common-place and perfect genome assemblies are not yet possible, there have been great strides on both fronts. However, it is increasingly apparent that in order to fully appreciate the rich genomic diversity of a population, and thus accurately map phenotypes to genotypes, a single reference genome assembly is insufficient. Too much genetic variation is lost with a single reference, and thus pangenomes, defined as the entire set of genetic information within a clade, are necessary to resolve complex genotype-to-phenotype relationships. This thesis presents several pieces of work related to the pangenome problem. First, we review and present a tutorial for k-mer based applications in genomics, specifically for efficiently modeling genomes from non-model species, a key first step in genome and pangenome assembly of these species. Next, we detail the construction and analysis of a genus-wide pangenome of Solanum (nightshades), with a focus on utilizing this pangenome for biological insights. We then detail a novel, alignment-free method for efficiently analyzing and visualizing large pangenomes. Finally, we discuss several applications of pangenomes, with a focus on the plant kingdom. Taken together, these chapters underscore the necessity of pangenomes to capture the full spectrum of genetic diversity and provide innovative methods and applications for their assembly and analysis, particularly within the plant kingdom
Contested Legacy: A Re-Examination of the Ideologies, Ethos, and Interventions of the Turkish Military
Full text linkThe military is one of the most consequential institutions in Turkey, shaping its founding, generating much of its political leadership, and repeatedly re-orienting its politics through a series of interventions. Given its historical significance, the military is deservingly one of the most studied institutions in the country. Despite extensive scholarship across diverse academic disciplines, studies of the institution present a remarkably consistent picture of the military as a unitary and monolithic actor, committed to a rigid set of Kemalist ideals. This thesis argues that this dominant scholarly view of the Turkish military is flawed, and by extension, has contributed to the failure of academics and policy makers alike to anticipate and comprehend key events in Turkey, including the failed 2016 coup. Using Turkish military interventions and major wars as markers of, and windows into, periods of political and ideological change, this thesis constructs a new periodization of Turkish political history, beginning with the late Ottoman Empire and culminating in the 2016 coup attempt. Re-examining much of the historical literature, including both primary and secondary sources, it argues that the military has historically possessed much more ideological diversity than generally assumed. Instead of a monolithic and unitary actor committed to strict Kemalist orthodoxy, it argues that the military has served as a theater of competition between robust and divergent political views inherent across the Turkish political spectrum. Moreover, it argues that these divergent ideologies played important roles in driving the actions and shaping the policies of the military, including within each coup
RADical Shifts: A Futurist's Guide to Ecological Transformation and Biodiversity Stewardship
Full text linkRADical Shifts: A Futurist’s Guide to Ecological Transformation and Biodiversity Stewardship provides natural resource managers and conservationists with practical tools to navigate the unprecedented challenges posed by climate change. The guide introduces the Resist—Accept—Direct (RAD) framework, offering a flexible, adaptive approach to managing ecological transformation. By using RAD, conservation leaders can identify strategies to resist, accept, or direct ecological changes in ways that enhance biodiversity stewardship.
The guide emphasizes that climate change can be meaningfully addressed through informed planning and on-the-ground action. It introduces the concepts of RAD menus, RAD portfolios, and RAD decision context, which help managers brainstorm adaptation strategies, track decisions over time and space, and adapt decision-making processes.
Successful RAD actions are often the culmination of many years of collective work, not easily visible from an outsider’s perspective. Radical Shifts helps to demystify these processes. It includes case studies that examine the behind-the-scenes realities of RAD decisions in different regions of the United States.
With RAD menus, managers can explore a full spectrum of adaptation actions. RAD portfolios assist in planning and tracking these decisions, accounting for both spatial and temporal factors. The guide also stresses the importance of collaborative, deliberative engagement to adjust social and institutional contexts in response to changing ecological conditions. It notes that failure to adapt decision-making processes can obstruct progress, as past values, rules, and knowledge may hinder the ability to respond to change.
Whether managing a small-scale project or leading larger efforts, this guidebook equips natural resource managers with adaptable approaches to manage biodiversity and ecosystem services in an era of ecological uncertainty. It empowers conservation leaders to act now while embracing the ongoing journey of learning and adaptation, making it an essential resource for navigating the complexities of climate change and ecological transformation
Healthcare Behaviors in Secondary Findings Recipients: A Qualitative Study
No full textAs more individuals receive broad genetic testing as access and technology improves, more individuals are receiving secondary findings (SFs), or genetic variants of medical value unrelated to the primary reason for testing (Katz et al., 2020). The principle justification for disclosing SFs to patients is the medical actionability of these variants (Green et al., 2013). However, the limited literature that exists suggests a significant portion of SF recipients are not adherent to recommended medical actions associated with their result (Sapp et al., 2021). The determinants of adherence in this population remain unexplored.
Through a secondary analysis of semi-structured interviews from participants in the Genomic Services Research Program (GSRP) at NHGRI, this research project begins to address the gaps noted above by examining individuals’ experiences considering and accessing care after receiving a SF. Analysis focused on SFs associated with Lynch syndrome (LS) and BRCA1/2-hereditary breast and ovarian cancer (HBOC), as these cancer predisposition syndromes have been flagged by the CDC as candidates with the greatest potential to improve public health if included in future genomic population screening efforts. Twenty-six interviews were selected from LS and HBOC SF recipients, maximizing time spent engaging with recommendations and demographic diversity. Barriers and facilitators to adherence were identified and mapped onto existing behavior change and implementation science frameworks to yield potential interventions to increase adherence. Interviews were analyzed thematically until saturation was reached.
SF recipients are a heterogeneous population, receiving results from a mixture of pathways and with variable personal and family histories of cancer or other illness. Overall, personal and family health histories informed SF recipients’ acceptance of health risk as well as their preparedness and rationale for healthcare action, but rarely featured in their experience with genetic counseling for their result. Pathway to receiving the SF and strength of existing care networks informed participants’ experiences and ease seeking care for the secondary finding. Adherence remained vulnerable to challenges and change over time. Greater use of family history in SF counseling as well as enhanced and active referral and follow-up plans are needed to overcome barriers to fulfilling the promise of SFs
12-Lead Electrocardiogram Anomaly Detection Using Hybrid 1D and 2D CNN+Transformer Architectures
Full text linkThis thesis presents two hybrid machine learning models for detecting anomalies in electrocardiograms (EKG): one that combines 1D convolutional neural networks (CNNs) with a transformer and another combining 2D CNNs with a transformer. The multilabel PTB-XL dataset includes over 21,000 cardiovascular conditions grouped into five superclasses: myocardial infarction, conduction disturbances, hypertrophy, ST-T wave changes, and normal EKGs. Due to its imbalance and multilabel nature, we implemented data augmentation techniques such as the Multi-Label Synthetic Minority Over-Sampling Technique (ML-SMOTE) and undersampling to handle the class imbalance. Additionally, the EKGs underwent advanced digital signal processing techniques to clean noise artifacts and convert the time-series signal into time-frequency space for two-dimensional representation. The 1D model achieved a 90.8% AUC, outperforming the 2D model, which achieved an 85.3% AUC