Dartmouth Institute for Health Policy and Clinical Practice
Dartmouth Digital Commons (Dartmouth College)Not a member yet
8214 research outputs found
Sort by
Predicting Ad Hoc Categories with Word Embeddings
No full textThe goal of the project is to develop a system which predicts what comes to mind for novel ad hoc categories using large language models. Ad hoc categories, formed in response to situational demands, reflect the flexibility of human thinking. Building on prior research into how word embeddings are situated in human feature spaces and representational techniques of language models, this project aims to investigate if language models can emulate the human process of creating ad hoc categories. Ad hoc category formation operates in a multidimensional feature space, where items with similar scores are clustered along contextually relevant dimensions. As such, word embeddings may model human conceptual representations, which our experiment aims to test. To test this, participants will interact with a web platform to generate members from novel prompts that combine base words and modifiers, for example, “A zoo animal you can bring on the plane”. Responses will be situated in a human-derived feature space and compared to word embeddings created using fastText. Dissimilarity matrices derived from the embedding space and the human derived feature space will be used to compare the structure of relationships between items in each space. This approach will reveal the dimensions most predictive of an ad hoc category and could be generalized to other bases. Findings would have broad implications on the representational potentials of large language models and their alignment with human conceptual organization.https://digitalcommons.dartmouth.edu/wetterhahn_2025/1001/thumbnail.jp
Together We Thrive: A Co-Production Toolkit for People with MS and Care Providers
No full textNearly a million people in the U.S. have Multiple Sclerosis (MS), a progressive neurological disease. People with MS (PwMS), are more likely to have a decreased quality of life (QoL) and are at greater risk for unemployment, loneliness, and depression than people without MS. As MS symptoms can include fatigue, balance issues, mobility problems, and changes to bladder and bowel function, managing MS requires patient engagement that includes coordination of care between neurology, primary care, and specialists. This thesis was researched and written by a PwMS who has worked in collaboration with her neurologist, Dr. Andrew D. Smith III, and a research team gathered through the ICoHN Co-Value study, an ongoing community of practice (established in 2019) that focuses on healthcare co-production: an approach that views healthcare as a service that can be co-designed by the provider of the service and the end-user.
Research shows that people with chronic conditions who can demonstrate patient activation, the skills and confidence that equip patients to become actively engaged in their healthcare, have better health outcomes and experiences. Educational programs that facilitate PwMS engagement with their care and self-management have led to reduced ER visits and hospitalizations. This thesis presents a toolkit that could be used to facilitate PwMS self-management and patient activation in clinical practice through educational activities designed to facilitate healthcare production between PwMS and providers.
The toolkit’s curriculum aligns with the Multiple Sclerosis Self-Management Scale (MSSM-R), an instrument that measures behaviors contributing to PwMS self-management, including MS Knowledge and Information, Healthcare Provider Relationship and Communication, Social/Family Support, Treatment Adherence/Barriers, and Health Maintenance Behaviors. Written for an audience of PwMS, the toolkit includes an educational overview of self-management factors presented in the MSSM-R. and is a prototype of an educational quality improvement initiative that could be tested and administered in clinical practice in conjunction with administration of the MSSM-R.
In addition to the curriculum for self-management, the thesis includes resources for building healthcare literacy, potential opportunities for co-production in practice, and the author’s reflections on her lived experience with MS
Growing Through Simulated Realities: Proxy-Tuning LLMs with Embodied Multi-task planning
Get PDFThis thesis presents a method for teaching large language models (LLMs) to perform embodied multi-task planning using simulation data and a lightweight training technique called proxy-tuning. Traditional LLMs are strong in language understanding but struggle with tasks that require reasoning and planning in real-world environments. To address this, we construct a dataset based on VirtualHome, a simulator that models household activities using input commands. We extract individual tasks from the simulator and combine them into multi-task plans, optimizing for coherence and efficiency. These plans are formatted as Q&A pairs to train models to generate step-by-step instructions that complete multiple tasks at once. Instead of directly fine-tuning large models, which is computationally expensive, we first fine-tune smaller models and then transfer their learned behavior to a larger model using proxy-tuning. This technique adjusts the output of the large model based on the differences between a fine-tuned and an unfine-tuned smaller model, without changing the large model’s parameters
Observed Decreases in Seasonal Drought Across the Northeastern United States
No full textClimate change has the potential to enhance hydrologic and agricultural drought risk in the northeastern United States, yet the Northeast has experienced observed total and extreme precipitation increases. Here, we investigate whether observed regional drought risk has changed over the last century despite well-documented hydrologic intensification. We use the Palmer Drought Severity Index (PDSI) calculated with station data from the Global Historical Climatology Network to assess seasonal drought from 1901 to 2022. We find that Northeast drought across all months is both less frequent and less severe over the last 40 years, decreasing by 59.1% and 18.7% respectively, relative to 1901-1983. Drought frequency decreased by 56.3% and intensity decreased by 16.2% during the warm season, when drought poses the greatest agricultural risks. Reduced hydrologic and agricultural drought risk is largely consistent throughout the Northeast and across stations. The variance in dry and wet events has not significantly changed over time, suggesting that the Northeast hydroclimate is not any more volatile at the monthly timescale than it has been in the past. These findings suggest that precipitation increases to date have outpaced increased evapotranspiration demand from warming, but do not preclude the possibility of future increases in drought risk with additional warming. Collectively, our findings contextualize recent trends in seasonal Northeast drought, finding a marked reduction in seasonal root zone drought and represent a stark contrast with the enhanced regional warm-season surface soil moisture drought risks consistently projected across model ensembles. Our results inform current and future water management decisions, from agricultural to residential water use, as well as efforts to prioritize climate change hazards across the Northeast
Profiling Bile Acids in the Stools of Humans and Animal Models of Cystic Fibrosis
Get PDFCystic fibrosis (CF) is associated with dysbiosis of the gut microbiome, alterations in intestinal mucus production, aberrant bile acid (BA) metabolism, fat malabsorption, and chronic inflammation. As little is known about BAs in CF, we performed both comprehensive and targeted BA profiling in stool of children with or without CF. Our results reveal that select BA species and metabolites are significantly different between children with CF (cwCF) and healthy controls. There is also a trend towards higher primary cBA and total BA levels for cwCF. Matched bacterial metagenomic analyses showed no change in a-diversity between groups in our small cohort, at odds with previous studies, whereas changes in relative abundance of Bacteroides (lower) and E. coli (increased) species is consistent with prior reports. A robust trend was noted toward reduced abundance of bsh gene families (Wilcox test, p = 0.052) – a key rate-limiting enzyme required for bacterial synthesis of secondary BAs – in cwCF. Modest changes in both BAs and microbial BA metabolism-related gene abundances may be attributable to small sample sizes, but also suggest likely combination defects in both host and microbial BA metabolic pathways in cwCF. Importantly, although fecal BA profiles from both ferret and mouse CF models showed significant differences from human BA profiles, only the ferret model reproduced significant differences between CF and nonCF animals, highlighting ferrets as a potentially more appropriate model for studying BA in stool in the context of CF. Together, these results provide new insights into CF-related BA dysmetabolism in cwCF and highlight limitations of CF animal models for BA functional studies
Magnetic reconnection in space: from two-dimensional Magnetohydrodynamic (MHD) simulations to three-dimensional particle-in-cell (PIC) simulations
Get PDFMagnetic reconnection is a physical process in plasmas that converts magnetic energy into plasma kinetic energy and thermal energy. It plays an important role in the generation of solar flares, interactions between the solar wind and Earth\u27s magnetosphere, and magnetic confinement fusion devices. This thesis focuses on studying magnetic reconnection using computational methods, including two-dimensional (2D) magnetohydrodynamics (MHD) simulations to study Petschek-type reconnection, estimating the timescale from a thin current sheet to fully developed reconnection using 2D particle-in-cell (PIC) simulations, and magnetic reconnection X-line spreading in the current direction using three-dimensional (3D) PIC simulations.
This thesis contributes to the understanding of magnetic reconnection physics in space plasmas. In Chapter 2, we found that a resistivity gradient along the outflow direction can generate Petschek-type reconnection in MHD simulations, and the reconnection rate is of the order 0.1 even for a very large exponential resistivity applied. This finding has possible applications in solar spicules formation or plasma thrusters. In Chapter 3, we found that X-line spreading will be affected by drift-kink instability. This finding might be important for modeling energy transport into Earth\u27s magnetosphere via reconnection in Earth\u27s magnetotail. In Chapter 4, a simplified model of the evolution from a thin current sheet to steady-state magnetic reconnection has been derived and compared with 2D PIC simulations. This finding could be important for modeling magnetic reconnection in Earth\u27s magnetotail
Basal autophagy promotes accurate chromosome segregation in Drosophila oocytes and declines during oocyte aging
No full textThe frequency of trisomic pregnancies exponentially increases with advanced maternal age, a phenomenon known as the maternal age effect. The majority of these aneuploid pregnancies arise from meiotic segregation errors in the oocyte and premature loss of sister chromatid cohesion is a contributing factor. However, the mechanisms that lead to premature cohesion loss during oocyte aging are not fully understood.
Using Drosophila melanogaster as a model organism, work described in this dissertation demonstrates that basal levels of autophagy, a degradation process for damaged organelles and protein aggregates, are required during meiotic prophase for arm cohesion maintenance and accurate meiotic segregation in Drosophila oocytes. Interestingly, immunofluorescence analysis indicates that basal autophagy significantly declines when Drosophila oocytes undergo aging, supporting the hypothesis that a decline in autophagy contributes to age-dependent meiotic segregation errors.
Additionally, I provide evidence that prophase-specific knockdown (KD) of Spermidine Synthase (SpdS), an enzyme involved in the synthesis of polyamine spermidine, causes meiotic chromosome segregation errors that are consistent with precocious loss of arm cohesion. Spermidine is known to stimulate autophagy across species and consistently, SpdS KD disrupts basal autophagy in Drosophila oocytes. This result suggests that dietary supplementation of spermidine may stimulate autophagy and suppress age-dependent segregation errors in oocytes. Together, the results presented in this dissertation broaden our understanding of how cellular processes become compromised in aging oocytes and may inform future strategies to improve oocyte quality and the fidelity of meiotic segregation in women of advanced maternal age
DESIGNING FLEXIBLE ELECTRONICS INTEGRATING LIQUID METALS AND LIQUID-METAL DERIVED SURFACE OXIDES
No full textThe pursuit of advanced, flexible, and wearable electronics necessitates innovative approaches to overcome the limitations of traditional materials and methods. This thesis explores a unified approach to liquid metal-based electronics across three domains: stretchable radio frequency sensors, efficient soft magnetic devices, and transparent oxide electronics for biointerfaces. Gallium-based room-temperature liquid metals offer unique advantages in deformability, conductivity, and surface reactivity, enabling innovations across all three areas. First, we introduced a method for high-frequency AC-enhanced resistive sensing using deformable liquid metals to improve low-power detection in wearable electronics. By modulating electromagnetic effects, such as current crowding due to the skin effect, this method can quantitatively distinguish mechanical deformation modes. Additionally, this method can produce an energy-efficient yet high SNR, producing a stretchable mechanical sensor that has been shown in sensing gloves and in a breathing monitoring setup. Second, we present a low-loss, high-Q inductor architecture using 3D-woven liquid metal \u27litz\u27 wires. By mitigating proximity and skin effects, this design enhances RF coil efficiency and offers a pathway toward soft wireless power interfaces. Finally, we demonstrated an automated method to extract and deposit a transparent, ultrathin yet highly conductive liquid metal oxide skin from the In-Tin liquid metal alloy onto flexible substrates. Non-invasive dry bioelectrodes made of transparent oxide films provide superior strain compliance, adhesion, and abrasion resistance compared to standard alternatives. The conductivity and transparency of 2D ITO were used for synchronous, multimodal measurements via electrocardiography (ECG) and pulse plethysmography (PPG). By creating a flexible electronic wearable system, we utilized these bendable and transparent oxide films as bioelectrode material in a multimodal biosignal acquisition setup
Antibody and protein engineering for cancer immunotherapy: advancing targeted and immune modulation therapies
No full textCancer remains one of the most challenging and deadly diseases worldwide due to the complexity of tumor biology and the ability of cancer cells to evade treatment. Despite advancements, cancer remains the second leading cause of death globally. Two major challenges are tumor heterogeneity and the immunosuppressive tumor microenvironment (TME). Tumor heterogeneity, driven by genetic and phenotypic changes, allows cancer cells to lose antigen expression, leading to resistance against single-target therapies. Targeting multiple tumor-specific antigens with antibody cocktails or modulating immune checkpoints that restore immune function in the TME could help overcome these challenges.
In this work, we applied directed evolution in yeast and phage display systems to engineer antibodies from human antibody fragment libraries that are specific to neoepitopes in the mouse B16F10 melanoma and CT26 colon carcinoma models. Many of these antibodies showed binding to tumor cells, elicited effector functions, and inhibited tumor growth in mice. Additionally, we explored the engineering of an immune checkpoint blockade antibody for canine oncology that showed promising results in a canine safety trial. Furthermore, we engineered two anti-PD-L2 checkpoint blockade antibodies with distinct receptor-blocking properties to explore how PD-L2 interacts with its binding partners, PD-1 and RGMB, in the context of tumor immunology.
The results of these studies highlight the feasibility of engineering patient-customized antibody therapies as a possible solution to overcoming tumor heterogeneity as well as offer immune checkpoint blockade strategies and insight for both human and veterinary oncology. Future work should focus on optimizing these therapies and evaluating their clinical potential in both human and canine cancer treatments
ENGINEERING EXTRACELLULAR VESICLES (EVs) TO COMBAT INFECTION AND INFLAMMATION IN CYSTIC FIBROSIS
No full textCystic Fibrosis (CF) is a multisystemic genetic disorder caused by mutations in the CFTR gene, resulting in chronic pulmonary infections, hyperinflammation, and progressive lung damage. Among the pathogens that colonize CF lungs, Pseudomonas aeruginosa (P. aeruginosa) is predominant, infecting over 50% of adults with CF and developing antibiotic resistance over time. Current modulator therapies fail to address persistent bacterial infections, chronic inflammation, or irreversible lung damage, underscoring the need for novel therapeutic strategies. In our previous in vitro studies, we demonstrated that extracellular vesicles (EVs) secreted by primary Human Bronchial Epithelial Cells (HBECs) inhibit P. aeruginosa growth by reducing the abundance of biofilm-associated proteins and disrupting antibiotic resistance pathways through the delivery of microRNA let-7b-5p. We expanded on these findings in our current work to show that these HBEC-derived EVs also reduce P. aeruginosa lung burden, attenuate inflammation, and lower neutrophil infiltration in a CF mouse model. To further address potential immunogenicity associated with HBEC-derived EVs, we engineered EVs from Mesenchymal Stromal Cells (MSCs), to contain the microRNA let-7b-5p. In vivo, these engineered EVs significantly reduced bacterial burden and inflammation in CF mice. Using in vitro systems, we further discovered that one of the mechanisms by which let-7b-5p loaded MSC EVs act in vivo involves reducing the inflammatory response of primary CF bronchial epithelial cells to Pseudomonas aeruginosa and inhibiting its ability to form biofilms on these cells. Our work highlights a novel dual anti-infective and anti-inflammatory therapeutic strategy for CF. This platform holds significant promise for optimizing delivery strategies and expanding its clinical applicability to target additional respiratory pathogens