Washington University Medical Center
Washington University St. Louis: Open ScholarshipNot a member yet
26344 research outputs found
Sort by
System Security Foundations for AI-Enabled Systems
Full text linkAI is being deployed broadly, from conventional computing systems like IoT systems to more advanced agentic systems. It is shifting from being a specialized component responsible for specific functions to becoming the core of agentic systems, where it drives autonomous decision-making and task execution. These AI-enabled systems bring tremendous benefits. For example, large language models can interpret user intent, select appropriate tools, and access data to complete tasks with minimal human guidance. However, they also introduce new security, privacy, and safety risks. These risks arise not only from the models themselves but also from the broader system design and integration. Since these risks can span multiple system layers, they are difficult to limit and may affect the entire system. This complexity makes securing such systems particularly challenging, especially without compromising performance or functionality. We argue that addressing these challenges requires a holistic view of the system and an adaptation of system security principles to the layered nature of AI-enabled systems to reduce risks at each layer and limit their propagation. In this dissertation, we focus on securing both conventional and agentic AI-enabled systems by first measuring risks and then mitigating them through the design of secure system architectures. For conventional IoT systems, we identify the lack of adequate security protections and the prevalence of logic vulnerabilities in firmware that can expose on-device AI assets. We address this by developing a system-level intellectual property protection mechanism that enables secure model execution. Agentic systems, on the other hand, pose unique challenges due to their dynamic and often unpredictable behavior. Through measurement of real-world platforms such as OpenAI’s GPTs, we identify security and privacy issues, including insecure tool usage and data exposure. To mitigate these risks, we introduce execution isolation and access control that support safe tool use and data sharing. We also propose an automated permission management framework that learns user preferences to assist users in making safer permission decisions. Although new challenges continue to emerge as AI grows increasingly widespread, autonomous, and interconnected, we believe this work provides a strong foundation for securely developing and deploying AI-enabled systems
Computational Methods and Optimization for Naturalistic Functional Brain Mapping Using Diffuse Optical Tomography
Full text linkNaturalistic neuroimaging has significantly advanced our understanding of the brain in everyday situations. Stimulus designs that previously targeted individual brain regions are now being replaced by activities such as watching movies, engaging in conversations, and interacting with virtual reality environments. However, traditional functional neuroimaging techniques like functional magnetic resonance imaging (fMRI) are inherently unnaturalistic, as participants must lay down and remain still within the magnet\u27s bore. Consequently, these naturalistic paradigms require an equally naturalistic neuroimaging modality. High-density diffuse optical tomography (HD-DOT) offers a wearable alternative to fMRI by utilizing overlapping optical measurements to densely sample cortical brain activity. HD-DOT has proven effective in mapping brain responses to features from audiovisual movies, indicating that this method is appropriate for naturalistic neuroimaging. Here, we aim to optimize computational methods to advance naturalistic neuroimaging using HD-DOT. With movie viewing as an accessible naturalistic task, we map (or encode) audiovisual features, including concurrent features such as speech and faces from animated clips. We further demonstrate the feasibility of multi-sensory decoding by predicting which movie clip a participant viewed based on their DOT data. To improve our DOT data, we constructed a very high-density DOT imaging system for whole-head, high-resolution optical neuroimaging. This was validated in healthy adults completing functional localizer and movie-viewing tasks and was directly compared to fMRI. Decoding tasks underscore the repeatability of our signal, which is essential for naturalistic neuroimaging studies. Finally, a simulation approach was introduced to estimate the performance of optical neuroimaging systems using large-scale fMRI datasets. This provides a data-driven method for decision-making regarding new imaging systems and stimulus designs. Overall, this dissertation establishes and optimizes computational methods for naturalistic neuroimaging using DOT as a wearable surrogate for fMRI
Stereotyping in Language (Technologies): An Examination of Racial and Gender Stereotypes in Natural Language and Language Models
Full text linkThis dissertation examines stereotyping across natural language and language technologies through three interconnected studies. The first chapter applies a contemporary model of race relations from social psychology to investigate America\u27s racial framework within American English, revealing how language encodes hierarchical associations between racial/ethnic groups and attributes of superiority and Americanness. The second chapter extends this analysis to Large Language Models (LLMs), finding that these language technologies portray socially subordinate groups as more homogeneous compared to dominant groups. The third chapter investigates stereotyping in Vision Language Models (VLMs), showing that these language technologies generate more uniform representations for women than men and for White Americans than Black Americans, with uniformity increasing for more gender-prototypical appearances. This research demonstrates how stereotypes persist across natural language and AI systems, with many biases in language technologies mirroring established patterns of human social cognition. By integrating the fields of natural language processing and social sciences through interdisciplinary research, this dissertation documents bias patterns in language technologies and demonstrates how social psychological theories provide valuable tools for detecting and measuring stereotyping of AI systems
Dynamic Spectral and Systems-Theoretic Approaches for Inference and Control of Heterogeneous Complex Networks
Full text linkNetworks of nonlinear systems are commonly employed to describe a diverse range of phenomena across physics, engineering, neuroscience, and biology. The undesirable behaviors of such systems, in the form of neurological disorders, power grid failures, or ecological collapses, have spurred significant interest in understanding their dynamic structures and developing effective control strategies. These systems are typically large-scale, consist of heterogeneous units, and are partially observable with limited measurement data, presenting theoretical and computational challenges for control design and connectivity inference. This thesis addresses these challenges by developing novel algorithms for pattern formation in populations of stable limit-cycle oscillators and connectivity reconstruction in complex networks from time-series data. The first part focuses on controlling nonlinear oscillator networks with heterogeneous units under aggregate population-level measurement constraints. The primary contribution is the development of a unified framework that leverages spectral approximation to transform the optimal control problem into a simple convex quadratic program with linear constraints, applicable to both open-loop and feedback control tasks within a model-based setting. We demonstrate its efficacy through numerical simulations and experimental validations on ensembles of electrochemical oscillators. The proposed framework is further extended to enable data-driven learning using aggregated measurements by characterizing the network synchronization patterns using the Fourier coefficients of the population mean. The second part of the thesis addresses the network reconstruction challenges arising from limited measurement data and partial node observations. We first present a data-efficient algorithm by formulating the network inference task as a bilinear optimization problem. An iterative algorithm with sequential initialization is proposed to solve the resulting bilinear program. We then tackle partial observability by integrating time-delay embedding with statistical learning techniques. The performance of the proposed algorithms is compared with existing methods across experimental and simulated datasets, comprising oscillatory, non-oscillatory, and chaotic dynamics. This thesis advances both control strategies for oscillator networks and inference techniques for nonlinear networks, contributing to the manipulation and understanding of complex dynamical networks
Nucleocytoplasmic Regulation of Epithelial Transitions
Full text linkNucleocytoplasmic transport fundamentally regulates cellular phenotype and function across tissues in the body. In epithelial tissues, various processes—including embryogenesis, tissue repair, and cancer invasion—require cells to go from a homeostatic, mostly stationary and solid group to a fluidlike migratory collective. Underlying this tissue destabilization are two physical transitions: the epithelial-mesenchymal transition (EMT) and the unjamming transition (UJT). EMT comprises the loss of cell-cell junctions, front-back cell repolarization, and elevated migration. In contrast, UJT preserves cell-cell junctions, but cell elongation and migration fluidity more significantly increase. While nucleocytoplasmic transport forms the basis of EMT, UJT has only recently gained recognition, and its underlying nucleocytoplasmic transport mechanisms are still being revealed. Nuclear export inhibition (NEI) offers a conducive method for studying these transitions by directly interfering with nucleocytoplasmic transport. NEI works by binding to a nuclear export receptor, thereby inhibiting the transport of its cargos out of the nucleus. In this way, NEI facilitates study of the retained cargos and their impact on cellular outcomes. NEI commonly targets the exportin CRM1, which has over 370 known cargo proteins, many of which regulate epithelial cell phenotypes and cancer progression. While CRM1-based NEI has been studied as a cancer therapeutic for its effects on growth and proliferation, reports covering effects in healthy cells are limited. Preliminary experiments for this body of work indicated that CRM1-based NEI induces EMT and apparent UJT in healthy cells. Therefore, subsequent studies leveraged NEI as a tool to explore fundamental EMT and UJT questions in different contexts. EMT is regulated by the nucleocytoplasmic transport of core (e.g., SNAIL, TWIST, ZEB) and affiliated EMT factors (e.g., YAP, IκBα). Multiple nucleocytoplasmic proteins have been individually characterized as switches, either promoting or inhibiting EMT. Yet, the combined interaction of opposing switch-like proteins has not been investigated. Chapter 2 explores how these opposing E-M switch-like proteins interrelate, using NEI to understand their combined influence on cell phenotype. UJT arises from increased cellular propulsion during conditions of maintained cell-cell adhesion. However, the intracellular processes that enable increased migration fluidity during UJT are not fully understood. Chapter 3 applies NEI to study the physical and biological mechanisms underlying fluidlike migration. Experiments investigate whether decreasing cytoplasmic viscosity facilitates migration fluidity, with particular consideration for potential connections to glycolytic metabolism. Measurements further examine how glycolytic metabolism might be linked to actin fiber formation to drive fluidlike migration, surveying involvement of the FMRP/CYFIP1/eIF4E/cofilin signaling axis. Finally, the analyses explore whether Fragile X Messenger Ribonucleoprotein (FMRP) could exhibit mechanosensitive nuclear localization, potentially triggering the fluidlike adaptation of cells to their physical environment. Together, these experiments reveal new biophysical mechanisms regulating epithelial migration. Chapter 2 demonstrates that NEI induces concurrent epithelial-mesenchymal states, leading to an expanded model of EMT. Chapter 3 classifies highly fluid migration that does not align with the standard protein and morphology markers for EMT or UJT. Results indicate that this fluidity might stem from increasing cellular deformability, potentially regulated by mechanosensitive FMRP signaling. Clinically, both studies highlight significant off-target effects of NEI, raising additional considerations for therapeutic use. Overall, the findings enhance our understanding of epithelial transitions by expanding traditional migration models, and the analyses introduce new questions concerning epithelial function in health and disease
The Crosstalk Between Muscle and Adipose Tissue in Health and Disease
No full textThe aim of this dissertation is to investigate the crosstalk between adipose and muscle tissues in physiology and pathophysiology, particularly focusing on the therapeutic potential of muscle-associated adipose tissue in muscle regeneration. Adipose tissue, an endocrine organ, secretes cytokines (a.k.a. adipokines) that significantly and negatively impact muscle function in conditions like diabetes, obesity, and aging. Whether similar signals mediate communication between muscle and its adjacent adipose tissue through paracrine signaling is unknown, as is whether this communication is therapeutically modifiable. Understanding the interactions between adipose and muscle tissues could lead to novel therapeutic strategies for improving muscle regeneration and function. To define and manipulate paracrine adipose-muscle signaling, we developed an experimental transplanted adipose-muscle interface in the mouse rotator cuff for in vivo studies and collected human muscle and adipose samples from individuals with and without diabetes for in vitro experiments. Our previous research demonstrated that transplanted brown adipose tissue (BAT) enhances muscle regeneration, compared with white adipose tissue (WAT), through paracrine signaling. Given the potential pro-regenerative role of BAT-secreted cytokines, we hypothesized that stimulating muscle-associated adipose to adopt BAT-like characteristics could improve muscle regeneration. To explore this hypothesis, we utilized RNA sequencing to profile transcriptional changes in muscle-resident cell populations during regeneration in our BAT transplant model, aiming to identify potential “batokines” and their target cells involved in muscle regeneration. Concurrently, we profiled intramuscular adipose tissue (IMAT) and subcutaneous fat in human subjects to characterize depot-specific differences in secreted cytokines and their browning and pro-myogenic potentials. Adipose progenitor cells (APCs) were isolated and used to condition primary human muscle myoblasts in vitro to assess the impact on myogenesis. Last but not least, we developed a tunable polyethylene glycol hydrogel embedded with nanosilicates, aiming for sustained local release of a browning factor, mirabegron. We then tested whether sustained local mirabegron will locally brown adipose tissue in vivo in mice without systemic off-target effects. Our key findings reveal that transplanted BAT increases muscle mass in the early stages of regeneration primarily by modulating fibro-adipogenic progenitor (FAP) signaling, potentially through immune regulation. Although there were no long-term improvements in muscle mass or function in wildtype mice, in the absence of endogenous BAT, BAT transplantation improved muscle contractile force independent of changes in muscle mass, suggesting potential direct regulation of muscle contraction by “batokines”. RNA sequencing of human IMAT and subcutaneous fat revealed distinct transcriptional profiles, highlighting depot-specific differences and their respective browning potentials. IMAT showed reduced inflammatory pathways and an anti-myogenic profile for secreted cytokines compared to SQ. Stimulation of isolated human APC from IMAT fail to push them toward a brown/beige profile, and thus did not improve myogenesis of myoblasts in vitro. The implantation of our biodegradable gel proved to be viable up to 2 weeks in vivo with minor surrounding inflammation. The mirabegron-binding efficiency to the NS was measured to be 87% ± 5 and released mirabegron retained bioactivity. The mouse injected with mirabegron-loaded gel as proof-of-principle exhibits evidence of increased lipolysis, but not substantial adipose browning, on the mirabegron treated side compared with the contralateral side treated with unloaded gel. Importantly, the gel-treated mice did not exhibit increased heart rate associated with bolus mirabegron injection, suggesting that gel loading the drug avoids unwanted off-target effects. These data support the potential efficacy of the biomaterials approach, but also the need to tailor dosing and timing. Taken together these studies shed new light on the role of muscle-associated adipose in regulating muscle physiology. While exogenous brown fat has a moderate impact on muscle regeneration, human IMAT secreted factors do not improve in vitro myogenesis and human IMAT lacks the capacity for browning, shifting our therapeutic interest to adipose outside the muscle. We show that it is possible to induce browning locally through sustained release mirabegron, suggesting that this could target endogenous epi-muscular adipose
Synthetic Gene Circuits for Self-Regulating Circadian- and Inflammatory-Driven Drug Delivery
No full textHomeostasis represents a critical balance between catabolic and anabolic states, which is especially evident in tissues like cartilage. When homeostasis is disrupted, such as in rheumatoid arthritis (RA), progressive joint destruction occurs. While inflammation is required in response to infection or tissue injury, aberrant immune activity can contribute to harmful outcomes in diseases like RA. The proinflammatory state in RA is perpetuated by cytokine signaling, notably including interleukin-1 (IL-1). Intracellular pathways activated by this signaling regulate networks such as the nuclear factor-κB (NF-κB) pathway, creating an environment of sustained inflammation. While no cure exists for RA, advances in disease modifying antirheumatic drugs (DMARDs), particularly in biologic drugs, have opened new opportunities for tailored treatment to target specific cytokines. For example, the targeted biologic IL-1 receptor antagonist (IL-1Ra; anakinra) is an FDA-approved DMARD. However, current treatment requires repeated, high doses to achieve clinical impact, resulting in significant side effects. Making disease management more challenging, patients with RA experience dynamic changes in inflammation across multiple timescales. On a short-term scale, inflammation peaks in the early morning each day, resulting in characteristic morning stiffness of joints. On a longer-term scale, disease varies between flares and remissions that can last weeks to months. Therefore, there is a need for drug delivery approaches that account for dynamic changes in disease state to optimize drug delivery. To address daily changes in disease, chronotherapy is an emerging concept that aligns therapeutics to the body’s circadian rhythm. While chronotherapy has shown success for synthetic DMARDs and anti-inflammatories in pre-clinical and clinical studies for RA, circadian medicine for biologic DMARDs has not been explored. Therefore, we sought to generate self-regulated drug delivery mechanisms for circadian IL-1Ra release. To accomplish this aim, we generated synthetic gene circuits driven by clock-controlled elements called E’-boxes, D-boxes, and RREs, which stimulate downstream gene expression according to the endogenous intercellular circadian feedback loops. We demonstrated that these chronogenetic gene circuits produced IL-1Ra at unique phases in a 24-hour cycle and protected an in vitro model of arthritis from circadian disruption and inflammatory degradation. Building on this concept, we engineered a dual-responsive synthetic gene circuit to address both short- and long-term dynamic inflammation. Utilizing a promoter with OR-gate logic driven by both inflammatory NF-κB-responsive elements and circadian E’-box elements, this design generated cyclical daily output to target circadian changes in inflammation plus enhanced therapeutic release on-demand in inflammatory environments, assessed in vitro. Finally, we translated the chronogenetic drug delivery mechanism in vivo, where we assessed hydrogel-encapsulated engineered cartilage constructs for self-regulated daily drug delivery in the K/BxN serum transfer arthritis (STA) model. We demonstrated that engineered cartilage cycled in vivo using bioluminescent reporters, and key clinical features of arthritis including joint swelling, pain, and bone erosion were improved with engineered implants. Together, this research adds to the toolkit for synthetic biology-based tailored drug delivery in arthritis. By utilizing disease-relevant input cues, the gene circuits described overcome limitations in current treatment approaches, demonstrated both in vitro and in vivo, and motivate new opportunities for personalized treatment
A Dissertation on Belonging in Occupational Therapy Education with Students from Historically Minoritized Identities
No full textThis dissertation provides a comprehensive exploration of the phenomenon of belonging specifically tailored to students with historically minoritized identities (HMIs) within the Washington University Program in Occupational Therapy (WUOT). Employing qualitative methodologies including scoping reviews, theoretical conceptualizations, focus groups, individual interviews, surveys, and participant video diaries, the study delves into the multifaceted nature of belonging within the context of occupational therapy (OT) education. The dissertation is organized into four key articles. The first two articles lay the theoretical foundation by examining the historical context of education as forced assimilation for minority groups and the nuanced use of professional behaviors to continue forced assimilation in OT education today and through the use of scoping review on belonging in health professions education to elucidate the factors that influence belonging in this educational milieu. The second two articles aim to provide a holistic examination of WUOT by exploring equity in the admissions process including how social/cultural capital are required to simply apply to WUOT and the longitudinal experiences of belonging for OT students who hold HMIs- focusing on the intersection of identity in context. Several major themes emerged from this dissertation. 1) Conceptual Understanding of Belonging: The dissertation emphasizes the need to understand how a sense of belonging is crafted for OT students with HMIs at WUOT. It examines the factors influencing belonging, such as prior educational experiences, the context of OT education, and students\u27 social identities. 2) The role of student perceptions of belonging - Students\u27 perceptions of belonging in OT education are shaped by their previous experiences with belonging in higher education and are influenced by their historical experiences with discrimination and bias in both education and society. These factors collectively impact their sense of belonging and success in the OT program. 3) The role social identities in context play in belonging at WUOT- At WUOT, social identities against the backdrop of the larger environment significantly impacts students\u27 sense of belonging, as individuals who hold HMIs are often impacted by multiple systems of power and oppression and may feel their identities are threatened choosing to alter their behaviors to mitigate potential negative treatment. The dissertation integrates Bronfenbrenner\u27s ecological system theory and Carter\u27s supplement to Bourdieu\u27s theory of social and cultural capital to reveal how students with HMIs experience belonging at WUOT. It emphasizes the need for WUOT to intentionally counteract previous educational messages about social identity acceptance and facilitate student success by addressing factors within the microsystem, such as relationships, policies, and visibility of non-dominant social identities, while being influenced by broader societal norms. In summary, this dissertation highlights the necessity of understanding and actively fostering a sense of belonging for students with HMIs through systemic changes, redefined success measures, and an intentional, identity-focused approach within the educational context of WUOT
Essays on Macroeconomic Growth in an Open Economy
No full textIn the last three decades, the world economy has become increasingly interconnected through multinational production and international trade. This dissertation examines how these factors influence the economic growth of individual countries. The first two chapters study the effect of multinational firms and trade in cross-country knowledge diffusion. The first chapter introduces a novel model of knowledge diffusion with trade and multinational production to analyze their effects on productivity growth across countries. The second chapter develops a method to bring empirical evidence to the intensity of knowledge diffusion. Using the Chinese outward investment promotion that occurred during the year 2004—2006, the result shows that a 1 percentage point increase in multinational firms originating from China leads to a 1.83% increase in the total factor productivity of the recipient country’s sector. We show that this methodology can provide evidence on key parameters in knowledge diffusion models. Counterfactual analysis uncovers the sets of countries that gain and suffer, depending on changes in trade or multinational production costs. The final chapter highlights the role of service trade in economic growth. Focusing on the case of India, I find that modeling services as tradable is crucial in accounting for India\u27s pattern of structural transformation. The model I developed indicates that India\u27s producer services experienced substantial productivity gains relative to the global average. Additionally, estimates indicate that producer services experienced the steepest decline in trade costs over the studied period. These factors drove structural transformation in India and also resulted in shifts towards more productive sectors, increasing allocative efficiency
Brief in Support of Respondents as Amici Curiae, Donald J. Trump, President of the United States, et al. v. Rebecca Kelly Slaughter, et. al, No. 25-332
No full textIn the years since this Court’s decision in Seila Law LLC v. Consumer Financial Protection Bureau, 591 U.S. 197 (2020), legal historians—including originalists—have unearthed a rich body of evidence challenging the conventional wisdom concerning the scope of the President’s removal powers at the Founding. Revisiting not only the Framers’ intellectual influences, but also their driving philosophical concerns and political incentives, this scholarship explains that removal authority was not understood to be vested exclusively in the President as a matter of settled constitutional law, but rather was permissibly exercised by various actors in various ways, intentionally not settled by the text of the Constitution but governed by practical and policy considerations.
To be sure, many early statutes vested removal power in the President. One of the Framers’ greatest fears was “to continue a bad man in office.” The Papers of James Madison, vol. 12, 2 March 1789 – 20 January 1790 and supplement 24 October 1775 – 24 January 1789 at 173-174 (1979). Impeachment was cumbersome, and joint removal of officers by the Senate and the President perhaps more so; but presidential removal was quick, aided by the President’s informational advantages. Thus, while the Founding generation disagreed broadly over what the executive power meant, at least some agreed that, in the President’s hands, the removal power would be most usefully exercised.
As recent scholarship demonstrates, however, this utilitarian approach to removal also explains why some statutes enacted during the Founding era declined to give the President unfettered removal power. Where expansive notions of presidential authority conflicted with the imperatives of good government, early Congresses (uncontroversially) sided with the latter. For example, certain statutes insulated early financial regulators from removal entirely on the grounds that those regulators required stability in their jobs and insulation from political manipulation. Other early statutes vested removal authority in actors other than the President, such as the courts. And for officers involved in the delicate job of managing the money supply—inspecting imports, collecting taxes, coining money, delivering the mail, and others—criminal prosecution was frequently the preferred mode of removal.
Against this backdrop, the late nineteenth-century rise of the modern “administrative state”—marked by the creation of new departments within the Executive Branch, helmed by a new class of executive officers with specialized areas of authority and expertise did not so much upset a tradition of unilateral presidential removal authority as apply the more heterodox set of early American removal practices to a new context. To be sure, the modern capitalist economy brought a proliferation of seemingly new forms of regulation, plus a merit-based civil service. But if Gilded Age Americans were uniquely fixated on officer independence as an antidote to corruption in Washington, the structures and arrangements they put into place to achieve those ends—including statutes that limited the President’s authority to remove—were not new. Rather, these structures were deeply rooted in longstanding American traditions.
Two important conclusions follow from recent scholarship’s understanding of removal at the Founding. First, although the President has always exercised considerable removal power, that power was not unfettered: it was circumscribed by law, including, at times, by the requirement that officials could only be removed for cause. Second, and relatedly, removal has not been treated as a settled feature of executive power emanating from the Constitution; rather, it has functioned as a management tool deployed by Congress to ensure an honest and effective administration. Understood this way, the limits on presidential removal authority that this Court recognized in Humphrey’s Executor v. United States, 295 U.S. 602 (1935), are consistent both with practice stretching back to the Founding and with other decisions from this Court recognizing robust presidential removal authority in certain contexts