22689 research outputs found
Sort by
ADAPTIVE MONDRIAN PARTITIONING AND GRADIENT DIMENSION REDUCTION FOR HIGH-DIMENSIONAL CLASSIFICATION
Get PDFThe increasing prevalence of high-dimensional data in genomics, image processing, and other fields necessitate methods for variable classification and dimensionality reduction. However, traditional approaches treat these tasks separately and usually overlook complex nonlinear relationships within the data. We extend a Transformed Iterative Mondrian (TRIM) algorithm for classification purposes with gradient-based techniques to handle high-dimensional data. The TRIM algorithm selects and transforms features by applying a Mondrian process—a hierarchical random partitioning of the input space—and iteratively refines the feature subspace based on predictive importance. To further enhance this process, we introduce the expected Jacobian outer product (EJOP), a gradient-based measure that highlights how variations in each direction affect the model’s output. We are left with the question of interpretability of these cuts in the algorithm, and to understand these directions better we introduce a simple sparsity threshold on the EJOP that preserves only the most salient variables.
The final model performs dimension reduction and classification. Empirical studies on synthetic data show that the sparse EJOP-TRIM classifier matches the accuracy of random forests and other gradient-covariation methods that involve optimization. The result is a computationally efficient route to sparse dimension reduction in modern high-dimension, large-sample settings
IMAGING BRAIN FUNCTION AND CEREBRAL BLOOD AND LYMPHATIC VESSELS USING ADVANCED MRI TECHNIQUES
Get PDFBlood oxygenation level-dependent (BOLD) functional MRI (fMRI) has played a crucial role in brain research, enabling the investigation of neural activity by detecting changes in blood oxygenation. This dissertation is aimed to develop novel MRI approaches to study brain function and the physiological parameters underlying the BOLD fMRI signal. First, we assessed the recently developed T2-prepared (T2prep) BOLD fMRI approach in event-related functional tasks. The T2prep BOLD fMRI method can provide fMRI images with minimal susceptibility artifacts, improving signal detection in regions prone to signal dropouts and distortion. The hemodynamic response functions (HRFs) measured using T2prep BOLD fMRI are compared with the widely used gradient-echo (GRE) and spin-echo (SE) echo-planar imaging (EPI) BOLD fMRI methods in human subjects.
Next, since blood vessels—especially pial arteries and veins—are the primary sources of fMRI signals, we developed novel MRI methods to enhancing their visualization in the human brain. Iron-based MRI contrast agents have shown promise in this regard. Our work showed that Iron Dextran can be used as an MRI contrast agent to enhance the visualization of small blood vessels using multi-echo susceptibility-weighted imaging (SWI) at 7T, enabling a more detailed examination of the cerebrovascular network.
Finally, we developed new MRI methods to study CSF circulation in the brain, another important aspect related to brain function and physiology. Recent discoveries have identified dural lymphatic vessels that are hypothesized to facilitate CSF drainage to peripheral lymph nodes. Using Gadolinium-based contrast agent (GBCA) based MRI methods, dynamic distribution of GBCAs from dural blood vessels to nearby fluid spaces were analyzed. The high spatial resolution in our methods allows us to measure the time courses of blood and CSF signal changes in two separate layers of the dura mater in the meninges
Stem Cell-Driving Signals Activate Cell-Intrinsic Mesenchymal and Immunosuppressive Mechanisms via TGFBR2 in Glioblastoma
Get PDFHigh-grade gliomas, including glioblastoma (GBM), are highly heterogeneous with a complex oncogenic microenvironment consisting of distinct tumor niches and remarkable cellular variability. A critical component of GBM malignancy derives from the distinct population of multipotent and tumor-propagating glioma stem-like cells (GSCs), which maintain the vast and diverse cell landscape implicated in therapeutic inefficacy. Notably, attempts to activate an anti-tumor immune response in GBM have been met with many challenges due to its inherently immunosuppressive tumor microenvironment, particularly in mesenchymal-like tumors. The degree and mechanisms by which molecularly and phenotypically diverse GSCs contribute to this state are poorly defined. In this study, we describe a mechanism by which stem cell-driving events coordinate the transition to a mesenchymal-like GSC state through activation of TGFBR2 signaling. Furthermore, our multifaceted approach combining bioinformatics analyses of clinical and experimental datasets, single-cell sequencing, and molecular and pharmacologic manipulation of patient-derived cells identified GSCs expressing immunosuppressive effectors mimicking regulatory T cells (Tregs). We show that this Immunosuppressive Treg-Like (ITL) GSC state is specific to the mesenchymal GSC subset and is associated with and driven specifically by TGFb type II receptor (TGFBR2) in contrast to TGFBR1. Transgenic TGFBR2 expression in patient-derived GBM neurospheres promoted a mesenchymal transition and induced a 6-gene ITL signature consisting of CD274 (PD-L1), NT5E (CD73), ENTPD1 (CD39), LGALS1 (galectin-1), PDCD1LG2 (PD-L2), and TGFB1. This TGFBR2-driven ITL signature was identified in clinical GBM specimens, patient-derived GSCs and systemic mesenchymal malignancies. TGFBR2High GSCs inhibited CD4+ and CD8+ T cell viability and their capacity to kill GBM cells, effects reversed by pharmacologic and shRNA-based TGFBR2 inhibition. Collectively, our data identify a mesenchymal-like, immunosuppressive GSC state that is TGFBR2-dependent and susceptible to TGFBR2-targeted therapeutics. The impact of TGFBR2 inhibition on the anti-tumor immune response and the effects of combining TGFBR2-targeted treatment with current immunotherapy against GBM remain unknown. Nonetheless, our in vitro findings suggest that alternative treatment options to precisely target TGFBR2 should be pursued and prioritized moving forward
Patient Education Visuals Communicating the Role of Salpingectomy in the Primary Prevention of Ovarian Cancer
Get PDFEpithelial ovarian cancer is one of the most lethal gynecologic malignancies, affecting 1 in 80 women, with fewer than 50% surviving five years after diagnosis. Approximately 90% of cases originate in the fallopian tubes (Lheureux et al., 2019). In 2015, the American College of Obstetricians and Gynecologists (ACOG) recommended bilateral salpingectomy as the standard of care for ovarian cancer prevention during surgical sterilization and hysterectomy (Committee on Gynecologic Practice 2015). Despite this, patient awareness of salpingectomy and its preventive benefits remains limited. Existing reproductive health and pregnancy prevention resources fail to distinguish between, or even include, salpingectomy and tubal ligation, contributing to confusion and missed opportunities for cancer prevention.
For the first time, salpingectomy-centered education was integrated into a comprehensive suite of multimedia resources, framed through patient perspective and guided by the lens of ovarian cancer prevention rather than pregnancy prevention alone. These materials were designed for patient use, with particular attention to visual clarity, accessible language, and cultural responsiveness. Development was informed by clinical shadowing, IRB-approved workshop participation, and collaboration with OB/GYN providers, legal advisors, and patient education specialists. All resources were translated into Spanish to expand accessibility. The prototypes developed include: 1) a 2D animation (ninety seconds) explaining the link between contraception and ovarian cancer risk; 2) three versions of comprehensive posters about contraception, juxtaposing pregnancy prevention and ovarian cancer risk reduction; 3) a trifold brochure contrasting salpingectomy with tubal ligation; 4) a Unity-based interactive module with manipulable 3D models of female reproductive anatomy; 5) a plain-language glossary with a simplified definition and pronunciation guide for forty relevant terms; and 6) the first procedure-specific surgical consent forms for salpingectomy and tubal ligation, approved for implementation at Johns Hopkins Hospital.
Developed with iterative feedback from patients and providers, these educational resources offer a scalable model for improving informed consent and lifelong reproductive health education. Future work will involve formal user feedback to assess impact and to support continued refinement. Together, these resources represent a novel, patient-centered advancement in ovarian cancer education. They offer opportunities to create actionable awareness about the benefits of salpingectomy for ovarian cancer prevention for all
AI's Energy Appetite: Lessons from Leading Hyperscalers on Sustainable Procurement
Get PDFThe rapid expansion of computation demand for artificial intelligence (AI) has created unprecedented push for data centers and the energy they use. In particular, this is driven by large hyperscalers, such as Amazon Web Services (AWS), Google, Meta, and Microsoft, who have also made commitments to achieve net zero emissions. The exponential growth of AI demand in the past few years is not only putting strain on electric grids but also on these clean energy commitments. This capstone paper examines the sustainable energy procurement strategies of each of the four aforementioned hyperscalers to assess the pros and cons of various approaches.
Drawing on case studies and industry reports, this research examines: how hyperscalers procure clean energy, potential roadblocks to sustainable procurement, and possible pathways forward. The findings reveal AI energy demand is leading to tremendous growth in renewable procurement but also new approaches such as investments in nuclear, emerging technologies, and behind-the-meter solutions that were not previously pursued by hyperscalers. This paper concludes an all of the above approach to clean energy procurement will be essential for meeting tomorrow’s AI demands and that this can be better facilitated with long-term financial commitments, enhanced collaboration, and policy reforms
Biomechanical Characterization of Pneumodissection in Deep Anterior Lamellar Keratoplasty
Get PDFDeep Anterior Lamellar Keratoplasty (DALK) is a partial-thickness corneal transplant technique that preserves the recipient’s Descemet’s membrane (DM) and endothelium, reducing the risk of endothelial rejection compared to full-thickness penetrating keratoplasty (PK). However, consistently achieving a successful dissection to the DM is technically challenging. The most common method—“big-bubble” (BB) pneumodissection—involves deep stromal needle insertion and air injection to cleave the posterior stroma from the DM. However, bubble formation is unreliable, and failure to generate a BB or rupture of DM often necessitates conversion to PK.
This thesis investigates the biomechanical factors that influence BB success. We hypothesize that tissue adhesion between stromal lamellae and DM governs the variability in bubble formation. Two experimental approaches were used: (1) Peel tests to quantify the adhesion strength between posterior stroma and DM, and (2) Big-bubble air injection tests to measure the pressure required to initiate and propagate bubbles under controlled conditions.
In the peel tests, bovine corneas were partially dissected and peeled using a motorized mechanical testing setup. Tension was recorded with a load cell, and interfacial toughness (Gc) was calculated. Results showed considerable variability in peel strength across samples, consistent with prior human studies. Peel strength was measured in grams per millimeter and converted to energy release rate values in J/m².
In the air injection tests, a robotic syringe pump delivered air through a vertically inserted needle to simulate clinical BB formation. Real-time pressure and imaging were used to detect bubble onset and growth. Type I BBs formed in most samples, and initial bubble pressures were lower than those reported in literature, likely due to larger injected air volumes in our setup.
By correlating the peel strength and injection pressure using fracture mechanics models, we estimated the critical pressure (pcrit) required for delamination and found reasonable agreement with measured pressures. These findings suggest that interfacial toughness is a major determinant of BB formation.
This work provides new biomechanical insight into the pneumodissection process and may inform improved surgical strategies and robotic systems for consistent DALK performance. Keywords: cornea transplant, DALK, big-bubble, biomechanics, peel test, pneumodissection
OPTIMIZATION OF ASPERGILLUS NIGER FEED USING A GENOME-SCALE METABOLIC MODEL TO PRODUCE METAL-LEACHING ORGANIC ACIDS
No full textThe increase in use of lithium-ion batteries has caused large amounts of spent battery
waste, the majority of which isn’t recycled and ends up in landfills or incinerators.1 Current
recycling methods, hydrometallurgy and pyrometallurgy, are not environmentally friendly,
requiring high amounts of energy or producing toxic waste streams.2 An alternative,
biohydrometallurgy, has emerged as a more sustainable form of hydrometallurgy, using
microbes to produce organic acids that leach spent battery metals in solution. Aspergillus
niger is a target microbe for bioleaching systems due to its ability to produce high titers of
target acids such as citric acid.
To make bioleaching more economically viable as a recycling method, production of the
microbes needs to be as efficient as possible. This work uses a genome-scale metabolic
model of Aspergillus niger to develop a fed-batch protocol of A. niger bioleaching cultures,
reducing nutrient use. Using calculated growth and glucose uptake rates from a
staggered-harvest experiment, a 4-day fed-batch protocol was developed from model
predictions of minimum nitrate uptake rates. This protocol resulted in no change in yield
of biomass per gram of glucose. The fed-batch protocol was extended past 4 days until
the glucose had been completely consumed, requiring an additional 3 days. The extended
fed-batch achieved the same dry weight and biomass yield per gram of glucose as the
batch cultures with 86% less nitrate. In addition, the extended fed-batch had no effect on
acid titer or yield per gram of glucose. Further work needs to be done to improve the
HPLC method used to better quantify and discern acid production
APPLYING PROTEIN ENGINEERING TO INVESTIGATE IMMUNE BIOLOGY AND ENABLE FACILE MANUFACTURING OF ANTIBODY-DRUG CONJUGATES
No full textRecombinant proteins are widely used as tools to explore fundamental questions in immunology and as drugs to treat various diseases. Despite their success in both areas, recombinant proteins have limitations due to their natural origins, such as pleiotropy, off-target effects, and manufacturing challenges. Here, molecular engineering strategies are used to address these issues by creating more specific and easily manufactured proteins.
In one example, rational engineering of the interleukin-4 (IL-4)/IL-13 system was applied for therapeutic immune modulation. Natural IL-4 and IL-13 cytokines, along with a hyper-stable IL-4 mimetic called Neo-4, were fused to mouse-serum albumin (MSA) to study IL-4 receptor signaling in vivo. These molecules showed that selective signaling through the type I IL-4 receptor with Neo-4 leads to tempered type 2 inflammation and different lung macrophage phenotypes compared to natural IL-4. Preliminary experiments indicated these molecules could speed up inflammation resolution in mouse models of acute lung injury. These fusion proteins are valuable tools for studying IL-4 and IL-13 biology and designing new therapies.
In a second example, IL-21 cytokine mimetics were engineered with restricted B cell signaling. These were engineered by reducing affinity and targeting B cells via fusion to an anti-CD19 monoclonal antibody. These constructs showed B-cell specific IL-21 signaling in primary mouse splenocytes and are currently used to study IL-21's effects on B cell fate in germinal centers in animal models.
In a third example, a platform for manufacturing antibody conjugates was developed by integrating metabolic glycoengineering and the insertion of N-linked glycosylation sites into the fragment crystallizable (Fc) region of human immunoglobulin G1. This platform supports multiple types of antibody conjugates, including fluorescent dye conjugation, cytotoxic drug conjugation, and biomaterial conjugation. The extent of conjugation was highly tunable using different production cell lines and various combinations of engineered sites, achieves conjugation levels comparable to state-of-the-art techniques. This glycoengineering platform provides a facile, efficient, and versatile method for generating antibody conjugates.
The work detailed herein describes the use of molecular engineering techniques to design protein products and molecular platforms for investigating fundamental biology as well as for the development of therapeutic drugs
UNDERSTANDING CUMULATIVE EXPOSURE TO FOOD ADDITIVES AND POTENTIAL IMPACTS ON THYROID HORMONE CONCENTRATIONS
No full textThyroid hormones regulate growth and metabolism and are vital for fetal brain development. Adverse neurodevelopmental outcomes in children have been linked to small changes in thyroxine (T4) levels during pregnancy. Hundreds of compounds in the food supply may affect thyroid hormone synthesis, but little is known about how combined exposures may affect thyroid function. We aimed to address this knowledge gap by investigating the cumulative burden of exposure to food additives with modes of action that may affect thyroid hormone concentrations. Understanding the overall impact of food additives on thyroid function, regardless of their source, is crucial to evaluating the safety of new food chemicals. We examined exposure to 15 compounds used as food additives and thyroxine concentrations in non-pregnant reproductive age women using the National Health and Nutrition Examination Survey (NHANES).
In aim 1, we characterized exposure to perchlorate, nitrate, thiocyanate, iodine, iron, vitamins A, D, E, quercetin, resveratrol, caffeine, aspartame, isoflavones, lutein and zeaxanthin, and epigallocatechin-3-gallate in women by race and iodine status. We identified potential vulnerabilities in non-Hispanic black women, and found significantly higher exposures to thyroid disruptors in iodine adequate women. We also identified weaknesses in the estimation of food additive exposure.
In aim 2, we sought to improve methods for food additive exposure by developing a novel method combining NHANES with NielsenIQ market share and the USDA Branded Foods ingredient databases, using aspartame as a case study. We found mean aspartame intakes to be lower, but the prevalence of exposure to be higher than previously published estimates.
In aim 3, we evaluated associations between the mixture and T4 concentrations in women. Overall associations between joint exposure to all food chemicals assessed and T4 concentrations were negative. We found associations to decrease in magnitude as phytochemicals were removed from the exposure set, indicating the potential for cumulative effects. We also observed nonmonotonic exposure-response curves for perchlorate and nitrate. These findings highlight the importance of examining cumulative exposures to thyroid disrupting chemicals, accounting for vulnerabilities due to low nutrient consumption
EFFECTS OF SHIFTING CYSTIC FIBROSIS THERAPIES ON SURVIVAL AND CYSTIC FIBROSIS-RELATED DIABETES
No full textBackground: Cystic Fibrosis transmembrane conductance regulator (CFTR) modulator therapies revolutionized treatment for people with cystic fibrosis (CF), but their effect on death and CF-related diabetes diagnosis has not been fully described. Furthermore, the changing therapeutic landscape complicates the estimation of survival for people living with CF, who have received very different clinical management according to their birth cohorts. Differences between observed survival curves for people with CF and those predicted using the period approach have not been quantified. Therefore, this dissertation sought to address these gaps by assessing the effects of shifting cystic fibrosis therapies on survival and development of CF-related diabetes.
Methods: For all aims we used the US CF Foundation patient registry data containing longitudinal demographic, clinical, and therapeutic data for over 55,000 people with CF living in the US from 1986 through 2023. In aim 1, we estimated the effectiveness of any CFTR modulator therapy initiation (compared to no initiation) on risk of death over 8-years. In aim 2, we estimated the effectiveness of triple combination CFTR modulator therapy initiation (compared to no initiation) on risk of CF-related diabetes diagnosis over 3-years. In aim 3, we quantified differences between observed and predicted (with the period approach) death risk curves for 3-year birth cohorts of people with CF.
Results: Initiation of any CFTR modulator therapy reduces the 8-year risk of death by 7%. Initiation of triple combination CFTR modulator therapy reduces the 3-year risk of CF-related diabetes by 2.5%. The period approach consistently overestimated the risk of death for past birth cohorts, with differences increasing over subsequent birth cohorts and older ages.
Conclusions: CFTR modulator therapies reduce the risk of death and of CF-related diabetes diagnosis for people living with CF, and therefore further research is needed to extend similar therapies to all people with CF. When clinicians and patients are planning for their futures, it is important for them to understand the limitations of period survival predictions and to consider the potential impact of therapeutic advancements