Environmental and Occupational Health Sciences Institute
Rutgers University Community RepositoryNot a member yet
58345 research outputs found
Sort by
Implementation of a foot screening tool to identify homeless residents at risk for foot disease
No full textPurpose of the Project: The purpose of this DNP project was to assess if the training of mental health nurses and implementation of a foot care assessment tool for the screening and treatment of foot disease would increase the identification of residents at risk for foot disease and foot related health complications when compared to the current admission screening.
Methodology: The plan for this DNP project used a quality improvement design. To determine if implementing a standardized foot assessment tool would increase the identification of residents at risk for foot disease or foot disease complications, psychiatric RNs were trained to use the screening tool. The RNs then implemented the use of the tool on all new admissions over 10 weeks. A chart audit of all admission assessments for the weeks prior to implementation was conducted for comparative statistical analysis.
Results: Results of the statistical analysis show improvement in screening and identification for patients with foot concerns. The screening rate improved significantly from 0% at pre-intervention to 92.3% at post-intervention, as indicated by p-values of less than .05. Identification of foot concerns was noted for over half of the participants (53.8%). The findings also demonstrated that over half of the patients received appropriate recommendations based on their foot health screening score.
Implications for Practice: To address the needs of this population, with complex healthcare needs, it is essential that psychiatric nurses identify, treat, and monitor health conditions that lead to increased mortality. The implementation of a nurse-driven foot care assessment model will identify homeless individuals at risk and prevent or treat their foot care problems.D.N.P.Includes bibliographical reference
Shock attenuation using skin-based resonators
No full textExplosive bolt separation of stages is common practice in launch vehicles that are used to reach orbit. The impulse of the force generated by the separation causes high energy mechanical waves to travel through the structure, potentially reaching sensitive electronics on the payload that can be damaged as a result. Resonators were investigated to attenuate the vibration. Cantilever resonators were explored, although were found to have little effect when the structure included a thin aerodynamic skin. Here, it is shown that skin-mounted resonators not attached to the lattice had a significant effect on attenuating the shock. Modal analysis of the resonators was conducted to determine the frequencies at which modes occur. This was compared to simulated spectral data to understand which modes play a key role in damping the shock. Energy from the shock was successfully trapped within the resonators, with evidence of modal energy capture in the ring mode frequency band.M.S.Includes bibliographical reference
Engineering magnetic topological insulators at the atomic level
No full textMagnetic topological insulators (MTIs) represent a class of materials that merge magnetic order with non-trivial band topology, offering possibilities to realize a plethora of magneto-topological states. From an experimental standpoint, design, growth and characterization of MTIs provide the foundation for observing these novel states and manipulating them to explore fundamental physics and accelerate their progress towards real-world applications. The main focus of this dissertation is artificially engineering the intrinsic magnetic topological insulator family, [MnTe][Bi2Te3]n, using molecular beam epitaxy (MBE). We begin by finding optimal growth templates and conditions for its building blocks, nickeline phase MnTe and Bi2Te3, which also leads us to the discovery of a zinc-blende/wurtzite mixed phase MnTe. Then, we employ the atomic layer-by-layer MBE technique to grow [MnTe][Bi2Te3]n with n as large as 15, well beyond the thermodynamic limits of bulk crystal growth. We discover that the “single-layer magnet (SLM)” phase, solely determined by intralayer ferromagnetic coupling, emerges for n > ~4 and remains little affected up to the extreme limit of n = 15. Nonetheless, we find that a small, yet non-zero, interlayer ferromagnetic coupling is necessary to stabilize the SLM phase. In the end, we describe some of our attempts towards quantum anomalous Hall effect through defect engineering of Cr doped Sb2Te3.Ph.D.Includes bibliographical reference
Unspeakable but not unspoken: the literary language of plague trauma in late Middle English texts
No full textThis dissertation argues that argue that literary elements in some canonical Middle English literature can be read as representative of the traumatic experience of living through the first waves of The Black Death in the second half of the fourteenth century in England. Even where they omit direct reference to pestilence, these texts nevertheless bear witness to the traumatic afterlife of the plague through the use of medical discourse, metaphoric and symbolic language, and thematic fixation upon sudden death and lingering grief. The project takes as its starting premise that the plague had a psychological impact on both the individuals that witnessed it, as well as upon English society as a whole; this impact bubbles to the textual surface in the ways these texts represent grief, depictions of the sick, and recollections of the deceased. Informed by trauma theory, the project takes an interdisciplinary approach, serving primarily as a literary study that borrows from recent work in psychology, cognitive linguistics, and history; it draws upon psychoanalytic thought even as it relies upon close reading of both literary and medical texts. Specific literary structures (repetition, circularity, omission, substitution); themes (death, mourning, corruption, woundedness, apocalypse); and language (metaphor, allegory, and the “discourse of phisik”) imitate the fundamental elements of trauma narratives, which are themselves characterized by symbolic or otherwise “literary” language—that is, language that is replete with omissions and gaps, couched in non-linear, atemporal, fragmented, repetitive structures. By examining what these circumlocutions aim to omit or repress and closely reading the symbols, images, and metaphors employed, traces of the plague thus become legible, in turn revealing the traumatic impact of the Black Death upon those who witnessed and survived it.Ph.D.Includes bibliographical reference
Exploring the application of process mining: process model discovery, data generation, and event prediction
No full textProcess mining is a prominent research field that has emerged in recent years. It offers substantial benefits in decision-making, efficiency enhancement, and error reduction across various real-world domains such as healthcare, business, and education. Process mining field mainly analyzes process data, typically stored in event logs, which capture case IDs, activities, and other chronological context. Processes exhibit variability based on different use cases or contexts. Researchers apply process mining techniques to extract non-trivial knowledge from the process data. However, a prevalent issue in this field is in the limited publicly available datasets that lack the complexity and diversity of real-world scenarios, which limited the development of generalized process mining methods and their practical applicability to complex problems. This dissertation aims to address this gap by exploring process mining applications across datasets of varying complexity. We proposed novel methods to enhance the representation of complex process data, mitigate challenges associated with limited data, and support human decision-making. Specifically, the dissertation presents three key contributions: (1) a method for discovering interpretable process model; (2) a deep generative model for augmenting process dataset; and (3) deep learning-based models for event prediction based on process data. The effectiveness of these methods is demonstrated through experiments conducted in real-world medical settings, such as trauma resuscitation, where they substantially improved medical providers' decision-making processes. Although the focus of this dissertation is on medical processes, the approaches developed are applicable to similar domains that using process datasets. First, we introduced a process model discovery method for complex medical processes. To develop a medical process model that improves patient outcomes, understanding the actual work (i.e., “work-as-done”) rather than theorized work (i.e., “work-as-imagined”) is critical. When applied to complex process data, traditional process model discovery methods often omits critical steps or produces cluttered and unreadable models. To address these limitations, we introduced a Trace Alignment-based Process Discovery method called TAD Miner to build interpretable process models. TAD Miner creates simple linear process models using a threshold metric that optimizes the consensus sequence to represent the backbone process, and identifies both concurrent activities and uncommon-but-critical activities to represent the side branches. TAD Miner also identifies the locations of repeated activities, which is an essential feature for representing medical treatment steps. We used the TAD process models to identify (1) the errors and (2) the best locations for the tentative steps in knowledge-driven expert models. The knowledge-driven models were revised based on the modifications suggested by the discovered models. The improved modeling using TAD Miner may enhance understanding of complex medical processes. Second, we introduced a data augmentation model for synthetic process data generation. Because process data with confidential information cannot be shared, research is limited using process data and analytics in the process mining domain. We introduced a generative adversarial network, called ProcessGAN, to address the limitation of shareable process data. Our model generates process data that consists of activity sequences and corresponding timestamps. ProcessGAN consists of a transformer-based network as the generator, and a time-aware self-attention network as the discriminator. It can generate privacy-preserving process data from random noise. ProcessGAN considers the duration of the process and time intervals between activities to generate realistic activity sequences with timestamps. To evaluate the synthetic data, we proposed statistical metrics and trained a supervised model to score the synthetic processes. We also applied our TAD Miner to discover process models (i.e., workflow diagrams) for synthetic medical processes and had domain experts evaluate the clinical applicability of the synthetic diagrams. ProcessGAN outperformed the existing generative models in generating complex processes with valid parallel pathways, and better represented the long-range dependencies between activities. The timestamps generated by the ProcessGAN model showed similar distributions and trends with the authentic timestamps. ProcessGAN can generate shareable synthetic process data indistinguishable from authentic data. Last, we introduced deep learning-based event prediction models specifically to assist the decision-making in the medical process. We introduced (1) a treatment activity recommendation model, and (2) a goal recognition model. Both models have the potential to improve workflow, reduce team errors, and improve patient outcomes. The treatment activity recommendation model aggregates patient context (e.g., injury features and physiological values) and process context (e.g., already performed activities) as input, and continuously predicts the activities to be performed. Our model is built on a tower-structured model of multi-layer perceptrons. It achieved a 0.89 F1-score on the sepsis activity predictions and 0.65 F1-score on the trauma resuscitation activity predictions, showing the feasibility of predicting activities using context and process data. The goal recognition model aims to track whether certain goals are being pursued by the medical providers throughout a medical process. Our experiments focused on recognizing the providers' goal of (1) airway stabilization and (2) circulatory support, which are two critical goals in trauma resuscitation. We designed a dual-gated recurrent unit (GRU) architecture to learn the contributions from timestamp and activity types for goal recognition. Our model achieved an average AUC score of 0.84 for airway stabilization and 0.83 for circulatory support. The results are explainable by identifying the significant time windows and the most contributory activities in deciding whether the goal is being pursuit. Our model outperformed existing medical event predictive models in both accuracy and interpretability. Integrating our model into a decision support system would automate the tracking of providers’ actions, optimizing workflow to ensure timely delivery of patient care.Ph.D.Includes bibliographical reference
Adaptive encoding of novel acoustic signals in the songbird forebrain
No full textOur sensory systems need to adaptively encode novel and variable sensory information from the external environment in the service of behavior. In the auditory domain, neural representations of acoustic signals are shaped by prior experience and reflect expectations. Stimulus discrimination and generalization are two major and complementary challenges faced by this encoding process. This dissertation investigates the neural mechanisms that balance discrimination between specific sensory signals and generalization across variability and how the resulting representations reflect the statistical structure and acoustic context of passively heard sounds. The auditory system of zebra finches shows substantial capacity to process complex communication signals. In particular, the secondary auditory area (caudomedial nidopallium, NCM) of zebra finches exhibits a process of dynamic stimulus-specific adaptation to repeated sounds in the absence of reinforcement. My work will show that this adaptation can be considered as a re-coding process that improves neural discrimination of acoustic signals. Experiment 1 investigates how NCM neurons encode phonetic categories in natural human speech. NCM neurons can discriminate phonetic features and form generalized representations for phonetic categories. Such representations improve with passive exposure and transfer to novel human speakers. Experiment 2 expands these findings using simpler synthetic acoustic stimuli. Passive exposure to sets of stimuli with parametrically varying acoustic features leads to both decreased responses (adaptation) and increased neural discrimination in NCM, and such familiarity effects generalize to similar acoustic features and adjacent feature ranges. Furthermore, these familiarity effects can be long-lasting. Experiment 3 investigates how neural coding of acoustic signals in NCM is modulated by pharmacological degradation of the noradrenergic system. Taken together, these experiments provide an understanding of how a sensory system can dynamically represent acoustic signals and how discrimination and generalization are balanced through the adaptive coding processes to efficiently provide information about the external world.Ph.D.Includes bibliographical reference
Dysregulated neuroimmune interactions in human stem cell derived neural models
No full textDementia is a debilitating neurodegenerative condition with the highest economic burden of all adult neurological disorders. While the cellular mechanisms underlying dementia-related pathology remain enigmatic, inflammation likely plays a major role in causing and exacerbating dementia-related neurodegeneration and cognitive decline. Chronic neuroinflammation is a hallmark of both alcohol use disorder (AUD)-related dementia and Human Immunodeficiency Virus-1 (HIV-1) related dementia. In this dissertation, I first addressed the need for human in vitro neuroimmune models by generating and combining human stem cell derived microglia, neurons, and astrocytes in two model systems. Second, I then interrogated the consequences of HIV-1 infection on human stem cell derived microglia and cerebral organoids. I found that HIV-1 productively infects iPSC-derived microglia and triggers inflammatory activation, resulting in sustained type I interferon signaling that persists in both microglia monoculture and in microglia-containing cerebral organoids. These data are consistent with recent studies suggesting a role for chronic interferon signaling in cognitive deficits and neurodegeneration. Third, I investigated the neuroinflammatory consequences of ethanol exposure on a microglia, neuron, and astrocyte tri-culture model. We found ethanol exposure leads to widespread inflammatory gene expression, activation of the NLRP3 inflammasome, microglia activation, and potent upregulation and alternative splicing of TREM2. These data suggest a complex role for ethanol in the dysregulation of neuroimmune interactions among microglia, astrocytes, and neurons. Our findings relating to TREM2, elevation of complement cascade proteins, and microglial activation point to possible alterations in synaptic connectivity and transmission. Overall, these results expand our understanding of underlying inflammatory processes incited by either HIV-1 or ethanol, both of which are known contributors to dementia, and offer new human neuroimmune models for therapeutic discovery.Ph.D.Includes bibliographical reference
Innovative statistical methods for biomedical and clinical applications
No full textThis thesis introduced novel statistical models and methods developed to address difficulties in analyzing data from cutting-edge biomedical and clinical applications. It addresses significant challenges in big data visualization, regression analysis between variables with errors, precision decision-making, and accurate quantitative estimation, with a focus on three key biomedical and clinical applications. Chapter 1 proposed a novel approach for visualizing and analyzing extremely large datasets using Projection Pursuit, Grand and Guided Tours, and Data Nuggets methods. An important application is analyzing vast datasets from cell flow cytometry experiments, to help investigations into surface protein functions of cells. The proposed methodology, along with a new Projection Pursuit index that can be easily computable for big data, enables the discovery of hidden structures such as clusters, outliers, and nonlinear patterns within extensive datasets. An R package, emph{PPbigdata}, was developed and published on CRAN to provide a powerful tool for performing 1-dim/2-dim projection pursuit for big data based on data nuggets. Chapters 2 and 3 focused on developing innovative regression frameworks and decision-making methods with clinical applications to personalized treatment recommendations. Chapter 2 proposed a two-stage Deming regression framework to accurately analyze associations between variables observed with errors of known or estimated variances. This framework is particularly useful in balancing clinical risks, but its versatility also makes it applicable to handling privacy-protected data containing multiple sources of uncertainty, such as in differential privacy scenarios. Chapter 3 introduced a precision decision-making method that quantitatively combines evidence-based knowledge with subjective preferences using prediction intervals to optimize decisions. A significant application of this method is in patient-centered treatment recommendations. Additionally, it can aid decision-making in other fields such as economics and demography. These methods were demonstrated through a case study of patients diagnosed with atrial fibrillation, providing patient-centered guidance for anticoagulant medication decisions by balancing the risks of stroke and bleeding. Chapters 4 and 5 addressed challenges in serial dilution experiments, which are widely used in immunology, virology, and other biomedical and pharmaceutical experiments. Chapter 4 proposed a new model based on joint likelihood to estimate the concentration of microorganisms in neat samples by modeling count data from the entire single dilution series, offering increased precision and accuracy compared to existing methods. Chapter 5 presented a complete workflow for estimating microorganism concentrations by enhancing automatic spot counting in images with spots representing viral plaque forming units (PFU), bacterial colony forming units (CFU), or spot forming units (SFU). The accuracy of the methods was significantly improved by introducing both empirical and theoretical bias correction techniques to address undercounting issues that arise when spots are densely populated. Overall, the proposed innovative statistical models and methods provide new tools for big data visualization, regression analysis, decision-making, and quantitative estimation, with applications extending beyond the medical and clinical fields to areas such as differential privacy, economics, and demography. This thesis advances data analysis techniques by developing novel statistical approaches specifically tailored to biomedical and clinical applications, while also offering generalizable tools for broader use.Ph.D.Includes bibliographical reference
Understanding viral polyproteins: insights from studies on human immunodeficiency virus type 1 and severe acute respiratory syndrome coronavirus 2
No full textIn recent decades, there has been a significant increase in viral epidemics and pandemics. From the global acquired immunodeficiency syndrome (AIDS) epidemic in the 1980s to the recent coronavirus disease 2019 (COVID-19) pandemic, both have caused thousands of deaths worldwide. These deadly viruses are evolving continuously despite the availability of FDA-approved direct antivirals for both HIV-1 and SARS-CoV-2, as well as successful vaccines for SARS-CoV-2. This constant viral evolution poses a significant challenge to scientists as they must deal with obstacles such as drug-resistance mutations or mutations to evade (natural or vaccine-mediated) immunity. Therefore, a comprehensive understanding of every aspect of the viral replication cycle is important for developing more effective treatments and potential cures.Human immunodeficiency virus 1 (HIV-1) is the causative agent of acquired immunodeficiency syndrome (AIDS), while severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is responsible for the coronavirus disease 2019 (COVID-19). They are classified as retroviruses and RNA viruses, respectively, and both utilize a “polyprotein strategy,” where the viral genome is translated into long precursor polyprotein(s) that undergo cleavage to produce essential viral proteins. This dissertation is dedicated to improving the understanding of viral polyproteins, particularly those from HIV-1 and SARS-CoV-2.
As of 2023, ~40 million people globally are living with HIV-1. While highly active antiretroviral therapy (HAART) has revolutionized HIV-1 treatment, challenges such as drug-resistance mutations, adherence to daily regimens, and reinfections exist. An in-depth knowledge of all the stages of the HIV-1 replication cycle is necessary to develop alternative strategies. HIV-1 polyproteins, specifically Gag and Gag-Pol, play an important role in the assembly and maturation stages of the replication cycle. Despite 40 years of extensive research on HIV-1, there is limited understanding of the Gag-Pol polyprotein, primarily due to the lack of suitable heterologous systems to produce this protein. This dissertation details protocols developed for producing soluble Gag-Pol in bacterial cells with sufficient yields to perform structural and biophysical studies. The single-particle cryo-EM structure of Gag-Pol showed a dimeric organization with a heterodimeric reverse transcriptase (RT) core. The structure analysis also showed low resolution density corresponding to PR which has provided insights into the role of RT in PR dimerization during maturation. HIV-1 assembly has been widely studied using Gag virus-like particles (VLPs) as a model system. Even though the Gag VLPs have significantly improved our current knowledge of the assembly process, they are incomplete mimics of the actual virus particles due to the absence of Gag-Pol in them. This dissertation has attempted to fill this gap by assembling VLPs in vitro containing both Gag and Gag-Pol polyproteins. These Gag:Gag-Pol VLPs may provide effective model systems for understanding the intricate mechanisms of the assembly and maturation process.
During the COVID-19 pandemic, SARS-CoV-2 has infected hundreds of millions and caused millions of deaths throughout the world. Polyprotein processing is a crucial process for viral replication. However, it is a poorly understood stage of the virus replication cycle. This dissertation has provided structural and biochemical insights into the processing of SARS-CoV-2 nsp7-11 and nsp7-8 polyprotein intermediates, whose mature products are key components of the replication-transcription complex (RTC). The integrative structural models of nsp7-11 and nsp7-8 were developed using data from cross-linking mass spectroscopy (XL-MS), hydrogen-deuterium exchange (HDX) MS, and small-angle X-ray scattering (SAXS) techniques. These models helped us understand the role of cleavage junction conformation and accessibility in determining the processing order. Additionally, this dissertation has also provided insights into two possible binding pathways by revealing the footprint of the polyprotein binding on Mpro. Finally, a proteolysis assay using the nsp7-11 polyprotein as substrate—benchmarked with nirmatrelvir—was used to characterize the effect of small molecule binders on Mpro-mediated processing. The results have provided insights into the allosteric regulation of Mpro activity. In summary, these structural and biochemical insights into SARS-CoV-2 polyproteins may help in understanding their role in the viral life cycle and provide a basis for targeting them using structure-based drug design and discovery.
SARS-CoV-2 nsp7 and nsp8 are essential cofactors of nsp12, which together form the viral RNA-dependent RNA polymerase or RdRp. This dissertation describes the expression and purification protocols of these two important viral proteins. XL-MS and HDX-MS techniques have been utilized to understand the dynamics of full-length nsp7:nsp8 heterodimer formation. These results have provided insights into the pathways of RTC formation which is required for the gene replication.
Lastly, this dissertation summarizes the polyprotein processing stage in the two viruses – HIV-1 and SARS-CoV-2. It combines the results from the above chapters to emphasize the pathways involved in polyprotein processing and the implications for antiviral drug discovery.Ph.D.Includes bibliographical reference
Material segmentation of remote sensing imagery
No full textLow-resolution remote sensing data makes it difficult to discern the composition of individual pixels, especially when isolated from their surroundings. This issue is accentuated by the limited and costly process of annotating this data, and high-resolution data is either publicly restricted, requires a significant time investment to provide meaning to our low-resolution data accurately, or is associated with considerable expense. In this thesis, we address these challenges by developing and improving methodologies for resolution material segmentation, which generates intuitive insights into the composition of remote-sensing images. Our major contribution is the creation of precise and accurate resolution consistent material maps trained on sparse datasets. Remarkably, our models achieve classification rates ranging from 82\% to 98\% with minimal training time, significantly reducing the computational resources required compared to modern deep learning methods, which typically necessitate hours to days of training. We explore various approaches to constructing and extending sparse datasets, employing pixel-wise segmentation methods that treat each pixel as a classification problem. The shift to pixel-wise framing allows us to leverage multiclass classification techniques and frame each pixel as a 14x1 feature vector, utilizing the rich multispectral information captured in our datasets. Additionally, we demonstrate that framing the problem as multiple binary classifications in a one-vs-rest framework results in a lower overall classification rate due to the inclusion of an uncertainty class. However, this approach enables us to generate high-quality material maps with classification rates exceeding 90\% in regions where expert models agree, while also identifying areas requiring further training or improved labeling. In parallel, we also employ our long training material segmentation pipeline, which uses modern encoder-decoder networks to produce over 99\% accuracy on our test set at the cost of many of the strengths afforded by our other models. From a research perspective, this thesis underscores the feasibility of research parallelism, illustrating how less hardware-intensive methods can be employed to gain valuable insights into model performance. Our findings not only advance the remote sensing material segmentation field but also provide a pathway for more accessible and efficient research in this domain.M.S.Includes bibliographical reference