Doctoral Dissertations (Winter 2014 to Present)

Permanent URI for this collection

https://udspace.udel.edu/handle/19716/12883

New submissions to the University of Delaware Doctoral Dissertations collection are added as they are released by the Graduate College. In most cases the Graduate College deposits all dissertations from a given semester after the official graduation date.

Doctoral dissertations from 1948 to present are also available online through Dissertations & Theses @ University of Delaware. Check DELCAT to locate print or microform copies of dissertations that are not available online.

More information is available at Dissertations & Theses.

Browse

Now showing 1 - 20 of 2904

Improving learning under data scarcity constraints: application in brain MRI, sonar, and natural images
(University of Delaware, 2025) Baker, Hassan
Lack of data significantly hampers machine learning approaches in domains with limited data. This shortage impedes the effective use of deep learning models, which are prone to overfitting and often perform poorly when processing data not seen during their training. The nature of this problem varies by application, necessitating tailored solutions. We focus on using machine learning to achieve different computer vision tasks on various imaging modalities: structural MRI from brain scans, sonar images, and natural images. ☐ To achieve abnormal tissue segmentation (brain lesion detection) from structural MRI, we propose a self-supervised task that exploits the intricate spatial structure in the brain. We take a patch taken from an MRI slice and attempt to learn the mapping to its location relative within a brain. We add to this task an estimation for the uncertainty of the predicted location. Then, for the downstream tasks of abnormality detection and segmentation, we use a combination of two scores, namely the estimated location error and the uncertainty, as an unsupervised abnormality score for the input patch. ☐ While this approach focuses on leveraging spatial context within available structural images, in many clinical scenarios, some MRI modalities may be missing or unavailable due to limited resources, acquisition time, or patient-specific constraints. To address this complementary challenge of modality scarcity, we propose a 3D two-stage model for many-to-many modality translation. This model achieves state-of-the-art performance in both reconstruction quality and inference time, making it a practical solution for completing missing modalities in multi-modal MRI pipelines. ☐ For natural images, we utilize the fact that they are composed of two parts: background and foreground objects, where the latter is defined as the salient parts of the images, in training a masking network to separate the two. In sonar images of the sea floor, this can separate objects from the background sea floor. To do this we propose a weakly-unsupervised training scheme to train a masking network that takes an input image and generates a mask for the foreground objects in the input image. This mask is used to generate a synthetic image with the foreground superimposed on a different background-only image, yielding a counterfactual image. We use the cluster assignments of background content of images to define a conditional statistical divergence between the generated counterfactual images and the real ones for each target background cluster. The trained model that minimizes this divergence can be used in downstream tasks such as foreground segmentation and classification. Additionally, counterfactual images composed of foreground objects overlaid onto different backgrounds that are not present in the training data are useful for data augmentation. ☐ While the proposed methods address core aspects of learning under data scarcity, they also reveal new directions for future work. First, finer-grained localization in Patch2Loc could be achieved by applying out-of-distribution detection techniques to spatially organized latent spaces, particularly to overcome the limitations imposed by fixed patch sizes. In the context of weakly supervised segmentation, the background clustering mechanism could be extended with dynamic or adaptive clustering methods to handle more complex, real-world backgrounds. Additionally, to mitigate hallucinations such as partial object removal, a discriminator could be employed. For modality translation, incorporating uncertainty modeling would help identify when a translation is ill-posed due to missing modality-specific content, thereby improving reliability in clinical settings. We also plan to extend the approach to other modalities such as stiffness maps estimated from MRE images. Furthermore, we observed that dynamic models can better estimate missing information during translation (e.g., the contrast of T1CE), but they may alter the structural integrity of the brain. Introducing structural regularization into these generative models could preserve anatomical fidelity and enhance translation performance.
Connecting phage genotypes to ecological function : ǂb replication protein modules reveal phage dynamics across diel and tidal cycles in Narragansett Bay
(University of Delaware, 2025) Ferrell, Barbra D.
Viruses are the most abundant biological entities on Earth and play critical roles in shaping microbial communities, influencing host metabolism, population dynamics, and nutrient cycling. Despite their ecological importance, the majority of viral diversity remains uncharacterized, and the connections between viral genomic features and infection phenotypes are poorly understood. Replication proteins, including Family A and B DNA polymerases (PolA, PolB), ribonucleotide reductases (RNR), and helicases, are central to viral genome replication and serve as informative markers for predicting viral infection strategies and ecological behavior. ☐ This dissertation establishes a framework for investigating the ecology of unknown viral populations across environmental gradients through replication module analysis. Aim 1 developed a reproducible workflow for identifying and quantifying viral populations in metagenomes based on co-occurring replication proteins, integrating contig assembly, functional annotation, abundance estimation, and phylogenetic placement. Aim 2 expanded detection to PolB–carrying populations, including cyanophage, by constructing a reference database and validating functional PolB proteins through active site and domain analyses. Aim 3 applied these methods to a 48-hour diel and tidal series in Narragansett Bay, Rhode Island, revealing distinct ecological patterns: temperate-associated populations (L762 PolA variants, E. coli numbering) remained stable across diel and tidal gradients, while populations encoding virulent-associated replication modules (F762/Y762 PolA, RNRs, superfamily 4 (SF4) helicases) exhibited condition-specific fluctuations linked to diel host metabolism and tidal transitions. ☐ This work demonstrates that replication module composition provides a robust approach for predicting viral infection strategy and ecological behavior, extending beyond single-protein analyses. The pipelines and PolB-focused developments enable systematic characterization of previously overlooked viral populations, linking genome content to ecological dynamics. By connecting replication machinery to viral population behavior, this research provides a scalable framework for exploring viral ecology across spatial and temporal gradients and lays the foundation for predictive bioinformatic approaches to infer infection strategies in uncultivated viral communities.
From the shore to the screen: how virtual reality and collective vulnerability influence perceptions of sea-level rise in Bowers, Delaware
(University of Delaware, 2025) McCusker, Joy
Bowers, Delaware, is a small, tight-knit coastal community that has lived with periodic flooding for generations. Rising sea levels and increasingly severe storms are now pushing the limits of the town’s infrastructure and testing its capacity to adapt. Despite these growing risks—exacerbated in Delaware by its low elevation—public engagement with climate adaptation remains limited. One key barrier is the psychological distance many feel toward climate change; its abstract nature often makes it seem distant or irrelevant to their daily lives. This study examines the impact of a virtual reality (VR) simulation of localized future sea-level rise (SLR) projections on participants’ perceptions of psychological distance to SLR. It also explores how residents perceive their vulnerability to SLR and the barriers identified in planning for the future. Using a quasi-experimental, mixed-methods single-case study design, the research collected quantitative data through Likert-scale surveys and qualitative data through semi-structured interviews. Three theoretical frameworks guided the analysis: Construal-Level Theory (CLT), Collective Vulnerability Theory (CVT), and the Social-Ecological Systems Framework (SESF). Integrating these frameworks enabled a transition from individual cognition to collective experience to systems-level insight. Findings reveal that the VR simulation significantly altered participants’ perceptions, on personal items, with statistically significant changes in 7 of the 12 measured items, which formed a Personal Impact Scale. This suggests that immersive, localized visualizations can reduce psychological distance, fostering engagement with climate risks. Findings highlight a sense of collective vulnerability within the Bowers community, despite perceived limited governmental influence. Despite a strong sense of social capital and collective action, Bowers still experiences vulnerability. Ultimately, this work sheds light on how perceptions of collective vulnerability, rooted in socioeconomic inequalities and institutional failures, influence individuals’ psychological distance to environmental threats. Findings demonstrate that psychological distance is not solely a cognitive construct but is deeply embedded within socio-political realities.
Offline politics, online control: rethinking the global rise of digital authoritarianism
(University of Delaware, 2025) Zaman, Fahmida
A growing body of literature which erupted in late 2000s have been studying the upsurge in authoritarian use of digital technologies. This phenomenon has been widely described as digital authoritarianism. Much of these extant literatures on this topic, however, a) often implicitly assume that regime type alone accounts for the emergence of digital authoritarianism resulting in an overemphasis on authoritarianism regimes, or b) focus disproportionately on external influences, particularly on China, as the primary driver of digital authoritarianism’s origin and diffusion. In contrast to such a one- dimensional approach which does not adequately engage with domestic dynamics, this dissertation project investigates domestic and international drivers of digital authoritarianism across regimes. The central question guiding this project is—what are the domestic and international forces shaping the rise of digital authoritarianism, and how do the dynamics of digital authoritarianism manifest across different political and regional contexts particularly in hybrid regimes? At chapter levels, the research questions are: to what extent does regime type explain variation in digital authoritarian practices across countries? Does a state's cyber capacity, whether technical, infrastructural, or resource-based, influence its tendency to engage in digital authoritarian practices? To what extent, if any, do digitally mobilized political actions contribute to the escalation of digital authoritarianism? (Chapter 2); to what extent is China actively exporting digital authoritarian practices to other states (Chapter 3); how do contemporary techniques of digital authoritarianism manifest in hybrid regimes and what underlying motivations drive such practice in these regimes? Across the three core chapters which combined cross-national statistical analysis (Chapter 2 and Chapter 3) and an in-depth case study of hybrid regime in Bangladesh (Chapter 4), the empirical evidence in this dissertation project underscores the significance of domestic factors in shaping driving the global emergence of digital authoritarianism. Furthermore, the findings challenge dominant narrative that attributes the global rise of digital authoritarianism primarily to China. Overall, this dissertation project draws attention to the role of domestic dynamics that shape adaption of digital authoritarianism and calls for greater caution in over-emphasizing on China’s role as the primary driver of digital authoritarian practices across the world.
Brain ECM-mimetic hydrogel platform to investigate cellular and extracellular dynamics in the early metastatic niche of triple-negative breast cancer
(University of Delaware, 2025) Freeman, Samuel
Brain metastasis of triple negative breast cancer (TNBC) rapidly progresses, causing severe neurological decline with a median survival of less than 6 months. This tragic disease is often difficult to identify with sufficient time for treatment and is exacerbated by a lack of effective pharmacological intervention. Replicating the biochemical and mechanical properties of the pre-metastatic niche in vitro is a critical step in expediting the development of new therapeutics. However, a high fidelity and reproducible model system is needed. In an effort to develop a model to simulate the adhesion signaling and cell to cell communication present in the pre-metastatic brain, we produced and characterized an in vitro, PEG based hydrogel and co-culture method to evaluate the cellular and extracellular dynamics of early cancer colonization. ☐ To quantify the influence of a brain-mimetic microenvironment on brain metastatic TNBC, we encapsulated and cultured the TNBC cell line, MDA-MB-231 (P231), and its braintropic subline MDA-MB-231-BrM2a-831 (BrM2a), in three premetastatic niches: a highly cell adhesive and highly cell degradable permissive niche, a highly adhesive but less degradable niche, and a non-adhesive but highly degradable niche. To mimic brain ECM, we functionalized the adhesive formulations with a brain-mimetic peptide cocktail and compared the cell responses to a ‘generic’ RGDS-only functionalization. This suite of conditions allowed us to investigate the influence of integrin-mediated adhesion, cell-mediated degradation, and cell type on the fate of P231s and BrM2as. By quantifying viability, viable cell density, apoptosis, proliferation, and morphology we demonstrate that brain-mimetic adhesion has little to no impact on P231 phenotype, but the BrM2as display reduced viable cell density, reduced proliferation, and a higher proportion of both spherical clusters and spherical individual cells compared to the generic RGDS-functionalized niches. ☐ With the demonstrated impact of brain-mimetic adhesion on brain preferential cells, we implemented the brain-mimetic functionalization to characterize the cellular response of cell native to the central nervous system. Primary human astrocytes and iPSC derived neural progenitor cells (NPC) were encapsulated and evaluated. In the astrocytes these metrics show modest differences between the brain-mimetic and RGDS functionalizations with reduced apoptosis in the brain permissive formulation. Both permissive formulations showed high viability and a larger portion of elongated cells than the adhesion restricted formulation. This shows that in monoculture, the astrocytes require ECM adhesion signaling to survive well but the diversity of the brain mimetic functionalization aids in reducing apoptosis. The iPSC derived NPCs were encapsulated in the brain mimetic permissive formulation as either single cells or spheres and differentiated toward forebrain neurons. Both seeding conditions produce clusters with robust neurite/axonal extension with the enlarged size of the spheres allowing extension across the surface of the hydrogel and the single cell encapsulated samples show projections that are more subdued in length but far more numerous. This demonstrates the compatibility of the brain-mimetic permissive formulation to cultivate complex neuronal cultures and further replicate the environment of the brain. ☐ Having characterized the baseline response of astrocytes and BrM2as, we proceeded to evaluate them in coculture. Similar to the astrocyte mono-culture, the difference between the brain-mimetic and RGDS functionalization was slight, with the brain permissive formulation producing a greater proportion of elongated clusters. To closer examine the protein expression differences induced by these culture conditions, proteomics analysis was performed. This analysis revealed coculture samples encapsulated in the brain mimetic permissive formulation have significant changes to structural and regulatory ECM expression as well as secreted immune signaling factors. Taken together these studies present a well characterized brain mimetic hydrogel platform to cultivate central nervous system cells and provide a tool to investigate brain metastatic cancer in a tissue relevant context. Future implementation of this platform could be to model other cancers that frequently metastasize to the brain and investigate the mechanisms involved in cancer progression and response of the native cell population.
Wall temperature and high enthalpy effects on hypersonic boundary layer stability and transition
(University of Delaware, 2025) Kafle, Sulav
The hypersonic boundary layer transition study is crucial for controlled and sustainable flight. Although crucial, the mechanisms underlying the transition are still poorly understood, even in a low-noise environment. Understanding of these extreme environment flow phenomena can lead to significant advances in aerospace flight technologies. Different modes of disturbances present in the hypersonic boundary layer undergo modal growth eventually leading to turbulence. ☐ The objective of this dissertation is to understand the dynamics of modes and their interactions due to wall temperature and high-enthalpy effects on hypersonic boundary layer transition. This research study utilizes computational fluid dynamics (CFD) as well as stability analysis tools such as linear stability theory (LST), linear parabolized stability equations (PSE), and non-linear parabolized stability equations (NPSE). This research combines theoretical understanding of first and second-mode instability with practical application to predict and mitigate turbulent transition in hypersonic boundary layers. The mean flow Lagrangian invariants are introduced to relate it with obliqueness of the first-mode instability. The effects of stream-wise thermal gradients on the growth of second-mode instability are investigated. The computational results for the pattern wall temperature study are compared with experiments conducted in the AFOSR–Notre Dame Large Mach-6 Quiet Tunnel at the University of Notre Dame and show good consistency. The wall thermal configurations proposed in this study significantly delay the laminar-to-turbulent transition that arises due to second-mode instabilities. In addition to that, this research presents unique wall thermal patterns that do not affect the growth of second-mode instabilities. The computational results for high-enthalpy studies are compared with other numerical codes. The sensitivity of high-enthalpy hypersonic boundary layer flows to non-linearity is investigated. A 1D-CNN machine learning model was proposed to predict the critical N-factor. This data-driven model presented in this dissertation is the one that can be used as a preliminary assessment to predict the transition rapidly with minimal computational effort.
Strengthening staff knowledge to support adults with disabilities in community fitness facilities
(University of Delaware, 2025) Jadach, John A.
Persistent health disparities among adults with disabilities are linked, in part, to limited disability-specific knowledge among community fitness staff, which creates a barrier to the consistent delivery of inclusive, evidence-informed wellness programming (Obrusnikova, Jadach, Cavalier, & Firkin, 2023; Rimmer & Vanderbom, 2016). In my role as co-founder of Endless Possibilities in the Community (EPIC), a 501(c)(3) nonprofit that partners with individuals with disabilities and community stakeholders to promote inclusive community fitness, I undertook this Educational Leadership Portfolio (ELP) with the specific goal of identifying, implementing, and evaluating an example of “good practice” in professional development training for EPIC’s fitness staff. The long term goal is to reduce health disparities among adults with disabilities by strengthening staff knowledge, self-efficacy, and instructional competence while increasing participant engagement in physical activity. ☐ This ELP study first identified the Empowerment Model (Moran, Block, & Taliaferro, 2014) as an example of a professional development framework for inclusive fitness. Building on that foundation, the study then implemented and evaluated the Strategies of Success (SOS) online instructional modules, designed to increase EPIC staff members’ knowledge in disability awareness and inclusive fitness practices within community fitness facilities (CFFs). Using a mixed-methods design, the study quantitatively assessed pre/post knowledge and self-efficacy of staff participants across four SOS modules (Addressing Challenging Behaviors, Planning Inclusive Programs, Modifying Instruction for Inclusiveness, Accessibility Considerations) and gathered qualitative feedback on application to practice through focus groups. Study participants demonstrated a mean knowledge gain of 22 points (scale of 0–100) on the four modules. Self-efficacy was examined as a complementary outcome and improved on average (0.5 on a 0–10 scale). Focus group data indicated that the modules were valuable for onboarding new staff, equipping them with foundational disability awareness and fitness-specific knowledge and strategies, while also providing experienced staff with a review and opportunities to expand their repertoire of inclusive instructional knowledge and practices. Focus group findings informed creation of a draft EPIC professional development policy, procedural guidelines, and a staff resource manual. The intent is to use these mechanisms to embed ongoing knowledge building into organizational routines and support sustainability. ☐ Taken together, the findings demonstrate that CFF staff training needs, such as those at EPIC, can be addressed through a research-informed professional development approach centered on knowledge acquisition, which may serve as a foundation to increase staff confidence and capacity in an effort to reduce persistent physical health disparities among adults with disabilities.
A multi-scale numerical study on coastal hydrodynamics, sediment transport, and morphodynamics
(University of Delaware, 2025) Zhang, Jiaye
This dissertation reports studies that advance the scientific understanding of coastal processes by elucidating the coupled dynamics among waves, flows, sediment transport, and morphodynamics across multiple spatiotemporal scales in the coastal environments. Utilizing high-fidelity computational fluid dynamics (CFD) with process-based morphodynamic models, this multi-scale numerical investigation spans three representative scales of coastal dynamics: (1) fine-scale sand particle sorting driven by grain-turbulence interactions, (2) small-scale wave-driven sand ripple evolution and the mobility of underwater munitions, and (3) intermediate-scale storm-induced cross-shore beach profile changes. By integrating insights across these scales, the dissertation seeks to reveal the fundamental coastal processes and underlying physical mechanisms observed in laboratory and field settings. ☐ At the fine scale, simulations using an Eulerian-Lagrangian two-phase model, CFD-DEM, investigate the vertical sorting of polydispersed native sand and denser nonnative particles (e.g., olivine for coastal carbon removal) under oscillatory sheet-flow conditions. Results show that competing upward and downward migration mechanisms control nonnative particle fate, offering insights for deploying the optimum size of nonnative particles that can stay in the active layer to maximize their weathering and carbon capture. ☐ At the small scale, large-eddy simulations (LES) using SedFoam, an Eulerian two-phase model, resolve turbulent coherent structures (TCS) that drive sub-orbital ripple formation from an initially flat sand bed under oscillatory flow. The results demonstrate that TCS are the dominant mechanism initiating the formation of three-dimensional (3D) bed features. At a later stage, when ripples grow sufficiently larger than the integral length scale of turbulence, the wave orbital motion takes over and becomes the dominating driver for the subsequent ripple evolution to equilibrium. These findings elucidate the fundamental coupling between TCS evolution and sediment transport during ripple development. Furthermore, by extending SedFoam to incorporate six-degree-of-freedom for object motion with complete flow-sediment-object interaction coupling, the new model was validated to simulate the onset motion behavior of underwater munitions. The simulation reveals that hydrodynamic forcing and object properties, such as object density, size, and initial burial depth, jointly influence the motion behavior of small objects driven by oscillatory flows. ☐ At the intermediate scale, cross-shore hydrodynamics and morphodynamics in the surf zone are first investigated using the process-based model XBeach-Surfbeat (XB-SB) and large-wave flume data for an erosive event. Simulations of storm-induced berm erosion, sediment transport, and sandbar formation reveal that default model settings overpredict undertow, leading to excessive berm erosion. Systematic calibration produces optimized coefficients that improve morphodynamic predictions based on a well-calibrated undertow. Extending this work to field conditions, XB-SB is applied to two 2023 experiments at the Field Research Facility in Duck, North Carolina, representing an accretive (March 2023) and an erosive event (November 2023). Results indicate that adjusting existing model parameters alone cannot achieve consistent agreement across the shoreline and sandbar regions, highlighting the need to incorporate geotechnical properties into morphodynamic models to represent increased sediment strength in the intertidal zone and to stabilize the foreshore under energetic wave conditions. ☐ Collectively, the findings of this dissertation establish a coherent linkage of physical processes across scales, demonstrating that integrating multi-scale insights yields a more unified understanding of coupled coastal dynamics and enhances the predictive capability of reduced-complexity models.
Catalytic activation of bioorthogonal chemistry without photochemistry
(University of Delaware, 2025) Neglia, Sophia
My research centers on developing novel strategies to activate the rapid tetrazine ligation reaction through “photochemistry-free” oxidation, enabling applications in biological systems. I demonstrated that the stable precursor dihydrotetrazine (DHTz) can be oxidized by enzymes and small molecules, via catalytic and stoichiometric pathways, to trigger subsequent tetrazine-trans-cyclooctene (TCO) ligation in the absence of light. I achieved the first intracellular enzymatic labeling reaction of DHTz in live cells using “dark” catalysis- defined as catalytic chemical reactions that occur without light. Building on this, I made progress toward development of new proximity labeling systems that leverage dark catalysis, and I have evaluated a range of dark catalysts and oxidants, comparing their efficiencies and biocompatibilities with the DHTz system. ☐ In chapter 1, I discuss my work with ascorbate peroxidase (APEX2) to turn on the bioorthogonal tetrazine ligation reaction. Kinetic studies revealed that APEX2-catalyzed oxidation of DHTz is enhanced by superoxide dismutase (SOD), a ubiquitous mammalian enzyme that regulates oxidative stress by converting superoxide into molecular oxygen (O2) and hydrogen peroxide (H2O2). The APEX2 oxidation with SOD achieved a catalytic efficiency of kcat/KM 4.90 × 103 M–1s–1 in vitro. While H2O2 is not strictly required, the addition of 10 µM H2O2 accelerated the oxidation reaction both in vitro and in live cells. Using a dual-transfection protocol expressing cytosolic APEX2 and HaloTag-DHTz conjugate, I demonstrated that APEX2 promotes DHTz oxidation and subsequent Diels- Alder chemistry in live HeLa cells. Labeling with a fluorophore-tagged TCO probe was confirmed via in-gel fluorescence, Western blot analysis, and confocal microscopy. In live PC3 cells, APEX2 also catalyzed the oxidation of a DHTz conjugated to an endogenous monoacylglycerol lipase (MAGL) through a selective covalent warhead. ☐ In chapter 2, I describe my screenings for proximity labeling using APEX2 fused to various proteins of interest (POI). Previous studies showed APEX2-biotin-phenol systems label proteins within a 20 nm radius. I wanted to compare this method for proximity labeling to a complementary approach based on enzymatically activatable bioorthogonal chemistry. I labeled lysine residues proteome-wide with a TCO-N-hydroxysuccinimide (NHS) ester, then activated a DHTz by APEX2 bearing a biological alkyne handle to assess proximity-based differences. The alkyne- labeled proteins were conjugated to biotin-azide or TAMRA-azide via a Cu(I)-catalyzed azide-alkyne cycloaddition (CuAAC). Among the constructs tested, cytosolic APEX2-GFP detected new proteins. While other constructs targeting cereblon (CRBN), the outer mitochondrial matrix (OMM), and a nuclear export signal (NES) did not show detectable labeling under these conditions, these results offered insight to the complexities of expression levels and accessibility for effective proximity labeling methods. To further analyze labeling targets, I optimized a small-scale streptavidin enrichment protocol. ☐ In chapter 3, I evaluate DHTz activation by small molecules for dark catalysis. Building on the heme enzyme of APEX2, I demonstrated that the iron porphyrin complex, Fe (III) tetrakis (N-methyl-4′-pyridyl) porphyrinato (FeTMPyP) activates DHTz in vitro and in extracellular environments. In addition, ferrocenium tetrafluoroborate promoted the rapid and stoichiometric oxidation of DHTz with a second-order rate constant of k2 = 1.82 x 105 M-1s-1. The low molecular weight of ferrocenium ion and the extremely rapid kinetics of DHTz oxidation make it a uniquely promising oxidant among the small molecules tested for tetrazine activation in biological applications. The compact structure of ferrocenium also offers an excellent scaffold for further functionalization. Beyond ferrocenium, I evaluated a range of other oxidants. Copper (II) sulfate (CuSO4) catalytically oxidized DHTz, while quinones acted as stoichiometric oxidants. Each class of oxidant presents unique advantages and limitations, which I critically analyzed in the context of their potential for biological compatibility, efficiency, and tunability.
Identification and regulation of heritable biomarkers for manufacturing stress tolerance in CHO to improve monoclonal antibody production performance
(University of Delaware, 2025) Grissom, Spencer
Biologics, a class of therapeutics derived from living systems, have revolutionized modern medicine and unlocked treatment options for a wide class of diseases. For its production, Chinese hamster ovary (CHO) cells have gradually become the preferred host organism due to its compatibility with large reactor systems, high production, and patterns of human-like post-translational modifications. Any efforts to increase productivity translates to lower operating costs and increases the affordability and accessibility to therapeutics. However, large-scale bioreactors and CHO cell’s inefficient overflow metabolism contribute to the accumulation of inhibitory environmental perturbations that invariably reduce cell growth and volumetric productivity. These include elevated osmolality, oxidative stress, ammonia, and lactate levels that disrupt cellular structure, cause cell-cycle dysregulation, and may induce apoptosis. Traditional cell line development (CLD) workflows often do not screen stress tolerance during clonal evaluation and therefore represents a vulnerability in isolating cell lines suited for scale-up and stress resiliency. Approaches to improve stress tolerance have largely been centered around bioreactor control strategies and knockout of metabolic or apoptotic regulating genes. These have relied on conventional heuristics for cell health, but do not consider or characterize the adaptative regulatory networks that form robust resistance. In this study, a novel approach utilizing population-based transcriptomics for the identification of unique biomarkers for bet-hedging and stress tolerance is demonstrated. Downstream genetic engineering was used to generate biomanufacturing stress-tolerant cell lines with improved growth characteristics across two different biomanufacturing-relevant environmental perturbations. ☐ Initially, three of the most commonly explored stress agents (ammonia, lactate, and osmolality) were used to stress shock monoclonal antibody (mAb) producing CHO cells in fed-batch to determine the phenotypic, morphological, and transcriptomic effects of perturbation. At supplemented concentrations of 10 mM ammonia and 100 mOsm/kg, cell specific growth rates, peak viable cell densities, and the integral of viable cell density (IVCD) were significantly reduced. This translated to a reduction in volumetric productivity or loss of titer. At lactate concentrations of 15 mM, inhibitory effects were not observed, possibly due to a higher baseline tolerance in the cell line or a context dependency in well-controlled environments. Disruption of lysosomal structure, hydrolase activity, and amino acid metabolism were observed in ammonia stress and an increase in surface bound transporters and translational activity were observed in osmotic stress. While this information provided insight regarding the phenotypic effects of stress, broad differential gene expression analysis highlighted confounding and ambiguous patterns of expression that convolutes the search for rational engineering targets. ☐ While traditional transcriptomics are useful, they are often insufficient in elucidating clear targets for genetic engineering. An alternative method for identifying stress-associated biomarkers was explored using a population-based transcriptomic tool known as MemorySeq. This method utilizes RNASeq fluctuation analysis of roughly 40 single-cell derived populations after 17 generations of growth to identify highly variable genes that correlate to intermediate, transient, and heritable memory states. These unique transgenerational properties have been linked to bet-hedging and broad stress resistance mechanisms in cancer, plant, and microbial cells. Using this tool, 199 unique genes with heritable properties were identified and found to be enriched in signaling/communication, regulation of cell proliferation, and apoptosis regulation functionalities. They also significantly overlapped with the differentially expressed genes in stress shocked populations, highlighting their role in early pre-stress resistance states. ☐ With genetic targets identified, stable and homogenous genetic engineering tools would permit replicable characterization of their effect on cell health. To streamline CHO cell line engineering efforts and regulation of native genes, a flexible and modular targeted integration toolkit was developed to accelerate vector construction and stable integration. This toolkit featured a 16 component one-pot Golden Gate (GG) reaction for plug-and-play assembly of complex mammalian expression cassettes. With efficiencies ranging from 100% in 7 element reactions to 35% in 16 element reactions, multifaceted vectors could be generated to optimize cis-acting regulatory elements. The toolkit also outlined a site-specific integration (SSI) workflow displaying 90-100% efficiency of complex payloads utilizing the Cre/lox recombinase system. SSI significantly reduces the transcriptional and transgene stability heterogeneity associated with random integration, therefore isolating intentional changes in vector design to phenotypic deviations. ☐ Finally, with the MemorySeq genetic targets and the SSI toolkit, regulating native gene expression allowed for the induction of stress-tolerant phenotypes. Development and optimization of CRISPR activation/interference (CRISPRa/i) systems allowed for activation and repression of three genes with heritable properties. These included activating transcription factor 3 (Atf3), immediate early response 3 (Ier3), and heme oxygenase-1 (Hmox1). These three genes have been indicated in other cell lines as playing a role in stress detection and response, but never in CHO. CRISPR facilitated activation of Atf3 and Hmox1 and repression of Ier3 resulted in a 30-40% increase in integral viable cell density and peak viable cell density in both ammonia and osmotic stress fed-batch conditions. This translated to measurable improvements in volumetric productivity that may be further compounded in an appropriately controlled bioreactor. Concomitant with improved growth was also a reduction in broad rates of apoptosis, indicating a cytoprotective feature of some of these genes or regulatory pathways. Overall, this thesis reflects a novel approach for identifying, characterizing, and engineering stress tolerant phenotypes in CHO using heritable properties as an early-stress resistance biomarker. Continued exploration of genes displaying these properties may highlight robust rational engineering targets for the development of novel CHO host strains with improved performance in manufacturing scale-bioreactors.
Demystifying water and carbon cycling of ghost forests: harbingers of coastal change in a warming world
(University of Delaware, 2025) O'Halloran, Robyn C.
Coastal forests are undergoing rapid, climate-driven mortality and are transitioning from carbon sinks to emergent carbon sources under accelerating sea-level rise, fundamentally reorganizing how carbon moves through these ecosystems. This study investigates how canopy decline in sweet gum trees, Liquidambar styraciflua, alters the generation, molecular composition, and redox functionality of dissolved organic matter (DOM) exported via stemflow, a concentrated pathway that links the canopy directly to the forest floor. Stemflow acts as a focal point of water and carbon delivery, creating transient biogeochemical hotspots at the base of trees where rainfall is funneled and chemically transformed. The overarching hypothesis guiding this work is that sea level rise driven tree stress and mortality disrupt canopy-groundwater carbon coupling by altering physiological processes within trees and, in turn, changing the quantity, composition, and routing of DOC and water through coastal forest ecosystems. ☐ Hydrologic monitoring across healthy, moribund, and dead trees revealed that localized water inputs from stemflow drive sharp, event-scale increases in near-trunk groundwater levels. Around healthy trees, small storms below 16 mm produced threshold-like rises in water table elevation, whereas recharge was more uniform in areas around dead trees. Mean stemflow volumes declined from healthy trees to moribund trees and were negligible in dead trees. Despite DOC concentrations that were nearly double in dead-tree stemflow relative to healthy trees, total DOM flux per unit basal area declined by more than 70 %, indicating a collapse of canopy-mediated carbon delivery. Groundwater DOM near mortality zones exhibited stronger humic and terrestrial fluorescence and reduced microbial and protein-like components, consistent with reduced vertical carbon transfer and altered subsurface processing. ☐ Molecular analysis confirmed that sugars - including neutral sugars, amino sugars, and raffinose - constitute a major and previously overlooked fraction of stemflow carbon. Total neutral sugar concentrations averaged 871 ± 1434 µM, roughly ten times higher than global forest mean values. Diagnostic ratio1s of neutral sugars distinguished moribund trees, indicating cell-wall remodeling and microbial turnover under physiological stress. Cumulative sugar fluxes followed a unique and episodic trend with moribund trees, consistent with transient leakage during metabolic decline. The episodic release of these highly labile carbohydrates during the moribund phase represents a substantial carbon pulse to near-trunk soils, fueling short-lived microbial and nutrient turnover. As canopy function collapsed, both sugar and DOC fluxes diminished sharply, signaling the loss of concentrated, high-quality energy inputs that once sustained microbial hotspots at the forest base. ☐ Optical, lignin, and electrochemical analyses revealed that the oxidative state of stemflow DOM shifted systematically with canopy health. Healthy trees contained reduced, antioxidant-rich organic matter, whereas moribund and dead trees showed oxidized, lignin-dominated DOM and diminished antioxidant content. Lignin phenol composition shifted toward greater vanillyl and syringyl oxidation, reflected in higher lignin oxidation ratios, capturing a transition from reactive to inert carbon. These patterns show that canopy decline directly reduces the redox buffering potential of exported DOM, constraining the ability of near-surface soils to maintain reduced conditions during storm events. ☐ Tree mortality is redefining carbon and nutrient cycling in coastal forests by coupling hydrological, molecular, and electrochemical change. The collapse of stemflow generation, loss of biolabile carbon fluxes, and oxidation of canopy-derived DOM mark a transition in forest function from systems that actively cycle and reduce carbon to those that passively release it. By tracing these transformations from rainfall to groundwater, this work identifies stemflow as a sensitive indicator of forest physiological decline and a mechanistic link between canopy health and subsurface carbon reactivity. As sea level rises, coastal forests lose not only their canopy cover but also their capacity to mediate the flow of carbon and electrons that sustain ecosystem resilience.
Mitigating product abuse through privacy-preserving and secure technologies in digital and industrial systems
(University of Delaware, 2025) Sari, Alparslan
The rapid digitalization of modern life has enabled unprecedented convenience and efficiency while simultaneously creating new opportunities for exploitation, misuse, and privacy violations. This dissertation investigates product abuse (the intentional misuse or manipulation of technological systems beyond their intended design) across distinct digital domains to surface its security, privacy, and operational implications. Using real world datasets, deployed prototypes, and empirical vulnerability assessments, it provides a cross domain examination of how abuse emerges and how defenses succeed or fail in practice. ☐ Case Study 1 (Email Tracking) analyzes how embedded tracking beacons in email communication can be repurposed as tools for covert surveillance and behavioral profiling. Through large scale measurement and analysis, the study exposes the privacy risks posed by such mechanisms, highlighting how legitimate business tools can cross the boundary into privacy abuse. ☐ Case Study 2 (CAPTCHA) surveys 24,000+ web pages from the Alexa Top 50K and correlates implementation patterns with 179 MITRE CVEs (2005–2025). The study finds that most failures stem from implementation errors, weak server-side validation, and supply chain issues, not the intrinsic design of challenges and documents how AI assisted solvers & paid solving economies further erode resilience, with practical hardening recommendations. ☐ Case Study 3 (Industrial IoT / WMS) examines abuse in industrial environments integrating Decision Support Systems, IoT devices, and Warehouse Management Systems. Drawing on a deployed prototype and operational data, it identifies attack surfaces that enable product manipulation, data leakage, and supply chain interference, and proposes blockchain backed audit trails, stronger authentication, and anomaly detection to enhance cyber resilience. ☐ Case Study 4 (Automated Crypto Trading) evaluates Mean Reversion, Arbitrage, Grid Trading, and Mean Deviation strategies as both efficiency enablers and abuse vectors. Experiments highlight how automation, if poorly designed or exploited, can induce market manipulation and systemic instability, motivating transparency, guardrails, and regulation aware algorithmic design. ☐ The dissertation (i) consolidates empirical evidence that product abuse recurs across heterogeneous systems; (ii) maps dominant failure modes from client-side exposure and automation to server-side validation gaps and supply chain weaknesses; (iii) demonstrates deployable countermeasures, privacy preserving email defenses, CAPTCHA hardening practices, IIoT/WMS auditability and access control, and ethics \& compliance aware algorithm design; and (iv) offers a practical threat informed rubric for engineering teams to anticipate misuse, not merely react to incidents. ☐ In conclusion, this dissertation offers both diagnostic and prescriptive perspectives on digital product abuse. It establishes that while product abuse cannot be fully eliminated, it can be systematically reduced through better architecture, stronger accountability, and adaptive security mechanisms that evolve alongside technological progress. By capturing the interplay between innovation, exploitation, and defense, this work contributes to the ongoing discourse on building secure, privacy preserving, and trustworthy digital ecosystems.
Applications of computational optimal transport in machine learning and signal processing
(University of Delaware, 2025) Riaz, Bilal
Recently, there has been a surge of interest in using optimal transport between probability distributions to measure the Wasserstein distance and enable better machine learning systems. More specifically, optimal transport can be used to define clustering algorithms, semi-supervised learning algorithms, and techniques for data compression and for correcting for covariate shifts in classification tasks. Additionally, Wasserstein distances can be used as cost functions in generative modeling and as constraints for robust modeling. The tremendous success of these techniques in wide application domains is due to the fact that optimal transport combines the related but distinct concepts of geometric distances and statistical divergences. ☐ The first work in this dissertation thoroughly investigates variants of optimal transport to deal with the cases where a subset of the support of one distribution aligns with complete support of another distribution, such as in the case of a carefully curated dataset that can be augmented by a source of less reliable data. In our experiments we demonstrated the utility of our approach in partial point cloud alignment, color transfer, positive-unlabeled (PU) learning and semi-supervised learning. Additionally, we propose to investigate the effect of partial alignment in generative modeling and to examine partial alignment in the case of global covariate-shift correction in classification tasks. ☐ In the second work for this dissertation, we investigate partial optimal transport in the case of two or more stochastic processes with application to matching bio-signals represented as univariate stochastic processes from a population of subjects, where the representation space underlying the transport is not Euclidean. In particular, we consider the case where spectral patterns observed in short-time windows can occur at different time scales for different processes. We seek a monotonic transformation of the spectra of each process that minimizes the Wasserstein distance between the distribution of spectra across windows. We anticipate that the spectral alignment for multiple subjects with different frequency spreads can enhance the performance of downstream learning systems. That is, learning on the aligned data performs better than learning on the original data. This has wide applications in cases where the machine learning system is better off learning to be invariant to the time scale. ☐ In the third work for this dissertation, we focus on the development of algorithms for neural network parametrized support subset selection approaches, where we only have access to the sample from underlying data distributions. More specifically, we developed algorithms for training neural network parameterized Monge-like maps in static formulation of continuous subset alignment and velocity-fields in dynamic formulation of continuous subset alignment. We applied our frameworks to PU-Learning and latent-space image alignment problems.
Evaluation of source water contribution to tidal marshland using stable isotopes (²H, ¹⁸O, ¹⁷O) of water
(University of Delaware, 2025) Bradach, Sophia M.
Sea-level rise is increasing saltwater intrusion into coastal marshes, altering porewater chemistry and threatening ecosystem functions such as nutrient cycling and carbon storage. Bulk salinity tracers (e.g., EC, Cl⁻) capture tidal mixing but struggle to distinguish precipitation inputs and evaporative enrichment. This study investigates whether stable water isotopes (δ²H, δ¹⁸O, δ¹⁷O) paired with end-member mixing analysis (EMMA) provide finer resolution of water sources and their biogeochemical impact than salinity alone. Porewater was sampled along a forest-to-channel transect across depths, seasons, and spring/neap tides. This work analyzed isotopes, EC, redox (Eh), and major ions, and compared simple two-endmember models with end-member mixing analysis (EMMA) using (a) isotopes only and (b) isotopes+EC, and the calculation of an Evaporative Enrichment Index (EEI). Direct comparison of isotope- and EC-based models revealed strong agreement at intermediate seawater fractions but divergence in interior zones. The isotope-only EMMA retained clear seasonal and tidal variability tied to recharge and evaporation, while the isotope + EC EMMA collapsed to a conservative, salinity-dominated axis. Incorporating an Evaporative Enrichment Index (EEI) corrected EC-derived estimates by isolating isotopic enrichment from evaporation and transpiration, producing a process-aware mixing framework. Spatial and chemical patterns aligned with marsh zonation: near-channel sites showed rapid flushing and redox oscillation; the transition zone exhibited prolonged residence, evaporative enrichment, and mobilization of Fe, Mn, and P. Isotope and salinity-informed models capture complementary process signals and not interchangeable estimates. Integrating isotopic corrections such as EEI enhances salinity-based models, providing a mechanistic framework for predicting how hydrologic and redox gradients reorganize as marsh zones compress under rising sea levels.
Populism talks: contemporary drivers of the decline in global climate change cooperation
(University of Delaware, 2025) Nam, Jiwon
The issue of climate change has been annually negotiated at the international level under the auspices of the United Nations Framework Convention on Climate Change (UNFCCC) for well over 25 years. Yet, and despite this quarter century of continuous, and robust negotiation, international climate change negotiations not only have repeatedly fallen short in reaching a comprehensive climate change agreement but have also worsened in their cooperative progress. What explains the ever-increasing gap between routine negotiation over climate change agreements and nation-states’ (in)abilities to reach effective and timely agreements on climate change? To answer this question, I posit that the recent global rise of populism has adversely influenced states’ abilities to reach international climate change agreements. To test this proposition, I first apply a Structural Topic Model to UNFCCC Conference of the Parties (COPs) speeches from the 16th COPs to the 25th COPs, as made by high-level country representatives. After extracting 25 topics from the speeches, I evaluate whether populist heads-of-state influence certain countries to negotiate over climate change in unique manners. I then pair this automated text analysis with qualitative case studies and a quantitative analysis of actual policy outcomes (i.e., annual changes in CO2 emissions and renewable energy consumption). In each respect, I find that in most cases populist leaders express and exhibit less supportive stances towards climate change cooperation in favor of greater anti-elitism, isolationism, and sovereignty-reinforcing stances, priorities, and policy outcomes. However, I find that the presence of right-wing populist leaders does not affect countries’ level of CO2 emissions, whereas the presence of right-wing populist leaders is associated with a decrease in a country’s renewable energy consumption as a percentage of total energy consumption. I also found that right-wing populist leaders lack in the implementation of effective environmental policies that will benefit the country in the long run.
Computational and experimental approaches to quantify the influence of pathlogical hemodynamics on hippocampal astrocyte dysfunction
(University of Delaware, 2025) Taitano, Ryan
Alzheimer’s disease (AD) and vascular dementia (VD) are major causes of disability and death in people over 65. While the cellular and molecular mechanisms that govern the initiation and progression of AD/VD are not fully understood, evidence suggests that changes in mechanical cues including high blood pressure, age-related arterial stiffening, and age-related brain softening are likely contributors. Arterial stiffening of elastic blood-vessels is thought to change normal blood flow patterns, thus resulting in neuronal inflammation and injury via mechanical strain-mediated mechanisms. Therefore, I hypothesize that exposure to pathological changes in elasticity, in both the µvessel wall and brain tissue, and high magnitude blood pressure exacerbates and astrocyte injury due to increased mechanical strain transmission to the surrounding tissue. Here, I propose investigating my hypothesis through the following means: (1) Develop methodology for characterizing micromechanical properties of hydrogels via microindentation. (2) Develop a predictive computational model of brain-tissue strain as a function of pressure and tissue elasticity. (3) Develop an in vitro microfluidic model to control pulse pressure to determine its influence on hippocampal astrocyte behavior.
Do protective factors matter?: examining the impact of cumulative risk and psychological well-being
(University of Delaware, 2025) Woodson, Christina
Exposure to cumulative risks during childhood and adolescence can have profound and lasting effects on psychological well-being. Fortunately, not all individuals who experience adverse circumstances develop negative outcomes. Protective factors may explain why some individuals flourish and thrive despite adversity. While a vast majority of research has focused on understanding risk factors and cumulative risk, research on protective factors remains understudied. The purpose of the current study is to identify protective factors, specifically clusters that can buffer the effects of cumulative risk. Rutter’s Resilience Theory and Bronfenbrenner’s Bioecological Theory guide this investigation. Latent Class Analysis (LCA) is used to examine how multiple protective factors can cluster together to influence psychological outcomes. Results indicate that cumulative risk significantly predicts lower psychological well-being, with perceived discrimination providing additional explanatory power. Regression analyses confirmed that cumulative risk significantly predicted reduced psychological well-being (p < .001), yet moderation analyses revealed that certain protective clusters, particularly Class 3, mitigated this impact. These findings emphasize the importance of a strength-based approach. Exploring strengths alongside risks allows for a more comprehensive understanding of how one’s existing protective resources (i.e., including confidence, self-perceived intelligence, self-rated health, life expectancy, religiosity, wealth, social support) promotes resilience and psychological well-being in the future.
Manipulating carbon, electron, and nitrogen exchange in syntrophic Clostridium co-cultures for robust, scalable, carbon-negative chemical production
(University of Delaware, 2025) Willis, Noah
The classic acetone-butanol-ethanol (“ABE”) fermentation, which is based on Clostridium acetobutylicum, suffers from low carbon efficiency because 33-50% of carbon from the sugar substrate is lost as CO2 due to decarboxylation reactions in C. acetobutylicum’s central carbon metabolism. Gas fermentation, using acetogenic organisms such as Clostridium autoethanogenum or Clostridium ljungdahlii, can effectively assimilate gaseous carbon in the form of CO, but these organisms grow much more poorly on CO2 compared to CO, and, when grown on CO2, they make primarily acetate instead of more reduced products like alcohols. Our group has recently shown that cocultures between C. acetobutylicum and C. ljungdahlii have the potential to overcome many of the disadvantages associated with using only one of the two organisms in isolation. These C. acetobutylicum-C. ljungdahlii cocultures show improved carbon efficiency, large substrate and product portfolios, and unprecedented prokaryotic interspecies cell fusion events which involve large scale exchange of protein, RNA, and DNA. ☐ Building on and extending this work, this thesis describes an RNAseq study designed to uncover genes in both C. acetobutylicum and C. ljungdahlii which are important to the unique coculture phenotype, especially the interspecies cell fusion events. This study utilized a (to our knowledge) novel approach to test differential gene expression in response to direct microbial interspecies contact by isolating RNA from C. acetobutylicum and C. ljungdahlii grown in a transwell system, separated physically by a permeable membrane, and comparing it with RNA extracted from a normal mixed coculture. Using this methodology, we identified several genes from a putative Type VII secretion system operon in C. acetobutylicum which are upregulated by direct contact with C. ljungdahlii and which may be “fusogen” proteins involved in interspecies cell fusion. The gene expression data also revealed major differential regulation of amino acid metabolism (most especially of arginine, histidine, and tryptophan) in both C. acetobutylicum and C. ljungdahlii in coculture, which, when combined with amino acid secretion kinetics, helped to identify transfer of amino acids from C. acetobutylicum to C. ljungdahlii as a previously unknown layer of syntrophic cross-feeding between the coculture partners. Using this gene expression data, we also reconstructed a (to our knowledge) novel histidine catabolism pathway in C. ljungdahlii which substantially increases the energy efficiency of C. ljungdahlii growth on CO2 and demonstrated that C. ljungdahlii monocultures grown on CO2 with supplemental histidine grow much faster and to higher cell densities than controls grown only on CO2. ☐ Next, this thesis describes how the coculture between C. acetobutylicum and C. ljungdahlii can be repurposed for carbon-negative production of isopropanol from glucose (in which all of the glucose carbon is assimilated to soluble products along with some external CO2). This study presents detailed analysis showing how, due to interspecies electron exchange, the presence of the acetogen, C. ljungdahlii, enables C. acetobutylicum to synthesize much higher yields of acetone (which is then converted to isopropanol by C. ljungdahlii) in the coculture than would be possible in a C. acetobutylicum monoculture. Using high density, small scale pseudo perfusion experiments, we show that higher cell densities (and thus tighter interspecies proximity) strengthen this electron exchange to enable enhanced acetone and isopropanol yields. Finally, we demonstrate how, using a perfusion bioreactor, prolonged high density cocultures of C. acetobutylicum and C. ljungdahlii can produce isopropanol as the sole alcohol product from carbon-negative fermentation for over 100 hours with strong productivity. ☐ Next, this thesis describes how the coculture between C. acetobutylicum and C. ljungdahlii can be converted from obligate commensalism (C. ljungdahlii requires C. acetobutylicum for carbon in the form of CO2, but C. acetobutylicum does not require C. ljungdahlii for growth) to obligate mutualism (C. acetobutylicum requires C. ljungdahlii for nitrogen). This was achieved by designing a minimal medium with nitrate as the sole nitrogen source. C. acetobutylicum cannot use nitrate, but C. ljungdahlii can use nitrate and, when it has more than it needs, converts the excess nitrate to ammonia (a nitrogen source C. acetobutylicum can use) and secretes it into the culture medium. Based on this strategy, we test and demonstrate how varying the nitrogen source ratio in batch cultures and varying the nitrate feed rate in fed-batch cultures can be used to maintain a stable species ratio in the coculture, increase carbon efficiency, and improve yields of isopropanol and butanol. ☐ Finally, this thesis addresses an important open question in the gas fermentation literature: why do acetogens like Clostridium ljungdahlii show such a strong preference for growing on CO compared to CO2? We show that, though the presence of high energy substrates, such as fructose, can produce some form of catabolite repression, the true limitation on CO2 fixation by C. ljungdahlii (and similar acetogens) is the extremely low solubility of H2 (the electron donor for CO2 fixation in C. ljungdahlii) relative to CO2 and CO. By alleviating the H2 mass transfer limitation with increased mixing (via roller bottles) and high H2 partial pressure, we demonstrate (by far) the fastest doubling time ever recorded for C. ljungdahlii (or similar acetogens, to our knowlege) growing on only CO2 and H2, a doubling time equivalent to the fastest doubling time ever recorded by C. ljungdahlii (or similar acetogen) using CO. We discuss the significance of these findings for the future of gas fermentation and describe how coculturing acetogens with solventogenic organisms, such as C. acetobutylicum, can help to overcome H2 insolubility and potentially enable scalable and economically competitive CO2-negative fermentation.
A multi-approach to the removal and analysis of micro-/nano- plastics (MNPs)
(University of Delaware, 2025) Qiu, Siyu
The wide occurrence of micro- and nano plastics (MNPs) in food, water, soil and air has caused increasing concerns on potential human health impacts and environmental pollution. Accurate identification and efficient elimination of these particles continue to pose difficulties owing to their diminutive size, chemical heterogeneity, and interactions within intricate matrices. This work explored a multi-modal methodology for the detection and removal of MNPs in liquids. A plastic-free, chemically densified carbon nanotube (DCN) membrane was assessed for its effectiveness in filtering MNPs and foodborne pathogens, using the fluorescent polystyrene (PS) particles and Escherichia coli and Listeria innocua as model targets in the studies. Additionally, our collaboration with Dr. Ruogu Tang and Dr. Juzhong Tan explored the use of biochar for removing MNPs. Both DCN and biochar filters exhibited excellent removal of MNPs, with biochar displaying a higher efficacy on removing bigger aggregated particles in HPLC water. DCN membranes consistently demonstrated superior removal of MNPs, with 0.95 and 2.1 µm particles typically decreased by over 90%. Biochar filters exhibited comparable performance for submicron particles (0.1 and 0.5 µm) and demonstrated a pronounced effectiveness for bigger MNPs, with removal rates for aggregated MNPs (2 µm) often surpassing 85–90%. Test solutions, comprising HPLC water, bottled water, and apple juice, were added with model MNPs and treated in vitro digestion, simulating stomach and intestinal conditions. MNPs in the solutions before and after digestion were evaluated by different methods. ☐ The quantitative and qualitative evaluation of MNPs was performed utilizing Nile Red fluorescence staining with fluorescence imaging or reading, Dynamic Light Scattering (DLS), Nanoparticle Tracking Analysis (NTA), and Scanning Electron Microscopy (SEM). In this phase of the study, commercial 0.5 and 2 µm MNPs (fluorescently labeled and non-labeled), alongside with various pretreatment methodologies (in vitro or chemical digestion, matrix-specific background correction rinsing, and centrifugation were added into HPLC water, bottled water and apple juice, and then quantified by the various methods. Nile Red staining of non-labeled PS particles generated fluorescence signals that were dependent on MNPs concentration and size, facilitating the creation of calibration curves for both 0.5 and 2 µm MNPs. The bigger MNPs often exhibited higher apparent recoveries owing to enhanced dye adsorption and aggregation. The fluorescence responses of labeled MNPs aligned with these trends and served as an internal validation of staining efficacy, especially in intricate matrices where apple juice displayed significant autofluorescence, necessitating pretreatment (e.g., dilution and/or centrifugation) to enhance the signal-to-background ratio. In all matrices, chemical digestion and filtration resulted in significant reductions in fluorescence intensity for both size categories, consistent with DLS, NTA, and SEM findings on particle removal or aggregation. The integrated Nile Red-based method is effective for monitoring both non-labeled MNPs in the test solutions although more research is still needed for improvement of recovery rate and detection sensitivity. These findings will be useful for creating a cohesive analytical and filtering platform to determine MNP in diverse food systems. Furthermore, the integration of improved characterization techniques with novel filtration materials such as DCN and biochar presents interesting approaches for future detection and mitigation tactics in evaluation of dietary MNPs and assistance in advancement of advance the regulatory policies on health impacts of MNPs.
Decentralized agents learning with generative communication models
(University of Delaware, 2025) Corder, Kevin
Reinforcement learning (RL) in multi-agent systems is an important and quickly growing domain with diverse applications. Single agent RL methods have been shown effective in low- and high-dimensional state spaces, such as game playing, robotics, and complex optimization problems requiring exploration. Multi-agent RL (MARL) has several additional challenges, including multi-agent credit assignment, the curse of dimensionality, non-stationary learning dynamics, and partial observability when each agent receives private observations. Despite these difficulties, many applications inherently benefit from multiple coordinated agents. ☐ Decentralized MARL offers a scalable and practical approach to coordination, where each agent executes its own policy conditioned solely on local observations and received communication from other agents. Effective decentralized learning typically depends on consistent and reliable communication. However, in real-world scenarios such as robotic teams operating in remote environments, communication channels are often sparse, unreliable, or bandwidth-constrained. ☐ This thesis explores methods that enable decentralized agents to learn effectively under limited communication conditions. Specifically, I extend a class of decentralized MARL algorithms that utilize centralized training with decentralized execution to impute missing communication, which enables continued learning in the decentralized phase. This is accomplished by equipping agents with generative models of joint observations or learned message encoders from teammates. A novel selective sampling approach is introduced that explicitly balances message transmission against model-based inference via a new counterfactual metric called the communication advantage. This value is proven to linearly approximate the associated global advantage, with experimental results demonstrating its efficacy in reducing communication overhead without sacrificing task performance compared to centralized baselines. Additionally, a comprehensive study and empirical analysis of centralization techniques is conducted, clarifying their effects across popular off-policy MARL algorithms and environments. Ultimately, this research provides practical methods and insights to improve MARL scalability and applicability in resource-constrained environments.

Browse

Recent Submissions