Washington University Medical Center
Washington University St. Louis: Open ScholarshipNot a member yet
26344 research outputs found
Sort by
Mechanisms of MLL-rearranged Infant Leukemogenesis
Full text linkInfant leukemias arise as either B cell acute lymphoblastic leukemia (B-ALL) or acute myeloid leukemia (AML) and are primarily driven by MLL/KMT2A-rearrangements (MLLr). Prior work has shown that MLLr transform fetal/neonatal hematopoietic progenitors more efficiently than adult progenitors. However, the genetics and epidemiology of MLLr leukemias present an important paradox. MLLr arise during late gestation under conditions that would seemingly favor leukemic transformation (i.e., fetal/neonatal progenitors have high proliferation rates and self-renewal). Furthermore, MLLr infant leukemias require very few cooperating mutations for transformation. Yet congenital leukemias are 10-fold less common than infant leukemias and \u3e100-fold less common than childhood leukemias overall. These observations raise the question of whether mechanisms exist to suppress leukemic transformation during fetal stages of life. Here, we use mouse models to show that fetal expression of MLL::ENL (also called KMT2A::MLLT1) creates a heritable, leukemia-resistant state that persists after birth and even after transplantation. When MLL::ENL is induced shortly after birth, transformation proceeds efficiently in this context. Heritable, fetal protection against leukemic transformation potentially explains the low incidence of congenital leukemias in humans, and it illustrates how ontogeny not only conveys leukemia-permissive states, but also leukemia resistant states. Furthermore, we show that MLL::ENL promotes a B cell bias in fetal/neonatal progenitors, even though the mouse models ultimately develop AML. We identified SKIDA1 as a temporally-restricted effector of this B cell bias, suggesting a potential explanation for why infant MLLr leukemias most often present as B-ALL, whereas non-infant MLLr leukemias more often present as AML. This work provides a better understanding of how leukemogenic mutations and normal developmental programs interact. If we can understand how ontological programs repress or potentiate leukemogenesis, we can potentially reprogram progenitors to treat high-risk pediatric leukemias
Development of a Prototype AWSM-PET System for Augmented Whole-body PET Imaging
Full text linkPositron Emission Tomography (PET) is a critical imaging modality for medical diagnostics, yet its accuracy is often limited by spatial resolution and sensitivity constraints. This dissertation presents the development of Augmented Whole-body Scan via Magnifying PET (AWSM-PET), a novel technique integrating high-resolution auxiliary detectors with a clinical PET scanner to enhance sensitivity and resolution while remaining fully compatible with standard clinical protocols. The study encompasses hardware development, image reconstruction methods, and performance evaluation through imaging studies, including human trials. High-resolution outsert panel detectors were designed, mounted on a mobile cart, and integrated with a Biograph Vision PET/CT scanner. Firmware and software modifications enabled simultaneous data acquisition under clinical protocols. The outsert detectors achieved an energy resolution of 11% full width at half maximum (FWHM) at 511 keV and a coincidence resolving time (CRT) of 183 ps FWHM. Sensitivity improved by up to 18.4%, depending on source location. A data-driven geometric alignment method was developed using point source measurements, achieving sub-millimeter precision without requiring special landmarks or repetitive measurements. This approach optimized the outsert detector positioning to minimize angular discrepancies in detected lines of response (LORs) via an iterative optimization method. Monte Carlo simulations confirmed that the deviation between the ground truth and estimated geometry remained well below 1 mm. Experimental validation demonstrated that point source images using the estimated geometry aligned well with those from the Biograph Vision scanner, with centroid deviations below 1 mm and angular deviations of line source images under 1 degree across the field of view. To support image reconstruction, a Maximum Likelihood Expecation Maximization (MLEM) listmode image reconstruction technique was developed for continuous-bed-motion (CBM) reconstruction in AWSM-PET. To balance reconstruction speed and accuracy, a single-ring approximation and grouped crystal approximation were introduced to accelerate sensitivity image calculation. Scatter and random corrections were implemented to achieve quantitative images, incorporating approximations to optimize computational efficiency. Additionally, a spatially variant point spread function (PSF) model was compared with a spatially invariant model. Rather than enhancing spatial resolution, the PSF model primarily contributes to improving contrast recovery and noise control. AWSM-PET was evaluated through NEMA-IQ, and mini-Derenzo phantoms, as well as human imaging studies. Compared to the Biograph Vision scanner, the mini-Derenzo phantom study demonstrated improved spatial resolution for spherical lesions ≤6 mm in diameter, and the NEMA-IQ study confirmed higher contrast recovery coefficients (CRC) for the smallest lesion (4.88 mm Ø). Initial human imaging validated clinical compatibility, showing enhanced resolution in high-count regions (e.g., brain) but increased noise in low-count regions (e.g., abdomen), highlighting areas for further refinement
Towards Secure and Privacy-Preserving Machine Learning Systems
Full text linkIn recent years, machine learning (ML) has advanced at an unprecedented pace, driving the widespread adoption of increasingly sophisticated models across a broad range of real-world applications, including healthcare, finance, autonomous systems, and critical infrastructure. While these models have delivered remarkable benefits and transformed numerous industries, they remain inherently vulnerable to a variety of security and privacy threats. Given their growing role in safety-critical domains, ensuring their security and privacy has become imperative. Systematically addressing these vulnerabilities requires comprehensive adversarial analyses to uncover weaknesses and inform the design of robust defenses. This dissertation systematically investigates ML vulnerabilities in adversarial settings from two complementary perspectives: adversarial capabilities and adversarial goals. From the capability perspective, I examine attacks launched by both cyber and physical domain adversaries. From the goal perspective, I explore attacks targeting three critical security properties: integrity, availability, and confidentiality. First, in the cyber domain, I investigate integrity threats to text-to-image generation models through tailored adversarial attacks, and propose a novel membership inference attack based on information bottleneck theory to compromise confidentiality. Second, in the physical domain, I explore integrity threats by crafting physically realizable adversarial examples against automatic speech recognition systems deployed in video conferencing platforms. I also design availability attacks that degrade the operational efficiency of LiDAR-based detection models. Finally, beyond individual attacks, I examine the unintended interactions among security properties and other essential ML system attributes. Specifically, I analyze trade-offs between privacy and explainability, both crucial for ensuring trustworthy ML systems, and further investigate the interplay between availability and privacy in federated learning environments. Through these comprehensive explorations, this dissertation provides an in-depth understanding of ML system vulnerabilities from multiple perspectives, thereby contributing to the development of more secure and privacy-preserving machine learning systems
Cyber-Physical Security Through the Lens of AI-Enabled Systems
No full textCyber-physical systems (CPS), powered by emerging artificial intelligence (AI) technologies, have become integral to various critical domains such as the Internet of Things (IoTs), medical devices, and autonomous vehicles. A unique aspect of these systems lies in their interactions with the physical world, by perceiving environments through heterogeneous modalities (perception), processing digital data with human-in-the-loop intelligence algorithms (computing), and autonomously actuating controls that affect physical processes (actuation). While this intricate fusion of cyber and physical components has unlocked unprecedented capabilities, it has also introduced new security challenges. However, traditional security measures often fall short in addressing these multifaceted threats. This dissertation aims to systematically explore and mitigate the threats inherent in AI-enabled cyber-physical systems. The research objectives are threefold: (1) investigating how the interplay of cyber and physical components opens up novel attack vectors, (2) developing robust defense strategies grounded by physical laws and constraints, and (3) benchmarking and theoretically analyzing security trade-offs from algorithmic, system-level, and human-centric perspectives. By bridging the gap between cyber and physical domains, my work seeks to enhance the resilience and trustworthiness of modern CPS while retaining system efficiency and usability
Biomedical Application of Hybrid Protein-based Hydrogel
No full textThe urgent need for internal bio-adhesives capable of maintaining robust performance in physiological environments has driven the development of novel biomaterials that combine strong wet adhesion, biocompatibility, and mechanical resilience. Such adhesives are essential in diverse clinical contexts, including bone tissue fixation and vascular sealing. However, conventional materials often fall short under physiological conditions due to their limited adhesion strength, long curing times, or insufficient compatibility with biological tissues. This dissertation addresses these limitations by presenting a systematic investigation into the rational design, synthesis, and performance evaluation of recombinant hybrid hydrogels composed of silk, amyloid, and mussel foot protein (SAM) domains. In Aim 1, the sequence–structure–property relationship of SAM hydrogels was elucidated through controlled variation of their β-sheet-forming amyloid components and amorphous mussel foot protein domains. By tuning the ratio of crystalline to amorphous regions, we demonstrate that β-sheet-rich domains significantly enhance cohesive strength, while extended mussel foot protein regions promote interfacial adhesion under wet conditions—key features for successful tissue integration and durability in vivo. In Aim 2, to address the challenge of insufficient mechanical strength, nanocomposite hydrogels were developed by integrating polydopamine-functionalized cellulose nanocrystals (CNCPDA) into the SAM matrix. These rod-like nanoparticles, known for their high stiffness and modifiable surface chemistry, were chosen to improve the hydrogel’s mechanical integrity without compromising flexibility. The catechol functionalities in polydopamine enhanced the interaction between the nanofillers and protein matrix, thereby reinforcing the hydrogel network and expanding its potential use in load-bearing applications such as orthopedic repair. In Aim 3, we focused on optimizing adhesion kinetics and usability under clinically relevant conditions. A composite hydrogel system was formulated by combining SAM proteins with polyacrylic acid (PAA), a biocompatible polyelectrolyte with strong water absorption capacity. The resulting SAM–PAA hydrogels exhibited rapid adhesion to moist biological tissues at low contact pressure (\u3c17 kPa) and short incubation time (\u3c15 minutes), making them promising candidates for vascular repair and other surgical procedures requiring fast, effective sealing with minimal handling. Altogether, this dissertation provides new insights into how molecular-level protein design and nanocomposite strategies that can be leveraged to fine-tune the adhesive and mechanical properties of hydrogels. The results presented here establish a framework for the development of next-generation recombinant protein-based materials tailored for complex biomedical applications. By integrating principles from synthetic biology, protein engineering, and materials science, this research contributes to the advancement of programmable bioadhesives for internal use in regenerative medicine and soft tissue repair
Characterizing the Regulatory Roles of pmeR in Indole-3-acetic acid (IAA) Responsiveness and Pathogenesis in Pseudomonas syringae pv. tomato strain DC3000.
No full textMicroorganisms rely on intricate signaling networks to respond to environmental changes. For plant-pathogenic microbes like Pseudomonas syringae pv. tomato strain DC3000 (PtoDC3000) reprogram gene expression in response to host-derived signals such as the phytohormone auxin (indole-3-acetic acid, IAA). In Arabidopsis-PtoDC3000 interactions, IAA promotes bacterial pathogenesis by suppressing plant salicylic acid-mediated immunity and regulating bacterial gene expression. However, the molecular mechanisms underlying bacterial IAA perception and transcriptional regulation remain largely unexplored. This study investigates the regulatory role of pmeR, which encodes the TetR-like family transcriptional regulator PmeR, in regulating auxin-mediated bacterial gene expression and virulence. PmeR can sense host-derived signals, such as flavonoids, and derepress gene expression of both its own gene (pmeR) and the mexAB-oprM efflux operon. Pevious transcriptomic evidence suggested that both pmeR and the mexAB-oprM operon are upregulated by IAA; however, whether PmeR regulates this auxin signaling pathway was not known. Using genetic, molecular, and biochemical approaches, I demonstrated that IAA induces expression of pmeR and the mexAB-oprM operon via both pmeR-dependent derepression and pmeR-independent mechanisms. Interestingly, IAA itself does not interfere with the binding of PmeR to the DNA sequences, indicating that PmeR does not act as a direct IAA sensor. Instead, its conjugated form, IAA-lysine, prevents the formation of PmeR/DNA complex, suggesting it is a ligand of PmeR. In addition to regulating its known targets, I showed, using molecular and biochemical approaches, that pmeR positively regulates additional auxin-upregulated genes, including PSPTO_1824 and PSPTO_4297. However, this does not appear to be through direct DNA binding. Instead, pmeR likely modulates intracellular IAA levels via mexAB-oprM-mediated export, indirectly influencing gene expression. Finally, to further delineate the scope of PmeR’s regulatory network, I used bioinformatic and biochemical analyses to identify additional auxin-responsive PmeR regulons, suggesting a broader role of PmeR in auxin-responsive transcriptional regulation. In summary, this dissertation advances our understanding of how P. syringae integrates host-derived auxin signals through regulatory networks mediated by the TetR-like transcriptional regulator PmeR, ultimately contributing to colonization within plant tissue. This study provides a foundation for future studies on hormone-mediated host-pathogen interactions and bacterial signal transduction mechanisms. Additionally, these findings suggest potential directions for developing strategies to mitigate bacterial diseases in crops by targeting auxin-responsive regulatory pathways
Parsing Heterogeneity in Depression
No full textDepression is a highly prevalent mental health disorder that often requires treatment. While current treatments are effective for a large portion of patients generally, the exact medication that works for any specific person is unclear, leading to large delays in symptom remittance if at all. This hurdle has motivated substantial efforts to identify markers of depression that might predict treatment outcomes. However, large sources of heterogeneity in depression have stymied progress on identifying robust and replicable neuroimaging markers. Parsing this heterogeneity is crucial but mired with challenges. My dissertation work has elucidated several key barriers in parsing the heterogeneity of depression. In chapter 2, we identified small but replicable biomarkers of depression in the UK Biobank. In chapter 3, we found these small effect sizes for biomarkers of depression can be increased by parsing clinical sources of heterogeneity. We also verified the presence of many-to-one mapping relationships between symptoms and the brain, providing novel insights into the mechanism of depression. This finding strongly indicates future work ought to relinquish the notion that clinical subtypes will explain neurobiological heterogeneity and vice versa. Both symptoms and neuroimaging must be accounted for when identifying subtypes or clinically relevant markers of depression. In chapter 3, we identified the impact of methodological variability on subtyping efforts, identifying subtyping techniques to avoid and validation analyses to include. For example, we recommend future work compare their findings to null data and to previous subtyping approaches. These advancements will hopefully allow the field to successfully identify and parse the sources of heterogeneity in depression. Such progress should allow for more robust biomarkers of depression and therefore treatment guidelines, ultimately improving patient care
Mechanisms of Voltage Gating in the Cyclic Nucleotide Binding Domain Channel Family
No full textKey physiological processes such as pacemaking in the heart, electrical signaling in the brain, photoreception in the eye, olfaction in olfactory sensory neurons, and stomata opening in plant leaves involve channels from the cyclic nucleotide binding domain (CNBD) ion channel family. CNBD channels are found throughout the tree of life, and for those with cryo-EM structures, show striking similarity in their architecture despite exhibiting diversity in gating polarity (i.e. channels that activate upon membrane hyperpolarization, depolarization, or are voltage-insensitive). Single mutations in CNBD channels can not only reverse gating polarity, but can also create channels that activate upon both membrane hyperpolarization and depolarization. To our knowledge, this plasticity in gating polarity is unique to the CNBD family within the superfamily of voltage-gated ion channels. In this thesis, I discuss the functional and molecular evolutionary approaches I used to determine the interactions between multiple structural elements involved in gating polarity. Voltage-gated members of the CNBD family are found primarily in three subfamilies: hyperpolarization-activated cyclic nucleotide-gated (HCN) channels, ether-à-go-go (EAG) channels, and Plant Voltage-Gated K+ (Plant VG K+) channels. Prior structural and functional studies show that different interactions between the pore domain and the voltage-sensing domain contribute to the diversity of gating polarity observed in these subfamilies. However, chimeragenesis studies reveal that other structural domains may also be involved in channel gating. In Chapter 2, I assess the impact of the C-terminal domain on gating polarity by swapping the C-terminus from members of the HCN, EAG, and Plant VG K+ channel families into a chimeric background. I then proceeded to perform alanine-scanning mutagenesis of the C-terminus to identify specific residues with the greatest contribution to gating polarity. In Chapter 3, I generated a state equilibrium model to interpret the gating scheme for all CNBD channels. To empirically test and validate the function of each subfamily, I used electrophysiology to characterize select extant CNBD channels in Chapter 4. These select extant channel species are least molecularly divergent to their respective subfamily’s ancestral channel, thus, giving us insight into residues and interactions that remain through evolutionary time key to channel gating polarity. Lastly, Chapter 5 is a discussion of the work presented in this thesis and future directions for molecular evolutionary and functional analyses of gating polarity in CNBD channels. Taken altogether, this body of work explores the molecular determinants for gating polarity through a functional and molecular evolutionary lens
Entanglement-Enhanced Metrology With Superconducting Circuits
No full textCircuit quantum electrodynamics provides a unique platform for investigating fundamental physics and practical quantum applications. In this thesis, I introduce the superconducting circuit platform from a foundational perspective. Drawing inspiration from quantum electrodynamics and utilizing the analog of closed time-like curves, this work achieves quantum enhancement over classical strategies. Specifically, I investigate agnostic phase estimation protocol and the associated approaches that leverage quantum entanglement to optimally estimate an unknown rotation angle without requiring prior knowledge of the rotation axis. This work not only demonstrates a proof of concept for a type of entanglement-assisted metrology but also highlights intriguing quantum effects. To establish the theoretical framework, I include a pedagogical introduction to quantum and classical Fisher information - the key concepts we utilize for quantifying sensor performance. Finally, I detail the experimental techniques that enable the demonstration of metrological advantage, weighing the benefits of quantum enhancement against the costs of entanglement manipulation
Essays on the Social Operations
Full text linkThis dissertation focuses on the intersection of operational decisions and their social implications, an area of growing importance as businesses increasingly optimize for efficiency while simultaneously facing pressures to consider environmental sustainability and technological adoption. In my research, I utilize field experiment and econometrics to study how innovations in delivery speed and algorithmic implementation influence human behavior and create broader impacts beyond immediate business outcomes. In Chapter 1, Green E-commerce: Environmental Impact of Fast Delivery\u27\u27, we study the impact of faster delivery on how consumers place orders (their order frequency and basket sizes) and the subsequent environmental implications. Specifically, we leverage a quasi-experiment involving the opening of a new local warehouse by Alibaba Group, which led to a half-day improvement in the delivery speed for local orders. Through a difference-in-differences analysis, we find that the delivery speed improvement not only increased consumers\u27 monthly purchasing amount by 6.70%, but also increased monthly order frequency by a higher percentage (i.e., 7.74%) and reduced the average order basket size by 0.79%. These results collectively suggest that with faster delivery, consumers purchase more on the platform but do so in more frequent and smaller orders, which implies more packaging and transportation costs for each unit of product sold. Based on these results, we conduct a detailed calibration using both public and company-specific data to estimate the increase in the platform\u27s carbon emissions due to faster delivery. We also explore and identify two mechanisms contributing to the phenomenon: order-splitting and category expansion. We combine these insights with heterogeneous treatment effect analysis to derive managerial implications for the e-commerce platform. In particular, we find that for a platform that implements a threshold shipping policy, raising the free shipping threshold may be more effective than raising the shipping fee to reduce the environmental and operational costs associated with faster delivery. In Chapter 2, How Forced Intervention Facilitates Long-term Algorithm Adoption\u27\u27, we investigate whether and why forced interventions can promote algorithm adoption and reduce algorithm aversion in the long term. Data from a leading online education company reveal that sales workers underutilize a new matching algorithm and often use it on low-quality leads. The company conducted a field experiment where sales workers were forced to use or not use the algorithm for three weeks. Experimental results show that forcing workers to use the algorithm during the experiment causally increases their algorithm usage one month after the experiment by 15.8 percentage points. We develop a theoretical model to derive empirical strategies for exploring the mechanisms behind this improvement. Contrary to the traditional literature focusing on habit formation, our findings suggest that learning is a key driver for long-term algorithm adoption among the workers. Specifically, forced algorithm usage allows workers to experience the algorithm\u27s unbiased performance firsthand and positively adjust their beliefs about it. Consequently, after the experiment, the workers use the algorithm not only more frequently but also more on high-quality leads. The study provides empirical evidence that forced intervention can effectively improve long-term algorithm adoption among workers, which is crucial for continuous development of these technologies. More importantly, we demonstrate that forced intervention works by enabling workers to experience an algorithm\u27s unbiased performance and adjust their prior misinformed assumptions about its effectiveness. This suggests that firms can implement extrinsic interventions or educational programs to help workers recognize the benefits of algorithms and develop unbiased beliefs about their capabilities, thus facilitating sustained algorithm usage