22689 research outputs found
Sort by
HMGA1 Chromatin Regulators Repress MHC II Antigen Presentation in MPN
No full textMyeloproliferative neoplasms (MPN) are at risk for transformation to highly lethal acute myeloid leukemia (AML), although targetable mechanisms to prevent progression remain elusive. Here, we unveil a novel role for High Mobility Group A1 (HMGA1) chromatin regulators as drivers of immune escape by disrupting three-dimensional genome architecture to repress antigen presentation (AP) genes at the major histocompatibility complex class II (MHC-II) locus. HMGA1 is up-regulated with MPN progression to AML and required for leukemic transformation in preclinical models. Integrating multi-omic approaches (RNA/ChIP/ATAC sequencing) reveal that HMGA1 directly represses genes involved in AP, with greatest effects on MHC-II. Mechanistically, HMGA1 binds throughout the MHC-II locus, recruiting repressive histones to compact chromatin and disrupt chromatin complexes comprised of the CCCTC binding factor (CTCF) and cohesin, thereby preventing CTCF-mediated chromatin looping and activation of MHC-II immune attack genes. HMGA1 also disrupts binding of CIITA, a transcriptional activator of MHC-II promoters. Functionally, HMGA1 depletion enhances T cell-mediated recognition and killing of MPN AML cells. Using artificial intelligence, we predicted that histone deacetylase inhibitors (HDACi) target HMGA1 networks. Strikingly, treatment with the HDACi, entinostat, recapitulates effects of HMGA1 silencing by activating MHC-II immune attack genes. HMGA1 silencing enhances effects of entinostat to induce MHC-II genes on MPN AML cells. Most importantly, HMGA1-mediated immune escape pathways are activated in MPN patients after transformation to leukemia. Together, these findings illuminate HMGA1 as an epigenetic key that locks down MHC-II and a promising therapeutic target to unlock immune attack genes and intercept disease progression
Robot Learning with Efficient Representation and Domain Adaptation
No full textRobot learning, integrating techniques from machine learning, computer vision, and robotics, has emerged as a promising field for developing robotic systems capable of performing a wide range of real-world tasks. In building successful learning-based system in robotics as in other domains, data plays a pivotal role. Despite significant achievement in collecting and training on large-scale robotic datasets, the real-world data acquired with physical hardware remains limited, particularly when compared to vast internet-scale image and text datasets. Consequently, a primary challenge for the robot learning community is constructing data-efficient systems that learn from scarce real-world data for deployment on physical hardware. This is particularly demanding for vision-based manipulation problems, where observations are high-dimensional, and tasks are complex.
This thesis addresses this challenge through two main themes: first, employing efficient visual representations to enhance learning efficiency; and second, leveraging abundant, low-cost simulation data for policy learning, followed by domain adaptation using minimal real-world data.
The first part of this thesis presents methods utilizing efficient visual representations to improve learning efficiency. In the "virtual in-hand eye transformer", we propose using virtual in-hand views instead of raw camera views, significantly enhancing performance when learning from a small number of demonstrations. Through self-supervision, in "proportional derivative controllable embedding", we learn embeddings from raw images that can be controlled with a simple proportional derivative-controller; in "keypoint-conditioned neural radiance field", we discover a set of keypoints of the scene that can be efficiently used by model predictive controller.
In the second part of this thesis, we study the problem of visual domain adaptation for sim-to-real transfer. Directly applying policies learned from simulation to real world would result in deteriorated performance because of the differences in appearances and physics. To bridge such domain gap between simulated and real environments, we propose leveraging exploratory experiences in the deployment environment to bridge the domain gap under domain adaptation settings.
To summarize, we propose multiple methods that learn visual policy with limited deployment domain data. Extensive experiments, both simulated and with real physical systems, validate our methods' superiorit
Hyperparameter Optimization for Neural Machine Translation Systems
Get PDFMachine translation, a sequence-to-sequence task, involves translating text from one language to another. Currently, transformer-based systems dominate the field of neural machine translation (NMT). To optimize these systems, various critical decisions regarding architecture design and training processes must be made—these decisions are the hyperparameters of the system. Typically, these hyperparameters are set before training begins and remain unchanged until convergence. Traditionally, they are tuned manually based on intuition, heuristics, previous studies, or default settings provided in open-source frameworks. This approach often leads to suboptimal exploration of the hyperparameter space, which can cause exaggerated performance differences and potentially misleading conclusions. Despite the proliferation of hyperparameter optimization (HPO) methods under the umbrella of Automated Machine Learning (AutoML), their effectiveness in NMT has not been thoroughly evaluated, primarily due to the significant computational demands of NMT models and their vast hyperparameter search spaces. This challenge is further complicated by the need to optimize multiple objectives simultaneously, such as translation accuracy and decoding speed.
This thesis addresses these challenges by conducting a comprehensive study of HPO specifically within the context of NMT. First, we introduce a benchmark dataset employing a ``table-lookup" based benchmarking procedure, designed to promote reproducible research in HPO for NMT. Second, we propose a novel HPO algorithm using graph-based optimization, which flexibly incorporates prior knowledge about hyperparameters. Third, we develop a post-hoc interpretation framework to better understand the significance and interrelationships of individual hyperparameters. Fourth, we evaluate the efficacy of a multi-fidelity HPO method, successive halving, and propose best practices for its application in NMT and large language models. Finally, this work includes the creation of an HPO toolkit tailored for NMT research, designed to streamline the experimental process and allow researchers to concentrate on innovation instead of the mundane
LEARNING FOR THE SAKE OF HEAVEN: LEARNING, RELIGIOSITY, AND COMMUNITY IN EARLY MODERN ASHKENAZI SOCIETY
No full text“Learning for the Sake of Heaven: Learning, Religiosity, and Community in Early Modern Ashkenazi Society” examines the structure and organization of early modern Ashkenazi educational institutions. This study illuminates the vital role learning and learning institutions had in consolidating Ashkenazi society in Central Europe between the 1580s and 1750s.
The dissertation presents a comprehensive analysis of Ashkenazi learning culture, focusing on the social history of ḥederim and yeshivot and the complex history of their administration. Additionally, it investigates the impact of broader historical developments on these institutions. By introducing a new comparative method for studying Ashkenazi communal records (pinkasim), the study demonstrates that Ashkenazi communities, from Metz in the west to Lviv in the east, shared a common educational framework. This educational system differed considerably from contemporary Confessionalized pedagogical principles in its aims and goals. Early modern Ashkenazi society prioritized "learning for the sake of heaven" (lernen) over conventional educational objectives. Yet, this system was incompatible with some Jewish laws and, in certain instances, even transgressed them. Thus, by distinguishing between lay administration and rabbinic sources, this research elucidates a notion of an independent Ashkenazi civil tradition that was distinct from rabbinical views. Exploring a fundamental chapter in the history of Jewish culture, this study offers a reevaluation of the period scholars have identified as the genesis of Jewish modernity. In doing so, the dissertation contributes to the historiography of Central European Jewry and intersects with the broader fields of history of culture, religion, knowledge, and education in early modern European societies
Efficacy Study of Filo in Jefferson County Schools
Get PDFThis quasi-experimental study evaluated the impact of the Filo tutoring program on math, ELA, and science achievement among Grades K-8 students in Jefferson County (AL) Schools (JCS). Using a student-level matched comparison design (N=1,089 treatment, 2,163 comparison), the study examined student performance on ACAP and i-Ready assessments. Results indicated directionally positive impacts on ACAP math, ELA, and science scores, with statistically significant gains in ACAP math observed for male and Black students—a notable outcome given the historical achievement gaps within the district. Although no significant effects emerged for i-Ready scores, a positive association was found between the number of Filo tutoring sessions and ACAP ELA performance
MITOCHONDRIAL MORPHODYNAMICS ARE MODULATED BY PHYSIOLOGICAL RANGE OF TEMPERATURE AND INFLUENCE HOST CELL OUTCOMES DURING INFLUENZA INFECTION
Get PDFInfluenza viruses replicate in both the cooler, upper portions of the airway and the warmer, lower portions of the respiratory tract. These are two distinct environments in the respiratory tract with the upper portion of the airway maintaining an average temperature of 33°C and the lower airways maintain an average temperature of 37°C. Prior research shows that infections at these two temperatures results in differences in virus production and host immune responses but it is not understood how environmental temperature mediates these differences. This research investigates how physiological ranges of temperature, specifically 33°C and 37°C, impact host cell biology and how temperature-dependent differences in host cells influence outcomes during Influenza A Virus infection. This study prioritizes describing mitochondrial networks via immunofluorescence, live cell microscopy, and cellular functional assays due to their importance in maintaining cellular homeostasis and mediating immune responses to viral infection. We find that the temperature at which cells are incubated significantly influences
mitochondrial network morphology and mitochondrial function. Moreover, temperature-dependent changes to mitochondrial networks prior to infection result in temperature-specific changes to host cell outcomes during infection. These findings indicate that mitochondrial structure alone can modulate host cell outcomes during viral infection and that both the form and function of mitochondria directly impact Influenza A Virus production. While not all mitochondrial processes were shown to be affected by temperature or infection, these results
highlight the importance of using physiologically relevant temperatures in respiratory pathogen
research and elucidate how mitochondrial dynamics contribute to host cell outcomes during
Influenza A Virus infection.
Respiratory viruses infect the upper and lower respiratory tract but rarely is the impact of
physiological ranges of temperature (33°C to 37°C) considered. Mitochondria are central
mediators of numerous physiological pathways, and their functions are often modified by virus
infection. Physiological ranges of temperature can alter mitochondrial form and function, which
encompasses aspects of the length, width, connectedness, and movement of mitochondria as well as their signaling capacity, interorganelle interactions, and functional outputs like ATP
production and oxygen consumption. Furthermore, the temperature-dependent differences
detailed in this research are further impacted by virus infection resulting in temperature-dependent, infection-specific phenotypes. Ultimately, this study sheds light on how temperature
can impact mitochondrial form and function in concert with virus infection to influence
outcomes of influenza infection
CARCINOGENIC INDUSTRIAL AIR POLLUTION AND RISK OF KIDNEY CANCER: A POOLED ANALYSIS OF 535,000 AMERICANS
Get PDFOur study investigated associations between industrial air emissions of known and probable carcinogens and kidney cancer risk using data from two large U.S. prospective cohorts (NIH-AARP and PLCO). After a median follow-up of 22 years among 535,007 participants, researchers observed 5,590 kidney cancer cases. Exposure to industrial air emissions was assessed using EPA's Toxics Release Inventory data and inverse distance-weighted annual emissions indices. Several chemicals showed significant associations with kidney cancer risk, particularly formaldehyde, vinyl chloride, pentachlorophenol, and arsenic at 5km and 10km distances. Sex-stratified analyses revealed stronger associations among women for certain chemicals, including cadmium and lead. There was also evidence suggesting stronger associations with arsenic among non-Hispanic Black participants compared to non-Hispanic White participants. This study extends previous occupational findings to the general population and identifies potential environmental risk factors for kidney cancer, with implications for addressing cancer disparities across populations
THERMODYNAMICS OF BIVALENT HETEROTRIMER ASSOCIATION IN THE NOTCH SIGNALING PATHWAY
Get PDFThe Notch signaling pathway regulates cellular differentiation by activating transcription through a heterotrimer comprising the Notch receptor’s intracellular domain (NICD), the DNA-binding protein CSL, and the coactivator MAML. NICD has two binding sites for CSL, a short motif in the RAM region and an ankyrin domain (ANK), connected by an intrinsically disordered linker which form a bivalent ternary complex with CSL and MAML. Although bivalency is required for maximal transcription activation, the energetic contributions of bivalency within the transcription activation complex are unknown.
In Chapter 2, I first use isothermal titration calorimetry to measure binding of the CSL-ANK-MAML heterotrimer, for which I develop an obligate heterotrimer model. By comparing this heterotrimerization reaction with binding reactions involving different regions of RAMANK, I determine the energetic contribution of bivalency. I show that bivalency through the disordered linker increases the effective concentration of ANK, and that the bivalent interaction enhances occupancy of RAM and ANK at their binding sites on CSL by about three orders of magnitude.
In Chapter 3, I further investigate CSL-RAMANK-MAML assembly using destabilizing substitutions at the RAM-CSL and ANK-CSL binding interfaces. I find that interface destabilization affects the energetics of bivalency, although the mechanism is unclear.
A previous study showed that CSL occupancy of its target DNA increases after Notch activation, suggesting assembly of CSL-RAMANK-MAML increases the affinity of CSL for DNA. In Chapter 4, I investigate whether the presence of DNA affects assembly of CSL-RAMANK-MAML, and find negligible effect.
Linear repeat proteins are useful in protein folding studies, because they can be analyzed using a nearest-neighbor 1D Ising model to determine the intrinsic and interfacial stability of repeats. It was previously observed that naturally-occurring repeat proteins with intrinsically stabilizing repeats have at least 58 residues per repeat, but the 43 to 56 residue repeats of de novo Designed Helical Repeat (DHR) proteins are intrinsically stable. To determine whether the high stability of DHRs results from their Rosetta-based design or their repeat length, in Chapter 5 I perform Ising analysis on shorter DHRs. I find that shorter DHRs, like shorter naturally-occurring repeat proteins, have intrinsically unstable repeats
Theoretical and Deep Learning Approaches to Predict and Improve Surgical Tool Tip Tracking during Robot-Assisted Photoacoustic-Guided Interventions
No full textModern interventional procedures (e.g., biopsy, catheterization) require real-time tracking and visualization of important structures within the body. Photoacoustic imaging has emerged as a promising real-time solution that augments the tips of interventional devices (e.g., needles, catheters) with optical fibers, enabling modeling of these device tips as acoustic sources. However, photoacoustic imaging performance depends on multiple factors (e.g., laser energy, source dimensions, background noise), limiting the ability to predict achievable system performance. In addition, photoacoustic images reconstructed using delay-and-sum (DAS) beamforming suffer from inherent information loss, which limits optimal tracking performance. Furthermore, information from the out-of-plane dimension of a two-dimensional image acquired with a one-dimensional array is necessary to accurately track surgical tool tips in three dimensions.
This dissertation presents two theoretical frameworks and novel deep learning approaches to address the limitations noted above, including: (1) theory to connect system performance based on a recently introduced metric (i.e., generalized contrast-to-noise ratio) to system parameters (e.g., laser energy, source dimensions, background noise) and predict system performance with DAS images; (2) theory relating three-dimensional source locations and sound speeds to the shapes of the corresponding waveforms in two-dimensional photoacoustic channel data (prior to any beamforming); and (3) deep learning object detection-based systems, instance segmentation-based approaches, and associated image processing techniques to localize photoacoustic sources in two and three dimensions from two dimensional channel data. The deep learning-based systems were integrated with robotic control to autonomously find and track needle and catheter tips in phantom, ex vivo, and in vivo settings, including the first known application of force control to photoacoustic visual servoing. Results have the potential to transform medical fields, such as cardiology and hepatology, with photoacoustic-guided interventional systems that will increase global access to quality healthcare
Applications of Clinical Pharmacology to the Study of Emerging Psychoactive Substances
No full textChapter I: Emerging psychoactive substances have poorly-understood human exposure-response relationships. Applied clinical pharmacology tools can help understand the health outcomes associated with emerging psychoactive substance use.
Chapter II: Delta-8 tetrahydrocannabinol (Δ-8 THC) is considered less psychoactive than Δ-9 THC. A population pharmacokinetic analysis and model was developed from two clinical trials (total N=23) comparing Δ-8 THC (10mg, 20mg, 40mg), Δ-9 THC (20mg), and placebo after oral and vaporized administration. Despite a similar pharmacokinetic pattern, Δ-8 THC was less metabolized to its more active metabolite than Δ-9 THC. Participant inhalation behaviors influenced the relative bioavailability of vaped Δ-8 THC compared to oral, with decreasing bioavailability at increasing doses captured by the model using covariates from measured puff topography. The validated model supports future analyses to compare Δ-8 and Δ-9 THC effects.
Chapter III: The pharmacokinetics of fentanyl are well-studied, but the clinical phenomena associated with illicitly manufactured fentanyl exposure are understudied in persons who use drugs. This narrative review applies fentanyl’s known absorption, distribution, metabolism, and excretion properties to illicit fentanyl. A notable gap in research exists due to differences between medicinal fentanyl and illicit fentanyl use patterns and composition. However, its properties suggest sustained, supratherapeutic use lends to the peripheral accumulation of fentanyl, leading to prolonged elimination in persons who use drugs.
Chapter IV: The pharmacology of fentanyl and xylazine is not characterized in persons regularly exposed to illicit fentanyl. This case series presents individual-level urine pharmacokinetics of fentanyl, its metabolite norfentanyl, and xylazine in persons with opioid use disorder. Participants (N=11) provided urine samples (n=95) for quantitative analysis. Urinary half-life ranges were: fentanyl 1.5-28.7 hours, norfentanyl 5.2-27.4 hours, xylazine 1.1-25.2 hours. Predicted detection window ranges were: fentanyl 13.0-130.9 hours (0.5-5.5 days), norfentanyl 64.8-255.9 hours (2.7-10.7 days), xylazine 13.8-123.4 hours (0.6-5.1 days). Individuals with a high magnitude of drug exposure are likely to evidence an extended duration of urinary detection.
Chapter V: Pharmacokinetic and pharmacokinetic-pharmacodynamic modeling and simulation techniques are promising translational tools to study the exposure-response relationship of emerging psychoactive substances. Potential uses of these techniques are discussed, including clinical applications to inform drug testing guidelines using real-world data