Dissertations

This dissertation develops a normative theory of victimhood that reclaims its politicalpotential while addressing its ethical complexity. Challenging dominant views that portray victimhood as politically disempowering or destructive, I argue that, when engaged responsibly, victimhood can serve as a valuable political resource in resisting oppression. Grounded in the acknowledgment of universal human vulnerability, I conceptualize victimhood as a dynamic recognition of the unequal distribution of vulnerability experiences, rather than as a fixed identity. This recognition can enhance victims’ political agency and enable affective communication, thereby contributing to collective resistance. Building on this account, I propose a normative framework for evaluating how victims engage their suffering and victim status. Rejecting both non-evaluative approaches and frameworks that focus solely on the imperative to avoid political uses of victimhood, I argue that victims may bear political responsibilities to engage their victimhood in ways that align with resistance to their oppression. To support this claim, I develop three model evaluative tests, accompanied by qualifying conditions that account for differences in types of harm, epistemic capacity, and social position within power structures. By reorienting ethical discourse on victimhood toward the political goal of undermining injustice, this dissertation offers a politically intentional ethics that guides, supports, and carefully assesses victims’ efforts to mobilize their suffering for transformative political action.

Postpartum sexual well-being is a complex, multidimensional concept yet most research and clinical care continue to emphasize biomedical and behavioral outcomes. This dissertation aimed to (1) develop an integrated, person-centered model of postpartum sexual well-being and (2) develop and evaluate a measure of internal, cognitive-affective factors that influence whether postpartum individuals can engage in sexual experiences aligned with their values and priorities.The study was conducted in three phases. Phase I used qualitative interviews (N=12) to refine a multidimensional conceptual model of postpartum sexual well-being. The final model identified important factors across four domains: cognitive-affective, interpersonal, cultural, and structural/system-level. Within the cognitive-affective domain, five key factors emerged: sexual self-image, self-determination, resilience, comfort, and acceptance. This work laid the foundation for Phase II, which focused on developing a new self-report measure targeting the cognitive-affective domain. Item development and cognitive interviews (N=13) were used to iteratively refine content based on participant and expert feedback. In Phase III, the 34-item Postpartum Sexual Well-Being Inventory: Cognitive-Affective Domain (PSWI-CA) was administered to a diverse sample of postpartum individuals (N=46). Exploratory factor analyses supported unidimensionality for most subscales, with items trimmed based on psychometric performance. Trimmed subscales demonstrated acceptable internal consistency and significant associations with related constructs, including sexual satisfaction, distress, psychological functioning, and alignment between sexual experiences and personal values. Findings support the PSWI-CA as a promising tool for capturing postpartum individuals’ internal experiences of sexual well-being, with future work needed to conduct continued modification and testing of the PSWI-CA items, and extend measurement across all domains of the model.

The work in this thesis presents the use of a florescent visible light collector (FVLC) receiver for implementation in underwater environments as part of a novel underwater optical communication (UWOC) system.

Specific methodologies for analysis are defined in a comprehensive model. Experimental verification in-air is presented and shows good agreement with the values predicted by the model. Further calculations are presented to show the potential of the current FVLC UWOC system in water and demonstrates the FVLC UWOC system has great advantage over fielded devices. The consumer-off-the-shelf system as designed here, totally upoptimized, show a separation distance of up to 7m which is comparable to fully optimized, fielded systems. Greater optimization offers further improvements.

Improvements, optimizations, and differing implementation schemes are identified and explored, showcasing the versatility of the FVLC system. Whether the mission requirements call for low data-rate, large separation distances to high data-rate with small separation distance, the modular approach shows the FVLC UWOC system and proposed model well define the behavior of the system deployed in the operational environment combined in a highly customizable platform.

Predicting food web dynamics under global climate change is a pressing goal in ecology. Rapid trait changes, whether through rapid evolution or phenotypic plasticity, significantly influence contemporary food web dynamics through altering species interactions and subsequent trophic cascades. Therefore, understanding how rapid trait changes in interacting species affect population dynamics and food web structure is crucial for predicting how food webs and ecosystems will respond to future climates. My dissertation addresses the overarching question of how species traits rapidly respond to biotic and abiotic changes and how these trait responses feed back through eco-phenotypic and eco-evolutionary processes across trophic levels, ultimately shaping microbial food webs.

In Chapter 2, I experimentally manipulated temperature and nutrient levels in microbial food webs to investigate how warming and eutrophication jointly influence ecological and phenotypic dynamics. Using time-series analysis, I found that the joint—and often interactive—effects of temperature and nutrients were strongest at higher trophic levels. Our results uncover how feedbacks between ecological and phenotypic dynamics mediate food web responses to environmental change. These results suggest important but previously unknown ways that temperature and nutrients might jointly control the rapid eco-phenotypic feedbacks that determine food web dynamics in a changing world.

In Chapter 3, using experimental evolution and mathematical modeling in microbial food webs of prey algae and ciliate predators, I tested 1) how temperature affects prey evolution; and 2) how the context of the food web—i.e., predator identity, abundance, and competition between predators—mediates prey evolutionary dynamics. Our findings reveal that evolutionary outcomes under warming are shaped by the broader food web context of species, suggesting that the same species may exhibit different eco-evolutionary responses in different food webs under novel climates.

In Chapter 4, I used experimental microcosms and mathematical modeling to ask 1) how prey rapid defensive responses affect predator-prey dynamics when predators also undergo trait shifts, and 2) how prey rapid defenses affect predator biomass. The observed trait-mediated trophic transfer efficiency and reversal of predator-prey cycles highlighted the key role of rapid trait changes in predicting food web and ecosystem functions.

Together, my dissertation shows that climate change-driven environmental stressors change species population dynamics and species interactions. Moreover, species traits respond rapidly to changes in environmental conditions and species interactions, and in turn mediate those interactions. These rapid trait responses can fundamentally change population dynamics, predator-prey cycles, and trophic transfer efficiency, ultimately shaping food web responses in a warming world.

The following dissertation presents three empirical papers and one qualitative conceptual framework related to the design and implementation of Sustainable Development (SD) programs. This thesis has the following structure: Chapters 1 & 2 focus on the topic that brought me to Duke, artisanal and small-scale gold mining (ASGM). Chapter 1 explores a case study of an international intervention focused on promoting mercury-free technology to ASGM miners in Guyana. Chapter 2 presents a second study on ASGM interventions, this time directed toward women waste-rock collectors (jancheras) in Ecuador. The intervention in Chapter 2 focuses on the promotion of a different solution to mercury-use, ore-selling, in which miners sell raw ore directly to processing plants instead of processing it themselves with mercury. In Chapter 3, I transition from studying adoption of alternatives to mercury in gold mining to discuss the promotion of trees on private agricultural land in Southern India for public good provision (exploring how farmers’ preferences match with intended biodiversity ecosystem service provision improvements). Finally, in Chapter 4, I discuss how projects like the ones I studied in this thesis could collect higher-quality evidence to progressively improve design and impact. Then, I briefly conclude by revisiting the commonalities across the chapters and discuss extensions.

The chapters of this thesis are unified by theme, sector, and methodology. Thematically, I study programs with SD objectives – with at least one human and one environmental intended outcome. In terms of sector, I focus on activities involved in land use and land cover change, with chapters focused on gold mining and reforestation. In each of the chapters, the individuals who would ultimately be responsible for changing production practices, the “target population”, are individuals or small groups of individuals working in rural areas of Low to Middle-Income Countries (LMICs) in the Global South. Methodologically, I employ quantitative social science research techniques, and, in two of the three chapters, I use discrete choice experiments (DCEs) to elicit preferences of program participants to inform program design.

Another common theme across the empirical chapters (1-3) is the consideration of how SD solutions must be designed in light of heterogeneity of the target population and target context where the intervention takes place. For instance, the differences between the solutions promoted in the ASGM Chapters (1 & 2) relate to geology -- primary deposits in Ecuador and secondary deposits in Guyana. We point out in these chapters that geology and other heterogeneities (i.e. related to miner type) are relevant for design of effective Sustainable Development Goal (SDG) solutions for ASGM. In Chapter 3, we explore farmers’ heterogeneous preferences both for adopting tree-planting generally and for adopting certain types of tree-planting designs (defined by tree type and planting configurations). Based on our findings about heterogeneity, we make recommendations for public policy design.

The main findings of my studies show that target populations are often not well identified and characterized before SD intervention implementation begins. SD interventions may be more effective in meeting their goals if they first clearly define target populations, then study the preferences of target populations, and finally use those preferences, along with expert technical guidance, to devise solutions that are “adoptable” and “impactful”. Ideally, well-design SD solutions should raise the well-being of program participants in order to make system changes permanent while also generating the public goods that justify SD funding. The studies in this thesis highlight that, even within a single sector and across relatively geographically small contexts, program participants are diverse. As a result, to ensure that programs scale equitably, programs should consider who can and cannot participate in which designs, and what complementary support may be necessary to support rural producers in the Global South to transition to more efficient, more sustainable practices.

Anharmonicity of many material behaviors including thermal transport properties, ferroelectricity, multiferroicity, superconductivity or soft-mode driven structural phase transitions arise from the couplings between microscopic degrees-of-freedom, such as phonons,spins and electrons. Understanding these microscopic processes in the transport and conversion of energy from atomic scale is of vital importance for development and improvement of next-generation materials for energy sustainability. While harmonic phonon picture provides insight into the basic understanding and modeling of atomic vibration, the anharmonicity is indeed essential to explain material behaviors including thermal expansion and structural phase transitions that cannot be thoroughly described by harmonic phonon theory. Therefore, it is important to better understand the anharmonicity of phonons and their coupling to other degrees-of-freedom in order to realize the controllable design and property improvement of energy materials characterized as halide perovskites. Recent progress has been made both experimentally and theoretically to probe and rationalize the microscopic processes that determine the phonon scattering rates and phonon lifetime of different solid materials. In particular, inelastic neutron scattering (INS) and inelastic x-ray scattering (IXS) provide direct measurements of the phonon dispersions, scattering rate and dynamical structure factor S(Q,E). Developments in first-principles simulations have also enabled the computation of phonon scattering processes and provide detailed accounts of anharmonic scattering channels that is crucial in understanding material properties. Combining these experimental and theoretical methods, the goal of this research is to systematically explore the influence of the anharmonic phonons to solid state material properties related to energy transport. As a prototypical system with strong temperature-dependent anharmonicity, elemental bismuth has long served as a platform for exploring phonon-phonon and electron-phonon interactions. Prior studies have demonstrated strong softening of interatomic potentials in photoexcited Bi, raising interest in how intrinsic anharmonic phonon scattering evolves with temperature. In this work, we use INS and first-principles simulations to probe thetemperature dependence of acoustic phonon energies and lifetimes in Bi, shedding light on the microscopic mechanisms governing its lattice thermal conductivity. In parallel, we explore the anharmonic lattice dynamics in inorganic halide perovskites, including CsMX3 (M = Ge, Sn, Pb) and the double perovskite Cs2AgBiBr6, which have shown excellent performance in photovoltaic and radiation detector applications. These materials are characterized by soft lattices, low thermal conductivities, and large thermal expansion—features that arise from strong phonon anharmonicity. Phase transitions in these systems are typically associated with octahedral tilts and rotations driven by zoneboundary soft modes. Recent studies suggest that such structural fluctuations can couple to the electronic structure, facilitating exciton dissociation and enhancing carrier lifetimes. Using INS, X-ray diffuse scattering, and molecular dynamics simulations, we identify 2D overdamped diffuse rods in the high-temperature phases of CsSnBr 3 and CsPbBr3, highlighting the universal nature of these dynamic instabilities across halide perovskites. Extending this investigation to hybrid and double perovskite systems, we focus on methylammonium lead iodide (MAPbI3, or MAPI) and the inorganic double perovskite Cs2AgBiBr6, both of which exhibit soft lattice dynamics and intriguing structural phase behaviors. In Cs2AgBiBr6, we identify an incommensurately modulated ground state that emerges upon cooling from the tetragonal phase, as revealed by single-crystal neutron diffraction. At higher temperatures, quasi-elastic diffuse rods associated with anharmonic octahedral rotations are observed in both the cubic and tetragonal phases, similar to those found in other halide perovskites. In MAPI, dynamic structural fluctuations are also evident in the form of diffuse rods in the cubic phase, which vanish abruptly below the cubic-totetragonal transition. These measurements were made possible by combining inelastic neutron scattering, diffuse X-ray scattering, polarized neutron techniques, and first-principles simulations. Together, our results reveal distinct temperature-dependent lattice dynamics in both systems and offer new insight into the role of phonon anharmonicity and structural fluctuations in shaping the electronic and thermal behavior of halide perovskites.

The evolution of adipocytes provides mammals with a dedicated compartment for efficient energy storage in the form of triglyceride. Fat cells possess the biochemical machinery to convert glucose and free fatty acids (FFAs) into triglyceride (lipogenesis) during times of nutrient excess. When energy demand is high, triglycerides can be catabolized to FFAs (lipolysis). FFAs released into circulation can be used to provide fuel to cells in times of nutrient deprivation. FFAs within adipocytes can be further oxidized to generate heat through the process of adaptive thermogenesis. These anabolic and catabolic programs are highly integrated and coordinated to match nutrient availability with energy demand. Importantly, an imbalance in these programs leads to dysregulated lipid metabolism and metabolic disease. Our prior studies identified the nucleosome remodeling and deacetylase (NuRD) corepressor complex as a binding partner of key transcriptional regulators controlling the thermogenic gene program of adipocytes. The physiological importance of this complex in fat cells remains unknown. To address this, we have generated a genetic Tet-On mouse model that allows for inducible adipocyte-specific inactivation of Chd4 (Chd4iAKO mice) in the presence of doxycycline. Loss of CHD4, a primary catalytic component of the NuRD complex, leads to loss of NuRD complex activity. I find that the loss of the adipocyte CHD4-dependent NuRD complex leads to a striking upregulation of key genes driving both anabolic and catabolic pathways of lipid metabolism, including genes that might elicit futile energy cycling. Explanted tissues from chow-fed Chd4iAKO mice exhibit heightened basal and stimulated lipolysis. Acute cold exposure or pharmacological β3-adrenergic receptor agonism triggers substantial inguinal WAT inflammation, presumably due to hyper-stimulated lipolysis. Notably, inducible deletion of Chd4 in adipocytes has strong effects on body weight. Chd4 inactivation at the onset of high fat diet feeding (HFD) leads to increased energy expenditure and confers resistance to diet-induced obesity. Inactivation of Chd4 in mice with obesity (after 10+ weeks of HFD feeding) leads to gradual weight loss. However, in both cases, Chd4iAKO mice exhibit insulin resistance that is associated with expanded liver mass and ectopic lipid deposition. Together, our data highlight the importance of chromatin regulation in fine-tuning the balance between anabolic and catabolic programs in adipocytes. Moreover, our data further underscore the detrimental metabolic consequences of lipolysis, despite potential benefits to body weight and adiposity.

The ionosphere is a region of the upper atmosphere, ranging from about 60 km to 1000 km above ground, consisting of free electrons and ions primarily due to solar radiation. Because of its charged nature, it is essential for long-distance radio communication, navigation, and space weather monitoring. The ionosphere changes constantly, and its variability can significantly affect radio wave propagation. As a result, there is ongoing interest in developing tools for observing ionospheric disturbances. Radio signals at frequencies of 30 MHz and below are reflected by the lower ionosphere and are commonly used for ionospheric remote sensing through both active and passive techniques. However, the medium-frequency (MF, 0.3-3 MHz) band remains relatively underutilized. In particular, commercial AM (amplitude modulation) radio transmitters operating between 530 and 1700 kHz offer several advantages, despite not being designed for scientific purposes. These transmitters are widely distributed across geographic regions and continuously broadcast radio waves, making them valuable sources for passive ionospheric remote sensing with broad spatial coverage.

The aim of this dissertation is to explore the use of AM radio signals for detecting ionospheric disturbances during both nighttime and daytime, using a low-cost software-defined radio (SDR)-based receiver developed specifically for this work. The receiver supports dual-channel data acquisition with time synchronization across multiple receivers. Real-time monitoring of ionospheric conditions is enabled through a developed real-time data processing program, which also reduces data storage requirements and makes long-term measurements more feasible.

To estimate the virtual reflection height of nighttime skywaves, a new approach is proposed that computes the audio delay between groundwave and skywave signals using synchronized recordings from distributed receivers. This method relies solely on AM radio signals and does not require pulsed transmissions or ionospheric modeling tools. It is capable of tracking virtual height variations with time resolution on the order of seconds and distinguishing between reflections from the E and F layers, providing a foundation for detecting more detailed variations within different ionospheric layers during nighttime.

For nighttime measurements, we observed numerous Doppler shifts associated with gravity waves propagating in the ionosphere. The virtual height estimation method helped identify distinct gravity wave characteristics in different layers, with slower, larger variations in the F layer and faster, smaller variations in the E layer. Traveling ionospheric disturbances (TIDs) were detected and quantified through phase analysis using data acquired by three distributed remote receivers. Spread features in both the E and F layers were also detected, showing associations with gravity waves that are consistent with previous findings using HF Doppler sounding. Measurements during solar terminators revealed clear, layer-dependent responses to sunrise and sunset transitions.

Finally, the nighttime measurement techniques were adapted for daytime experiments, which present greater challenges primarily due to strong absorption from the D layer. By comparing received signal amplitudes with predicted groundwave amplitudes from an ITU-R (International Telecommunication Union Radiocommunication Sector) model, we confirmed that skywave components become increasingly dominant beyond 500 km and can be reliably detected beyond 1000 km. Daytime measurements revealed a variety of ionospheric disturbances, including MF signal blackout effects with the presence of increased solar activity, daytime gravity waves, and temporary reductions in plasma density associated with the 2024 total solar eclipse.

Adult metabolic disease and cancer are multifaceted and regulated by a host of diverse causes which can be immediate or long-term. The medical concept of the Developmental Origins of Health and Disease (DOHaD) has been established to characterize and investigate how conditions during gestation and development can predispose or directly cause diseases later in life. These can include modifications to the epigenome and changes in metabolic signaling. Research in mammals and human cohorts is limited by the availability of viable case-studies, complex ethical and social influences, and challenges associated with tracking individuals with dynamic and longitudinal development. Therefore, establishing a viable model system to investigate the interaction between physiology, genetics, and the environment is pivotal to understanding the developmental origins of adult disease.

Here, we implement the nematode Caenorhabditis elegans to study how extended starvation and developmental arrest affects the formation of developmental abnormalities and tumors in the reproductive system. We leverage the ability of C. elegans to arrest growth after hatching in response to the absence of food as a model for adverse developmental conditions. It has been previously demonstrated that extended early larval arrest increases the incidence of tumors in recovered adults, however the full scope of their regulation and proximal causes is not known.

We show that insulin-like/insulin-like growth factor signaling (IIS), a critical regulator of starvation resistance and recovery, is subject to dynamic regulation via ubiquitination. Specific ubiquitination sites and ubiquitin ligase genes function in a context-dependent manner, altering phenotypes at different life stages and environmental conditions. We also show that starvation-induced gonad abnormalities are regulated independently of IIS and its downstream effectors via IIS-independent activity of DAF-18/PTEN through the pocket protein LIN-35/Rb. We implicate C. elegans Hh-related signaling downstream of lin-35 and demonstrate that genes in this pathway transcriptionally regulate components of the innate immune system following extended larval arrest. These innate immunity genes are crucial to the regulation of starvation-induced

abnormalities and pathogen resistance. We also extend our previous analysis of the Wnt and lipid synthesis pathways, previously shown to act downstream of IIS to regulate the formation of gonad abnormalities and characterize their genetic interactions with Hh-related signaling. Additionally, we investigate the role of the C. elegans TOR pathway in the regulation of starvation-induced abnormalities and characterize its interactions with other relevant pathways. Overall, we expand previous work analyzing the regulation of complex developmental phenotypes following early-life stress and show that a potential fitness trade-off may exist in the induction of innate immunity and pathogen resistance at the cost of reduced developmental fidelity.

The aging population in the United States (U.S.) has led to substantial growth in the number of people living with dementia (PWD). Dementia is associated with increased healthcare utilization and expenditures, as well as decreased quality of life (QOL)—a multidimensional construct encompassing physical, psychological, and social domains. While many PWD live in the community and prefer to age in their own homes, a high level of care is often needed as dementia progresses, typically provided in institutional care settings. In the past few decades, there has been substantial growth in alternatives to nursing homes, including residential care facilities and home and community-based services (HCBS) programs. Current research provides limited understanding of how QOL varies across different living and care arrangements for PWD, or how these care arrangements impact QOL outcomes. This information is essential for identifying settings that best support the well-being of PWD and for informing policies and interventions aimed at improving their QOL. This dissertation addressed these gaps in the literature through three aims focusing on older adults (aged 65 and older) with cognitive impairment. Aim 1 conducted a systematic review and meta-analysis to quantitatively synthesize evidence on how QOL differs across three living situations for older PWD living at home, in nursing homes, and residential care settings. Aim 2 described 12-year trends in five QOL indicators for older adults with cognitive impairment, both overall and by living and care arrangements. Aim 3 assessed the impact of adding formal home care to existing informal care on depressive symptoms and self-rated health among older Medicaid beneficiaries with cognitive and functional impairments. To address potential endogeneity in care decisions, Aim 3 employed an instrumental variable (IV) approach using state-level HCBS generosity as an instrument for formal home care use. Aims 2 and 3 used population-based, repeated cross-sectional data from the Health and Retirement Study (HRS). Aim 1 included thirty-one studies with 13,064 PWD that met the inclusion criteria. No significant differences in global QOL were observed across living settings. However, moderator analyses by country showed significantly higher global QOL in residential care and home-based settings than in nursing homes in the United States and Spain. PWD in residential care settings showed significantly better QOL than those in nursing homes in the meaningful activities, emotion, and social relations domains. No significant differences in domain-specific QOL were observed between home-based and nursing home settings. Aim 2 results showed that trends in five QOL indicators were stable over time, except for a slight upward trend in the prevalence of a high level of life satisfaction, primarily driven by community-dwelling PWD with met home care needs. Trends in QOL indicators also varied substantially based on living and care arrangements. Community-dwelling PWD without functional limitations consistently reported better outcomes across all five QOL indicators, followed by those with either met or unmet home care needs, wherea nursing home residents consistently had the poorest outcomes. Compared to community-dwelling PWD with met home care needs, those with unmet needs experienced worse outcomes in terms of loneliness and life satisfaction. Aim 3 found no significant effects of combined formal and informal care on depressive symptoms and self-rated health, compared to informal care alone, in both the naïve Probit and 2SRI models. These findings remained consistent across several sensitivity analyses, including the use of broader definitions of formal care and cognitive impairment, as well as varying timing of the outcomes. In conclusion, this dissertation found that global and domain-specific QOL and QOL indicators varied substantially based on living and care arrangements among PWD. Adding formal home care to existing informal care did not result in significant improvements in mental and overall health among older Medicaid beneficiaries with cognitive and functional impairments. These findings highlight the need for psychosocial interventions that address multiple dimensions of QOL for PWD, especially for community-dwelling individuals with unmet care needs. Additionally, there is a pressing need to improve the diagnosis and treatment of mental health issues among PWD with functional limitations in the community. Integrating mental health services into formal home care may better address the complex emotional and psychological needs in this high-need population. High-quality evidence using rigorous causal inference methods is needed to examine the impact of different living settings on QOL.

This dissertation examines rebellion as a structural and recurrent feature of the early modern British Empire. It argues that colonial uprisings from the 1670s to the 1690s were not ruptures from imperial allegiance but political acts embedded in a contested grammar of English subjecthood. In an empire defined by overlapping jurisdictions, legal pluralism, and improvised authority, subjects repeatedly invoked royal obligation and protection to contest local misrule. Rebellion became a mechanism through which imperial actors—settlers, soldiers, officials—asserted their place within the empire, challenged the legitimacy of devolved governance, and negotiated their political identity. Drawing on case studies from Virginia, Carolina, Maryland, St. Helena, and Bombay, the dissertation shows how rebels framed their resistance in the idioms of loyalty, law, and subject-rights. These episodes reveal a political culture in which subjecthood was not passively received but actively articulated through petitioning, remonstrance, and revolt. The study integrates legal, political, and imperial history to reconceptualize rebellion not as an interruption of order, but as a visible expression of its internal contradictions.

Parental participation in the care of hospitalized children is an important element of the parental role and is associated with positive child and parent outcomes including parenting confidence. How parental participation is defined, how parents describe their participation, and the influences of social determinants of health and social interactions in the healthcare setting on parental participation are not well understood. The purpose of this dissertation is to describe the relationships among social determinants of health, healthcare social interactions, parental participation, and parenting confidence among parents of children diagnosed with cardiac conditions requiring hospital care utilizing various research methods including conceptual delineation, quantitative study, and qualitative study using primary and secondary analysis. The findings suggest that social determinants of health influence parental participation in their child’s hospital care via parenting confidence. Additionally, findings illuminate dynamic relationships between parental participation and confidence influenced by healthcare social interactions. This research emphasizes the need for empathetic care from providers to support parents’ confidence and enable parental participation in children hospitalized for cardiac care. Future research should examine the instrumental and relational support parents need to enhance their parenting confidence, and the resources providers need to offer such support.

Small cell lung cancer (SCLC) and olfactory neuroblastoma (ONB) are aggressive malignancies with neuroendocrine (NE) features, poor outcomes, and limited therapeutic options. Despite being clinically treated as a uniform entity, SCLC exhibits remarkable molecular and phenotypic heterogeneity, including distinct subtypes defined by the lineage-related transcription factors ASCL1, NEUROD1, POU2F3, and controversially— YAP1 (SCLC-A, -N, -P, and -Y). Similarly, ONB—a rare tumor arising in the olfactory epithelium—displays histologic diversity and variable NE differentiation, yet its origins and classification remain poorly defined. In both SCLC and ONB, there are major gaps in knowledge surrounding the cellular and molecular determinants of phenotypic heterogeneity. Clarifying whether SCLC and ONB phenotypic diversity stems from genetic drivers, distinct cells of origin, dynamic plasticity, or a combination of these factors is essential for advancing classification frameworks and identifying more effective, targeted treatments for these diseases.To address these knowledge gaps, we leveraged insights from normal development and tissue repair alongside time-series single-cell transcriptomics and lineage tracing, genetically-engineered mouse models (GEMMs) and organoids, and extensive single-cell and bulk analyses of primary human tumors. We began investigating SCLC subtype diversity from a NE cell of origin—the widely accepted origin for SCLC. We developed a powerful ex vivo platform to isolate early NE-high lesions from MYC- high and -low SCLC GEMMs that lack Rb1 and Trp53, enabling study of SCLC tumor progression in multiple genetic contexts. This, paired with studies in human cell lines, led us to discover that MYC is sufficient to drive SCLC subtype evolution from SCLC-A to - N to -Y states from NE cells of origin. However, regardless of genetic context, we find that NE cells fail to generate the full spectrum of phenotypic diversity observed in human tumors, including SCLC-P, motivating us to explore alternative cells of origin. We next investigated the lung basal cell as an origin for SCLC, given the ability of basal cells to generate all lung cell types, including rare ASCL1/NEUROD1+ NE cells and POU2F3+ tuft cells that mirror SCLC-A, -N, and -P phenotypes, respectively. We create highly-tractable basal-origin models and perform in vivo lineage-tracing under multiple relevant, genetic contexts to find that basal cells can produce the full spectrum of known SCLC phenotypes—including SCLC-A, -N, -P, -Y, and a recently identified ATOH1+ state. Moreover, we discover mixed NE/neuronal, hybrid tuft-ionocyte-progenitor (“TIP”-like), and inflammatory basal-like states in our models that are conserved in human SCLC tumors, but previously unrecognized. Single-cell clonal analyses reveal an immense capacity of SCLC for plasticity and uncover preferred plasticity trajectories in SCLC and putative plasticity drivers. Altogether, our findings establish basal cells as the most likely cell of origin for SCLC—fundamentally reframing our understanding of SCLC heterogeneity and plasticity and offering new avenues for modeling and treating the disease. In parallel, we developed a novel GEMM of ONB using Rb1/Trp53/Myc (RPM) alterations that generates high-grade, NEUROD1+, metastatic ONB. We identify the globose basal cell (GBC)—a progenitor population capable of regenerating neuronal and non-neuronal cell types in the olfactory epithelium—as a permissive and likely cell of origin for ONB. We demonstrate that ASCL1 loss in autochthonous RPM GEMMs and GBC organoids permits emergence of non-neuronal states, including POU2F3+ tuft-like lineages, alongside NEUROD1+ neuronal states. Consistently, analysis of human ONB tumors using scRNA-seq, spatial transcriptomics, and immunofluorescence reveals tumor heterogeneity, including NEUROD1+ neuronal and POU2F3+ microvillar states, reminiscent of normal GBC differentiation trajectories. Thus, highly similar to SCLC, mouse and human ONB tumors arise from a basal-like cell of origin, harbor mutually exclusive NEUROD1 and POU2F3-like states, and exhibit extensive plasticity between neuronal and non-neuronal lineages. Altogether, this body of work, informed by principles of normal development and injury repair, resolves the origins and lineage relationships of SCLC subtypes and reveals molecular parallels between SCLC and ONB that have important therapeutic implications. These results challenge the traditional view of fixed lineage identity in NE tumors and reveal that tumor heterogeneity and plasticity are collectively influenced by cell of origin, genetic alterations, and normal developmental or tissue repair pathways. Insights gleaned from this work establish a new conceptual framework for studying and treating SCLC, ONB, and other similar NE tumors.

Bayesian computational algorithms tend to scale poorly as data size increases, partic- ularly for dependent data such as long time series. This has motivated the development of scalable inference methods, including divide-and-conquer and sub-sampling-based ap- proaches. We study the problem of Bayesian inference for time series, where the literature predominantly focuses on approximate methods that often lack rigorous theoretical guar- antees and may result in poor practical accuracy. To address this, we propose a simple and scalable divide-and-conquer method for long time series, with provable accuracy guarantees.

In addition, we address the computational inefficiency of Markov chain Monte Carlo (MCMC) algorithms for hidden Markov models, which often rely on the forward-backward sampler and become slow with increasing time series length. We develop a targeted sub- sampling (TASS) approach that over-samples observations corresponding to rare latent states when estimating gradients in stochastic gradient MCMC. TASS improves sampling efficiency by reducing variance in gradient estimation, especially when rare states corre- spond to extreme observations. Real and synthetic data demonstrate substantial gains in predictive and inferential accuracy.

We further propose a decision framework for testing multiple hypotheses with a nat- ural tree structure, common in multiscale inference problems. We model dependence in hypothesis probabilities using a hidden Markov tree model (HMTM) and develop an or- acle procedure that minimizes the false non-discovery rate (FNR) under a false discovery rate (FDR) constraint. A data-driven procedure is introduced and shown to be asymptot- ically equivalent to the oracle. Motivated by human brain connectome data, we apply the framework to a one-sided two-sample testing problem for heterogeneous count data, and demonstrate its effectiveness through simulations.

Spina bifida (SB), the most prevalent subtype of neural tube defect, is a complex congenital condition with both genetic and environmental contributions. While recent advances in genomic sequencing have enabled progress in identifying potential risk loci, the genetic architecture underlying SB remains largely unresolved. To address this, I employed a comprehensive analytic strategy leveraging whole-genome and whole-exome sequencing data from case-parent trios and ancestry-matched public controls. My analyses included both single-variant association testing and gene-based mutational burden tests, enabling the detection of genetic risk signals driven by individual variants as well as the cumulative effect of multiple variants within genes.Using this approach, I identified 16 novel candidate genes associated with SB. These genes span diverse biological processes relevant to neural tube closure, including cytoskeletal remodeling, transcriptional regulation, immunity, and metabolism, and the majority have data supporting their expression in the developing human neural tube. Some have also been implicated in other neurodevelopmental and craniofacial disorders, further supporting their potential relevance to SB. Associations were driven by both common and rare variants, including two genes (GLB1L2 and PLA1A) where rare single nucleotide variants (SNVs) were implicated by both ExWAS and gene burden testing. Additionally, three genes (SPIRE2, TVP23B, CHD5) were identified solely through gene-level burden, indicating a cumulative mutational effect in the absence of individually significant variants. To assess whether genetic risk is also detectable in unaffected parents, I performed a parallel burden analysis in the parental cohort compared to controls. Ten genes were significantly associated in parents, four of which overlapped with those found in probands, suggesting the presence of latent genetic risk in the parental genome. The remaining six were uniquely associated with parental samples, though the biological interpretation of these signals remains uncertain and may reflect differences in power or unmeasured genetic modifiers. Together, these findings expand the list of candidate genes for SB, highlight the value of complementary association strategies, and suggest a more complex, polygenic architecture than previously appreciated. This work also underscores the importance of including both common and rare variation in efforts to resolve the missing heritability of neural tube defects.

My dissertation, "How to Save America: The Art and Culture of Hoarding," explores the fascination with junk that emerged in the mid-twentieth century by turning to artists and writers who work with found and collected material. I understand hoarding, whether or not it emerges from a pathology, as a life practice, a repeated, productive behavior of refusing to dispose that someone performs over an extended period of time. It serves as a reminder of the process by which we routinely transform objects into belongings, things into possessions, and matter into memory. I begin by narrating the story of Homer and Langley Collyer, a pair of reclusive brothers who were made famous for living and dying inside a Harlem brownstone they filled with belongings. My examination of the American interest in hoarding at once exposes patterns of consumerism and excess and reveals how a psychiatric pathology became the subject of cultural fascination. While hoarders raise questions about the value of what others would deem to be garbage, artists and writers turn disposables into a material index of consumer culture. Building on the work of a range of studies of hoarding, my project maps the various attempts to pathologize this peculiar behavior and reveals the connections among psychiatry, art, literature, and popular media. In an introduction, four chapters, and a conclusion, I draw on the methods of literary and cultural analysis to explore the phenomenon of hoarding.

The exponential growth in design complexity of very large-scale integrated (VLSI) circuits driven by continuous CMOS technology scaling has created significant challenges for traditional electronic design automation (EDA) methodologies. While machine learning (ML) techniques offer promising solutions to address these limitations, their practical adoption in EDA faces three critical challenges: the complexity and time-consuming nature of ML model development, limited availability of high-quality training data due to confidentiality concerns, and security vulnerabilities of ML models to adversarial attacks.

This dissertation addresses these challenges through six primary contributions:1. AutoML for EDA: An automated machine learning model development framework with application to routability prediction as a case study. 2. Privacy-preserving collaborative learning: A novel federated learning (FL) framework that enables collaborative training among design companies without explicit data sharing, addressing the critical data availability challenge. 3. HFL-LA for lithography hotspot detection: A heterogeneous FL approach with local adaptation (HFL-LA) specifically targets lithography hotspot detection while handling data heterogeneity across different design companies. 4. EDALearn: A comprehensive benchmark suite for evaluating ML-based EDA approaches with an end-to-end flow from synthesis to physical implementation. 5. DRC-guided CURE: A robust defense mechanism that defends ML-based lithography hotspot detectors against adversarial attacks using DRC-guided techniques. 6. CROP: A LLM-driven circuit retrieval and optimization framework for design-aware VLSI flow parameter tuning.

These contributions significantly advance the field of machine learning for EDA by establishing methodologies for automated model development, privacy-preserving collaborative learning, robust model deployment, standardized benchmarking, and intelligent design optimization, thereby paving the way for truly intelligent electronic design automation that can adapt, learn, and operate securely and reliably in industrial environments.

Viral infection are dynamic processes, covering vast spatiotemporal scales within heterogeneous biological environments. Extracellular viral diffusion features high diffusivity (~ 2 μm2/sec) across large axial range. Virus-receptor binding happens in milliseconds on a molecular scale while the whole infection cycle lasts from hours to days in cellular to tissue level. Capturing these events in real time with high spatiotemporal resolution has remained a persistent challenge due to limitations in conventional microscopy techniques. This dissertation presents a suite of innovations in high-speed 3D tracking microscopy and labelling strategies aimed at enabling enriched spatiotemporal mapping of single-virus dynamics in both cellular and complex environments. To overcome the limitations of existing single virus tracking (SVT) methods, we developed 3D tracking and imaging (3D-TrIm) microscopy, an active-feedback single-particle tracking platform (3D-SMART) integrated with simultaneous volumetric live-cell imaging (3D-FASTR). This technique offers unprecedented resolution of the extracellular phase of viral infection, visualizing insightful viral infection events such as membrane skimming, receptor binding, viral trafficking along protrusion, and intercellular trafficking. Furthermore, 3D-TrIm is capable of extending SVT beyond simple monolayer cell culture to more complex tissue-like epithelial systems, enabling comprehensive mapping of viral behavior within more biological relevant environments. Despite the performance of 3D-TrIm, SVT trajectories typically only last for minutes due to photobleaching of fluorescent labels, failing to visualize the entire virus infection cycle. Recognizing the need for photostable and bright fluorescent labels to sustain long-duration high-speed tracking, we incorporated StayGold, a highly photostable green fluorescent protein, to generate virus-like particles (StayGold-VLPs). These newly engineered particles enabled continuous 3D SVT in live cells for over an hour while maintaining the spatial resolution and kilohertz sampling rates, dramatically increasing the photon budget and information content of individual trajectories. This enhanced capability facilitates detailed studies of viral trafficking dynamics over extended timescales and offers the potential to map the complete life cycle of a single virus in the future. Previous studies have demonstrated the capability of 3D-TrIm microscopy to perform single virus tracking with high spatiotemporal resolution, as well as the effectiveness of StayGold-based VLP labelling strategy in extending tracking duration. We combined these two advancements to investigate the viral invasion dynamics of SARS-CoV-2. Long-term and eventful infection trajectories were captured, including free diffusion, viral landing, diffusion on cellular protrusion, internalization and internal trafficking. Among those events, a novel diffusion pattern was observed: viral linear trafficking on plasma membrane, which helps the cell to explore the cellular surface rapidly to localize the optimal entry spot. Further study revealed that this trafficking pattern is actin-dependent and positively associated with the ACE2 expression. Finally, we applied these high-speed 3D tracking methodologies to investigate nanoparticle diffusion in 3D porous environments. Using 3D-SMART microscopy, we resolved nanoparticle dynamics within agarose hydrogels at super resolution (~ 10 nm in XY, ~ 30 nm in Z). These experiments uncovered various transport behaviors, highlighting here ‘hopping diffusion’, where particles intermittently escape confinement pockets, providing insights into hydrogel microstructures. Long, highly sampled trajectories enable extraction of kinetic parameters, confinement structure and size, and thermodynamic barriers, offering new perspectives on nanoparticle diffusion and the structural dynamics of porous materials, with implications for drug delivery, material science, and biological systems.

Breast cancer is a global challenge, accounting for 670,000 recorded deaths in 2022. It is the most prevalent cancer and the cancer with the highest incident rate, with 2.3 million new cases annually. Incidence of breast cancer has increased since 1990 and is projected to continue to rise through 2030. While this is true, great strides have been made in the patient outcomes. Better treatment and widespread mammography screening have contributed to a decrease in breast cancer deaths, with age adjusted mortality rate decreasing from 48 per 100,000 women in 1975 to 27 per 100,000 women in 2019. However regardless of early detection, mortality from recurrent or metastatic disease has not improved in the past 20 years, presenting an outstanding clinical challenge in breast cancer care. Over 90% of breast cancer deaths occur from recurrent disease after treatment of the primary tumor.Breast cancer is a heterogenous disease, resulting in a stratification of prognoses based on molecular subtype. Hormone receptor positive subtypes (progesterone, PR+ and estrogen, ER+) have the best prognoses and are expected to respond well to chemotherapy. Patients positive for enrichment of the human epidermal growth factor 2 (HER2+) or negative for all three receptors (triple negative breast cancer, TNBC) suffer shorter time to recurrence and have greater likelihood of developing visceral metastases than those with ER+ and PR+ disease. However, HER2+ patients have benefitted from the advent of antibody-based therapies that target HER2 on the cell surface. Conversely, as TNBC lacks receptor targets, this aggressive and highly metastatic subtype of breast cancer currently does not benefit from targeted therapies and accumulates resistance to chemotherapy. Thus far, no targeted therapies have been developed that prove efficacious against TNBC. Metabolic heterogeneity is emerging as a promising area of study that can lead to translational benefits in breast cancer survival. Dysregulated cellular energetics was recognized as a hallmark of cancer in 2011 after being an area of study for nearly 100 years. In 1923, Dr. Otto Warburg observed that cancers rely heavily on the uptake of glucose and the glycolytic pathway to support rapid proliferation. This propensity of tumors to rely on aerobic glycolysis as a primary energy source is known as the Warburg effect. The catabolism of glucose and subsequent conversion of pyruvate to lactate provides byproducts that can be used for biosynthesis, paramount to tumor biomass production. An increased production of lactate modifies the tumor microenvironment to be more acidic, advantageous to tumors. Additionally, as tumors rapidly increase in size, angiogenesis lags, creating a tortuous, incomplete, non-uniform vascular network resulting in areas of hypoxia. Hypoxia triggers transcription factor HIF1-alpha to further upregulate glycolysis. Reprogramming of glucose metabolism provides only a partial picture of the complex and dynamic metabolic landscape of tumor cellular energetics. Oxidative phosphorylation (OXPHOS) has been shown to support the energetic needs of tumors alongside aerobic glycolysis, especially in TNBC. OXPHOS has been reported by our group and others to be upregulated in tumor tissue following treatment, through regression and ultimately recurrence. Recently, metabolism of fats has emerged as a key driver of aggressive breast cancer metabolism. Situated in the adipocyte-replete breast environment, tumors uptake fatty acids from nearby lipids, even communicating with adipocytes through gap junctions to stimulate the release of free fatty acids for tumor consumption. Upregulation of the oncogene MYC in TNBC has been shown to significantly upregulate fatty acid β-oxidation (FAO), and the inhibition of FAO in preclinical models of MYC-driven TNBC has been shown to be lethal to the tumor. The metabolic pathways discussed above are intrinsically linked with the tumor microenvironment, with vasculature and tumor-associated adipocytes both playing a role. Molecular heterogeneity as well as heterogeneity in the tumor microenvironment can cause differential response to treatment spatially across a tumor. Sub-populations of cells within a tumor that survive treatment and remain as residual disease threaten to return as recurrent disease. It is likely that just as molecular subtype can provide prognostic information and guide therapeutic strategies that are likely to be effective, understanding heterogeneity in a patient’s specific cancer can predict risk of recurrence and effectiveness of different interventions. Taken together, this body of evidence points to a strong correlation between heterogeneous tumor metabolism and vasculature which drives treatment resistance and patient mortality. OXPHOS, glycolysis, and fatty acid metabolism are key metabolic phenotypes that contribute to the aggression of breast cancer across a tumor’s lifespan. These phenomena have been observed by myriad technologies such as immunohistochemistry, LC-MS and metabolomics, FDG-PET, MR(S)I, optical spectroscopy, and optical imaging. Gaps in our knowledge remain due to a dearth of technology able to longitudinally measure multiple metabolic pathways along with vasculature concurrently in a living animal while preserving spatial information. Thus, the overarching goal of this project was to develop a multi-modal optical toolkit using both fluorescence spectroscopy and microscopy that can measure OXPHOS, glucose uptake, fatty acid uptake, and vascular parameters in a single living animal. This goal was achieved through three specific aims. Aim 1 establishes a technique to measure fatty acid oxidation in vivo and characterizes fatty acid oxidation longitudinally in multiple metastatic preclinical models, and in healthy tissue controls. Previously, the fluorophore Bodipy FL C16 had been shown by our group to be a robust marker of fatty acid uptake, and the fluorophore TMRE had been used as a marker of OXPHOS. Combining a measurement of fatty acid uptake with a measurement of OXPHOS can provide information on whether the fatty acids are being oxidized via fatty acid β-oxidation. Chemical, optical, and biological crosstalk between TMRE and Bodipy FL C16 was assessed, and the two were shown to be compatible. Application of the technique developed in this aim established the utility of multiple metabolic endpoints for discrimination between healthy and tumor-bearing mammary tissue, and between TNBC models with varying metastatic potential. Aim 2 establishes a technique to measure OXPHOS and the uptake of fats and sugars that drive it concurrently in vivo using both fluorescence spectroscopy and fluorescence imaging. Fluorophores Bodipy FL C16 (uptake of fats) and 2-NBDG (uptake of sugars) had not previously been used in combination due to challenges posed by both optical and biological crosstalk. Optical crosstalk between spectrally overlapping fluorophores Bodipy FL C16 and 2-NBDG was mitigated using a spectral unmixing technique developed using spectroscopy and adapted for imaging. The success of this technique was validated for both modalities using tissue-mimicking optical phantoms. To mitigate biological crosstalk, this aim validates a concurrent injection scheme using both spectroscopy and imaging. Leveraging the unique endpoints and datasets captured with each modality, relationships between metabolic and vascular endpoints were explored and the utility of combining metabolic and vascular information exploited. Spectroscopy is a non-invasive technique that enables longitudinal study, and collection of oxygen saturation (SO2) and total hemoglobin concentration ([THB]) improves discrimination between tumor and normal tissue. The addition of spatial information preserved with imaging allows intra-tissue heterogeneity to be investigated. Aim 2 shows increased intra-tissue heterogeneity in tumor tissue compared to normal tissue. Aim 2 also demonstrates the observation of compartments within tumor tissue displaying differing metabolic phenotypes, glycolysis and FAO). The oncogene MYC is known to drive tumor aggression in part by regulating metabolism. Studies have shown that the upregulation of MYC increases both glycolysis and FAO. Inhibition of FAO in MYC-driven tumors lead to tumor regression in multiple preclinical models. Having developed and robustly validated an optical toolkit capable of imaging OXPHOS, glucose uptake, fatty acid uptake, and vasculature concurrently, in Aim 3 we sought to investigate the relationship between the oncogene MYC and metabolism in preclinical TNBC models. We showed increased fatty acid uptake in two MYC-driven models compared to normal tissue. We characterized a metabolic switch from a FAO signature in growing, primary MYC-driven tumors to oxidation of glucose in regressing tumors. We captured inter- and intra-tumoral heterogeneity across tumors, showing different metabolic signatures exist in separate regions of a tumor, contributing to heterogeneity that drives aggressive disease. Finally, we identified correlation between proximity to vasculature and metabolic phenotype across different fields of view in a single tumor. This optical toolkit has translational potential and is poised to enable wide-reaching clinical and pre-clinical studies. Patient-derived organoids (PDOs) have been shown to preserve phenotypes observed in the individual patients’ cancers they are derived from; imaging of PDOs can provide real-time data on a patient’s likelihood to respond to treatment. This is critical information to inform treatment and can be implemented in an adaptive clinical trial. Preclinically, this optical toolkit can leverage the non-invasive and longitudinal spectroscopy modality to identify temporal windows of particular interest, such as early or late regression, dormancy, residual disease, or early recurrence. Imaging can then be used at those timepoints to capture holistic information efficiently. Finally, imaging of tumor-immune cell interactions, immune cell metabolism, and cancer-associated adipocytes will enhance our understanding of the dynamic interplay of the tumor and its microenvironment. Broadly, this technology may be applied across many cancers.

This dissertation advances the field of interpretable machine learning for high-dimensional neural time series by developing new methodologies for decoding latent structure from brain-wide neural recordings. Motivated by the challenge of identifying robust, biologically interpretable, brain networks descriptive and predictive of internal emotional states, particularly anxiety, this work presents a series of methodological innovations that address critical limitations in current machine learning approaches for structured neural data. Beyond developing novel methodology for brain network discovery, this work also presents an anxiety brain network which is validated broadly via numerical, qualitative, and behavioral strategies.Our brain networks consist of spectral features derived from local field potentials (LFPs), which operate as a neighborhood voltage in the brain around a given brain region. By recording LFPs while mice are simultaneously engaged in behaviors relevant to an emotional state of interest, machine learning methodologies can be applied to learn latent networks of LFP activity that explain behaviors. Previous works have made use of an interpretable supervised factor model known as discriminative cross-spectral factor analysis with nonnegative matrix factorization (dCSFA-NMF) to identify brain networks descriptive of stress, aggression, and social behavior. Traditionally, this framework has been applied by training on a carefully chosen behavioral context and then subsequently evaluating on new subjects in the same or related contexts. While this framework has been successful before, we demonstrate that the traditional modeling framework fails for our anxiety contexts – inspiring developments both in experimental training strategies and optimization frameworks. Key machine learning contributions include: (1) the development of a multi-assay supervision framework that enables the model to learn shared representations across heterogeneous behavioral paradigms; (2) the introduction of a cosine distance-based stability criterion for brain network representations across training iterations to assess and enforce the reproducibility of learned latent factors; and (3) a staged training procedure that integrates unsupervised NMF-based factor initialization, encoder pretraining for representational alignment, and domain-aware optimization with sample rebalancing. These methods are explicitly designed to identify brain networks that are independent of confounding behavioral states and more likely to generalize broadly in the population for predicting behavior and emotion. Scientific utility of the novel methodologies is validated in the context of decoding internal anxiety states in mice. Using neural recordings from eight brain regions across three distinct anxiety-related experimental paradigms – elevated plus maze, bright open field, and acute fluoxetine administration – we demonstrate that single-assay models fail to generalize beyond their training context. In contrast, the proposed multi-assay dCSFA-NMF framework identifies a brain-wide electome network that consistently and specifically encode an internal anxious state and a brain-wide electome network that may be more broadly predictive of negative affect. These networks generalize to held-out subjects and conditions, remain interpretable in terms of neuroanatomical structure and frequency content, and are validated through behavioral assays, disease models, and optogenetic manipulation. Further extending the generality of the modeling framework, I present preliminary work toward a "Universal Mouse Code of Emotion" - a multi-behavior, multi-dataset modeling architecture capable of decoding multiple internal states (anxiety and social behavior) across datasets with missing data due to non-overlapping recording configurations or sensor failure. This is achieved through the integration of a transformer-based encoder, variational inference, and a reconstruction-prediction joint training objective. The resulting model is shown to effectively impute missing data, maintain predictive accuracy across behaviors, and preserve interpretability through linearly constrained decoder weights and network sparsity. Collectively, this dissertation provides a comprehensive framework for interpretable, domain-general machine learning in the analysis of high-dimensional neural time series. The contributions made here extend the theoretical and practical utility of structured factor models in neuroscience, while also laying foundational tools applicable to other domains involving complex temporal signals with missing data, multi-task structure, and interpretability constraints. The anxiety biomarker serves as a case study that rigorously tests and demonstrates the efficacy of these methods, but the modeling principles and algorithms developed herein are broadly extensible to challenges in physiological signal analysis, wearable computing, behavioral sensing, and adaptive closed-loop systems. These contributions thus represent a significant advancement at the intersection of machine learning, neuroscience, and biomedical signal processing.