22689 research outputs found
Sort by
ZOOPLANKTON ABUNDANCE IN THE CONTEXT OF CLIMATE CHANGE: COMPARING ARCTIC AND NORTH PACIFIC BIOVOLUME FROM 1953 TO 1962
Get PDFDespite their ecological importance, knowledge of zooplankton community dynamics is limited. This study aims to fill the gaps in the scientific understanding of zooplankton abundance in the past to better protect marine ecosystems in the face of climate change. Although zooplankton research is over a century old, published literature includes few in-depth analyses of data from the twentieth century. This research focuses on zooplankton abundance from two marginal seas of the Arctic Ocean and the California Current System within the North Pacific Ocean over a decade (1953-1962) to draw conclusions about the history of these organisms. To strengthen future population predictions and drive effective marine policy, this research answers the following question: How did zooplankton abundance in the Arctic and North Pacific Oceans respond to El Niño-Southern Oscillation phases between 1953 and 1962, and what can this tell us about zooplankton resiliency in different oceanic regions as climate change persists? Biovolume data extracted from the Biological Atlas of the Arctic Seas and California Cooperative Oceanic Fisheries Investigation was analyzed using RStudio and Microsoft Excel. Results showed that Arctic zooplankton were more vulnerable to climatic changes than populations in the California Current. Changes in biovolume between La Niña and El Niño events were drastic in Arctic zooplankton. For both regions, abundance was greater during La Niña, whereas El Niño appeared to limit population size. Similar seasonal trends were strong throughout the timescale. Arctic abundance decreased over time during La Niña and increased over the El Niño event. In the Arctic, productivity is linked to ocean mixing due to increased freshwater caused by melting sea ice and terrestrial glaciers. Increased abundance in this region during La Niña may have resulted from a delay in Arctic zooplankton response to changing sea surface temperatures, or specific atmospheric conditions that reduce sea ice in high latitudes during multi-year La Niña events. California Current abundance was more predictable and closely linked with upwelling as a mechanism stimulating productivity. Cold, nutrient-rich waters driven by La Niña increased upwelling and nutrient availability, promoting zooplankton abundance. Warm El Niño waters weakened upwelling within the California Current, thus decreasing zooplankton abundance. Overall, zooplankton within this region are more resilient to climate change. This study reinforces the importance of regional understanding of zooplankton populations, along with a need for continued long-term monitoring as climate change persists
Statistical Methods for Acute Myeloid Leukemia Research: From Cell to Population
No full textAcute Myeloid Leukemia (AML) is a rare cancer that originates in the bone marrow. It is a deadly cancer with a five-year survival rate estimated to be around 28%. In this thesis, we compare, develop, and apply statistical methods to further understand heterogeneity in the disease and treatment response of AML, at both cellular and population levels.
Measurable residual disease (MRD) is a biomarker for patients with AML in complete remission that is associated with clinical outcomes. Flow cytometry-based MRD detection has been widely adopted in clinical setting, with the limitation of uncontrolled variability in measurement and interpretation across institutions. We develop an algorithm to enhance standardization of MRD assignment. It achieves high concordance with manual gating to identify patients with positive MRD status.
Next-generation sequencing (NGS) technologies are becoming more prevalent in MRD detection because they provide flexibility to cover more aberrant types while generating consistent and reproducible results. We validate the clinical utility of NGS for MRD targets on multiple AML disease related genes in Pre-MEASURE, a multi-center retrospective observational study supported by Center for International Blood and Marrow Transplant Research.
The Pre-MEASURE study reveals that conditioning intensity, a preparative treatment given to patients before transplant, can modify the MRD effect. To quantify the degree of heterogeneous treatment effects considering all patient characteristics, we apply five commonly-used semi-parametric and machine learning methods to our data, comparing patients given high intensity versus low intensity conditioning. However, the results are quantitatively and qualitatively different among methods, which would imply different clinical strategies for treatment recommendation. To identify the preferred method in our research setting, we design and implement a simulation study by creating "data" and "causal" neighborhoods to evaluate the performance of competing methods for data like ours. The work establishes a framework for analyzing heterogeneous treatment effect in conditioning intensity.
For patients who have refractory or relapsed AML that could not achieve complete remission, we led a pilot study using a novel combination of anti-PD-1 and hypomethylating therapy (NCT02996474). Longitudinal samples collected throughout the treatment course are analyzed using diverse single-cell experimental and computational approaches to elucidate temporal cellular dynamics that could relate with clinical response. We also develop a spatial statistical approach to quantify the distance change between T cells and leukemic cells across times.
The findings and methodologies presented in this thesis advance our ability to analyze and interpret the increasingly rich data in AML research, contributing to the goal of improving patient outcomes through more personalized treatment approaches
An Open Simulation Platform for Team Training in Robotic Surgery
Get PDFThe predominant paradigm for robotic minimally-invasive surgery places the main surgeon at a console, teleoperating robotic instruments inside the patient, with an assistant surgeon providing support at the bedside.
We present an open source simulation platform for training this robotic surgery team that integrates the console from a da Vinci Research Kit (dVRK) for the main surgeon, a haptic device (repurposed from an existing simulator) for the assistant surgeon, and a simulation environment built on the Asynchronous Multi-Body Framework (AMBF) to provide a common training environment. Our system emulates a realistic surgical scenario in which the dVRK console enables the surgeon to control the patient-side manipulators while the haptic device provides both control and tactile feedback for the first assistant (FA) operating a virtual laparoscopic grasper.
A user study with 7 teams (14 subjects total), each performing a retraction and suturing task with both a physical system and our proposed simulator, demonstrates that our simulation platform effectively replicates key aspects of team-based surgical training
INVESTIGATING THERAPEUTIC STRATEGIES FOR PREVENTION OF PATHOLOGICAL RESPONSE TO GROWTH PLATE INJURY
No full textInjuries to the epiphyseal growth plate (physis) in pediatric patients often result in the formation of a bony bridge (bony bar) that disrupts normal endochondral ossification, potentially leading to limb length discrepancies and angular deformities. Current clinical interventions, such as bar resection and fat grafting, offer limited efficacy and are associated with recurrence and incomplete restoration of growth. This thesis investigates the neurobiological mechanisms underlying pathological ossification following growth plate injury, with a focus on the role of sensory neurons and nerve growth factor (NGF)–TrkA signaling.
We developed a murine drill-hole injury model targeting the distal femoral growth plate. This model reliably produces a reproducible bony bar, enabling quantitative assessments of injury outcomes using a newly established scoring system and micro–computed tomography (µCT). Utilizing a chemical-genetic TrkAF592A mouse model and the small molecule inhibitor 1NMPP1, we selectively inhibited TrkA signaling and evaluated its effects on neurovascular invasion, osteogenic differentiation, and bony bar formation.
Our results demonstrate that sensory neurons are active regulators of pathological growth plate healing. Growth plate injury induces robust ingrowth of TrkA+ sensory fibers and associated vascularization, driving ectopic osteogenesis. Inhibiting TrkA signaling significantly reduced nerve and vessel infiltration, suppressed osteoblastic activity, and preserved cartilaginous architecture at the injury site. Transcriptomic analyses further revealed that TrkA inhibition repressed pro-osteogenic pathways while maintaining chondrogenic gene expression signatures.
These findings identify NGF-TrkA signaling as a critical mediator of neurovascular-driven ossification and suggest that its targeted inhibition may serve as a promising therapeutic strategy to prevent growth plate bridging. The work not only provides mechanistic insight into nerve-skeletal interactions during pathological repair but also establishes a preclinical foundation for future therapies aimed at improving outcomes in pediatric growth plate injuries
Understanding the Impact of Culturally Adapted Health Education on Breast and Cervical Cancer Screening Among American Indian and Alaska Native Women: A Scoping Review
Get PDFBackground:
American Indian and Alaska Native (AI/AN) women continue to experience a disproportionate burden of breast and cervical cancer, often facing delayed diagnoses and limited access to timely care. These inequities are compounded by cultural, geographic, and systemic challenges, highlighting the critical need for screening interventions that are responsive to the cultural contexts of AI/AN communities.
Objective:
This scoping review explores how culturally adapted educational strategies have been used to promote cancer screening and follow-up among AI/AN women, with a focus on identifying effective approaches and persistent barriers.
Methods:
A comprehensive search of PubMed, Embase, and the Cochrane Library was conducted to identify studies published up to February 7, 2025. Studies were eligible if they were conducted in the United States, involved American Indian or Alaska Native (AI/AN) women, addressed breast and/or cervical cancer screening interventions, employed any study design except literature reviews, were published in English, and explicitly incorporated culturally tailored educational elements. Data was then extracted on study design, intervention type, use of cultural tailored elements—like community-based participatory research (CBPR), home-based education, use of Indigenous languages, or storytelling—as well as reported outcomes. Implementation challenges related to recruitment, geographic isolation, small sample sizes, and structural barriers were also captured and summarized.
Results:
A diverse range of culturally aligned strategies was employed across studies, including storytelling, talking circles, video-based education, mobile apps, brochures, and home visits. Most interventions reported improvements in knowledge, self-efficacy, and reported screening behavior. While breast cancer-focused programs often incorporated digital and tech-based tools, cervical cancer initiatives more frequently utilized interpersonal and community-based formats. Programs that involved lay health workers and tribal leadership were more successful in fostering trust and engagement. Persistent barriers included transportation limitations, lack of culturally appropriate materials, fear or stigma surrounding cancer, and low health literacy. Several studies emphasized the value of collaborative design and community ownership as keys to sustainability and impact.
Conclusion:
Educational interventions that are grounded in cultural relevance and developed in partnership with AI/AN communities can meaningfully enhance cancer screening rates and health outcomes. Future programs should build on these insights, prioritizing culturally safe, flexible, and community-led approaches to close screening gaps and improve care equity
Transferable and Scalable Vision-Language Models for Reasoning and Grounding
No full textVision-language pre-trained (VLP) systems have demonstrated strong performance by learning transferable visual representations through language supervision. These models redefine the training paradigm for large visual systems by leveraging weakly paired image- and video-text data, eliminating the need for manually curated annotations. This dissertation presents a comprehensive study of vision-language models across diverse downstream domains, focusing on their efficiency, scalability, and generalizability.
Firstly, we introduce compute efficient methods for image-language and egocentric video-language pre-training by integrating multimodal fusion directly into unimodal backbones, requiring fewer cross-modal parameters than traditional fusion-specific transformer layers. Moreover, the proposed methods allow switching between dual and fusion encoders, enabling us to use the same pre-trained weights for various image and video understanding and grounding tasks. Next, we explore large-scale training of multimodal large language models (MLLMs) for fine-grained image and video understanding. We begin with VistaLLM, a unified framework that addresses coarse- and fine-grained vision language tasks across single and multiple photos. VistaLLM is followed by ED-VTG, a two-stage approach for fine-grained temporal grounding in untrimmed videos that enriches underspecified queries using strong captioning systems to enhance grounding accuracy. We leverage strong captioning systems to transform underspecified language queries into enriched sentences, adding missing details and cues to facilitate grounding. These enhanced queries are then localized using a lightweight decoder. To address noise and minimize hallucinations, our model employs a multiple instance learning (MIL) objective, which dynamically selects the most suitable version of the query for each training sample. Finally, we examine the capacity of current MLLMs to interpret complex visual content such as figures and tables in scientific literature. Collectively, these contributions provide a holistic investigation of vision-language pre-training and highlight its critical role in building scalable, adaptable, and transferable multimodal systems
STRUCTURAL DISSECTION OF AMPA RECEPTORS IN HUMAN DISEASE
Get PDFExcitatory neurotransmission is the process by which neurons communicate with each other via the neurotransmitter glutamate (Glu) to propagate signal transduction. When one neuron releases Glu, specialized receptors, the AMPA-subtype ionotropic glutamate receptors (AMPARs), on the post-synaptic neuron bind Glu and depolarize the neuron. The function of AMPARs is tightly regulated in neurons by transmembrane AMPA receptor regulatory proteins (TARPs), and their dysfunction is implicated in a wide range of neurological and neurodevelopmental diseases, in addition to glioblastoma. Despite how fundamental AMPARs are for health, targeting them therapeutically has remained challenging, which is reflected by only a single-FDA approved drug that target AMPARs. I am particularly interested in uncovering the molecular and structural underpinnings that orchestrate AMPAR function to provide a foundation for precision therapeutic design. To this end, during my PhD work, I discovered: 1) how a neurodevelopmental mutation affects ion selectivity and permeability, 2) how this mutation enables repurposing of an FDA-approved drug to attenuate the effects of the mutation, and 3) the molecular and structural motifs for how TARPs tune AMPAR function
COMPARISON OF REPRODUCTIVE HORMONES IN ADOLESCENT AND ADULT POPULATIONS WITH EXPOSURE TO BISPHENOL A AND PHTHALATE ESTERS FROM EPIDEMIOLOGICAL AND OCCUPATIONAL STUDIES
Get PDFBisphenol A (BPA) and phthalate esters are synthetic endocrine-disrupting chemicals (EDCs) widely present in consumer and industrial products. This study investigates the impact of BPA and phthalate exposure on reproductive hormones across key life stages—from in utero development and adolescence to adulthood—through a systematic review of epidemiological and occupational studies. Findings indicate that prenatal exposure to BPA and phthalates is associated with altered reproductive hormone levels in children during puberty, including elevated testosterone in females and estradiol in males. In adult populations, particularly workers in industrial settings, BPA exposure was linked to abnormal levels of prolactin and estradiol in males. Unfortunately, research on occupational phthalate exposure remains limited. The study underscores significant sex-specific differences in hormonal outcomes and highlights critical research gaps, especially concerning occupational exposures in females and hormone-specific data. These findings support the need for stricter regulations on BPA and phthalates in both consumer products and workplace environments. Policy recommendations include continuing to ban these chemicals for products marketed toward children, enhancing workplace safety measures, and further investigation for BPA alternatives and health effects from phthalates in adult populations
SYNTHESIS AND PROCESSING OF ULTRA-LARGE PROTEIN BIOMATERIALS DIRECTLY FROM BACTERIA VIA A COUPLED LOOPABLE-TRANSLATION SECRETION SYSTEM
No full textUltra-large proteins, particularly those with fibrous or repetitive architectures, offer extraordinary mechanical and structural properties valuable for biomaterials. However, their recombinant synthesis remains a persistent challenge due to the instability of repetitive DNA sequences, difficulty in codon optimization, high metabolic burden, and cellular toxicity caused by intracellular accumulation. These limitations have hindered the scalable production of proteinaceous materials at the mega-Dalton scale.
To overcome these obstacles, we developed a platform that couples circular mRNA-based loopable translation with co-translational protein secretion to enable autonomous biosynthesis and extracellular assembly of ultra-large polypeptides. As a foundational step, we adopted a permuted intron-exon (PIE) strategy using a group I self-splicing intron from T4 bacteriophage to generate circular mRNA in vivo. This circular architecture allows ribosomes to continuously translate an open reading frame without termination, effectively minimizing construct length while enabling the synthesis of ultra-high molecular weight proteins. Although this system enables loopable translation of model proteins green fluorescence protein, we found that fibrous protein sequences impose unique barriers to both expression and circularization.
To allow repetitive fibrous to be continuously exported for living assembly, we engineered a co-translational secretion system in Bacillus subtilis using the TasA_ss/SipW orthogonal signal peptide and peptidase pair. This system not only enables efficient secretion of fibrous proteins via the SecYEG translocon but also promotes their spontaneous assembly into nanofibers on the membrane surface, a process we define as secretion-catalyzed assembly (SCA). To overcome circularization inefficiencies caused by the high GC content and structured nature of fibrous mRNAs, we developed a combined strategy involving computational synonymous codon optimization and trans-acting RNA folding factors. This approach raised circularization efficiency from less than 10% to over 90% across a diverse set of fibrous sequences.
With optimized loopable translation, elastin-like polypeptides can be loopable-translated exceeding 3.8 MDa—surpassing the size of titin, the largest known natural protein. These proteins were further processed into robust fibers and transparent films with superior mechanical properties. We eventually integrate loopable translation and co-translational secretion to achieve living secretion of ultra-large fibrous proteins. Our work establishes a generalizable platform for microbial production of ultra-large proteins, paving the way for scalable, living assembly of programmable biomaterials
Designing Conversation Experience: From Traditional to LLM-Powered Voice Assistants
Get PDFVoice assistants (VAs) are widely integrated into consumer technologies such as mobile phones, smartwatches, smart speakers, and in-car systems, but they remain limited in their ability to support natural, adaptive, and personalized conversations. These limitations are apparent in interactions with older adults, who face unique usability challenges as technology is often not designed with them in mind. With the improvements in language technologies, there is an opportunity to reimagine how VAs support diverse users through richer, adaptable conversation experiences.
In this dissertation, I investigate how to design conversation experiences that improve usability, enrich user engagement, and offer personalized support as voice assistants transition from traditional rule-based systems to generative models such as large language models (LLMs). My approach spans both understanding user challenges and designing new interactions. First, I examine how older adults experience breakdowns with traditional VAs by conducting a month-long, in-home deployment, capturing real-world interaction data situated in their daily lives. I then explore trust repair as a design goal by studying how mitigation strategies and voice gender affect user perceptions of VAs, especially during error scenarios. As LLMs become more accessible, I turn to investigating how these generative models can be integrated into VAs and how this integration fundamentally alters the nature of interaction. Unlike traditional systems, LLM-powered VAs enable context-aware, multi-turn conversations and exhibit greater flexibility in handling ambiguous or complex speech. I built a ChatGPT-powered VA and conducted a lab study across diverse scenarios, followed by a field deployment with older adults to evaluate how the LLM-powered VA supports more fluid and adaptable interaction. Finally, I apply this understanding to a real-world application by co-designing and developing an LLM-powered VA to provide personalized and meaningful assistance for older adults in their health self-management.
Through these studies, I provide empirical insights and design guidelines for creating next-generation conversational agents that move beyond command-response interactions toward adaptive, user-aligned, and socially capable conversational systems. This dissertation contributes to our understanding of how LLMs can be integrated into everyday voice interfaces and how their capabilities can be shaped through user-centered design to better support diverse users—especially aging populations