22689 research outputs found
Sort by
Low-Power On-Device Deep Learning and Functional Imaging to Personalize Vagus Nerve Stimulation Therapy
No full textEpilepsy is a neurological disorder involving spontaneous seizure activity, and vagus nerve stimulation (VNS) was the first and is still a widely used neuromodulation therapy to reduce seizure frequency in refractory epilepsy patients. Current VNS administration applies fixed stimulation schedules using static parameters found using rudimentary trial-and-error methods, which can take months to years to optimize in individual patients. Many hypothesize that targeted stimulation in advance of or response to seizure activity may lead to greater seizure reduction efficacy, however several clinical trials using heart rate-based algorithms have struggled to show improvement beyond current methods. Based on the success of cortical responsive neurostimulation in focal epilepsy patients, I aim to bring cortical recording modalities to VNS devices to improve therapy for a broader range of patients, including children. To enable investigation of cortical recording modalities with VNS in animal models of epilepsy, I have developed an embedded implantable device capable of low-power on-device Deep Learning processing for rapid seizure prediction and responsive VNS mitigation of seizure activity, and rapid-feedback methods to optimize cortical activation driven by VNS.
First, a new deep learning architecture called the Time Scale Network is presented, which simultaneously and efficiently incorporates short-term and long-term features in a shallow neural network. The network provides competitive seizure prediction accuracy with a focus on minimizing computational requirements, demonstrating over 90% average seizure prediction accuracy with as few as 1,133 parameters and significantly fewer operations required for inference compared to similar popular models. These shallow neural networks are then translated to run on a low-power microcontroller platform enabling real-time recording, real-time processing, and real-time responsive stimulation to mitigate incoming seizure activity.
Further, this work also develops methodology to rapidly optimize cortical activation using a non-contact label-free holographic optical imaging system to measure cortical activation dynamics driven by VNS. This system is used to characterize effects of stimulation parameter adjustments in cortical neural responses, enabling the rapid tuning of stimulation parameters in several minutes compared to months or years. This work has enabled future chronic animal studies aimed towards optimizing VNS therapy using novel machine learning and optical imaging technologies
EXPLORING THE INTERPLAY OF HEAVY METAL TOXICITY AND SPECIFIC GENETIC BACKGROUND IN AUTISM SPECTRUM DISORDER USING A 3D BRAIN MODEL
No full textThe attention towards understanding the etiology and mechanisms of neurodevelopmental disorders has significantly increased in recent years. It has been observed that exposure to environmental chemicals and toxicants can have adverse health effects on developing nervous system. One hypothesis gaining recognition is the potential contribution of these exposures to the rising prevalence of neurodevelopmental disorders, such as autism spectrum disorder (ASD). ASD is a major public health concern, with a significant increase in prevalence in the United States over the past decade. Heavy metals, as environmental toxicants, are implicated in impairing neurodevelopment through multiple pathways. While animal models have been extensively used to study neurodevelopmental disorders and their phenotypic outcomes, they have limitations in predicting human health due to the interspecies differences and ethical concerns. Hence, there is a need for more human-relevant models to better represent the complexities of these disorders. In this study, we propose the use of brain organoid microphysiological systems (bMPS) as an advanced alternative model. We focused on Chromium, Cadmium, and a mixture of both as target environmental chemicals to investigate how they perturbate neurodevelopment in ASD. Specifically, brain organoids were generated from induced pluripotent stem cell (iPSC) lines derived from male patients with 16p11.2 deletion associated with varying ASD severity and age matched typically developed individual. An additional female line was used as an additional control. These brain organoids were differentiated and were exposed to the two heavy metals Chromium and Cadmium and their mixtures during critical period of brain development, from 4 to 8 weeks (long-term low dose) or 72 hours at week 8 (acute exposure). By evaluating and analyzing neuronal differentiation, synaptic connectivity, and functional neuronal activity at the cellular and molecular level, we obtained preliminary findings that shed light on the complex interactions between heavy metal exposure and neural differentiation of iPS cells with and without 16.p11.2 deletion and with and without autistic genetic backgrounds. Further mechanistic research is needed to understand the link between exposures and ASD and gene-environment interaction (GxE) in the context of ASD
EXPLORING PROGRAMMATICALLY OPTIMIZED UTILITY SOLAR PV SITE SELECTION IN TEXAS
Get PDFThis study delves into the programmatic optimization of utility-scale solar site selection in Texas, strongly emphasizing integrating geographical, economic, and environmental considerations. Texas, with its expansive landscapes and high solar irradiance, offers a unique platform for solar energy projects. However, the challenge lies in efficiently selecting sites that maximize energy production while accounting for localized conditions. This research endeavors to develop a model that not only streamlines the site selection process but also incorporates the multifaceted criteria crucial for the success of utility-scale solar farms. It aims to provide a valuable tool for renewable policymakers, investors, developers, and regulatory planners. The topic of this study is of personal interest and closely aligns with my professional experience in evaluating solar projects across the United States.
During this investigation, much was learned about the nuances of Texas energy markets and why solar energy has recently surged in the region. This study details the various model inputs and discusses the research from which these methods are derived. Then, a discussion and analysis of the critical model findings will be presented, detailing which regions of Texas have the highest financial potential for new utility-scale solar photovoltaic (PV). Additionally, the advantages and disadvantages of programmatic analysis will be highlighted to answer how programmatic optimization can aid in utility-scale site selection. As the parameters for solar site selection are numerous, this discussion will focus on items seen as the most consequential for project viability. Furthermore, as the number of analyses that could be done with this methodology is also plentiful, only a few key findings will be highlighted. The codebase for this project is available via this link: https://github.com/will-kable/SolarSite
Investigating Protein Kinase Activation through the Hippo Pathway Kinase MST2
No full textThe Hippo pathway regulates cell growth by balancing proliferation and apoptosis. The pathway has been implicated in wound healing, development, and maintenance of the stem cell niche with dysregulation of the pathway being involved in carcinomas and sarcomas. At the core of the Hippo pathway are two serine/threonine kinases, MST1/2 and LATS1/2, and the two accessory proteins, SAV1 and MOB1A/B. This core kinase cassette controls the activity of the co-transcriptional activators YAP and TAZ through phosphorylation resulting in cytoplasmic retention and degradation. Diverse stimuli, from cell polarity and mechanotransduction to energy sensing and cellular stress, can activate the Hippo pathway. Mechanistically how such a diverse range of stimuli can activate the Hippo pathway has remained an unanswered question.
In these studies, we investigate the molecular mechanisms regulating MST1/2 activation and suggest broader applications of these mechanisms. MST1/2 activation occurs through trans-autophosphorylation where two MST1/2 molecules phosphorylate one another. Biologically, MST1/2 activation is regulated through homo- or heterodimerization with SAV1 and RASSF via the SARAH domain, but how these interactions promote pathway activation was unclear. We determined the biophysical parameters using purified SARAH domains from Hippo, the Drosophila melanogaster homolog to MST1/2, dRassF, and Salvador. We found the interactions are exclusively dimeric with Hippo:Salvador having the highest affinity. Based on affinity, MST1/2 and SAV1 form constitutive dimers and SAV1 homodimerizes to form a tetramer increasing the local concentration of MST1/2 leading to MST1/2 activation. As such, we investigated if dimerization or an increase of MST1/2 concentration was sufficient for activation. First, we wondered if a specific interface contributed to the kinase domain interaction. We identified the αG-helix mediated kinase domain interactions as substitutions of the αG-helix prevented dimerization and autophosphorylation but did not disrupt catalytic activity or protein fold suggesting the αG-helix mediates a dimer critical for trans-autophosphorylation. Finally, we wondered if other kinases activated by trans-autophosphorylation utilized the αG-helix. Upon literature review and analysis, we found the αG-helix mediated trans-autophosphorylation across the kinome and substitutions within the αG-helix decreased kinase activation. We identified the mechanism governing MST1/2 activation and a potential role for the αG-helix mediating trans-autophosphorylation of protein kinases
A MULTI-TIME-SCALE WALL MODEL FOR LARGE EDDY SIMULATION OF NON-EQUILIBRIUM FLOWS
No full textThe prohibitive cost of resolving near-wall flow features has led to the usage of wall models for Large Eddy Simulation (LES) of wall bounded turbulent flows. Wall models typically rely on simplified equations to model, as opposed to resolve, the underlying physics such that the computational overhead is reduced. The equilibrium wall model (EQWM) remains the most popular wall model due to its simplicity and relatively good performance over many flows. However it is conceptually only valid for equilibrium flows while many wall-bounded turbulent flows occur far from equilibrium. This motivates developing the Lagrangian relaxation towards equilibrium (LaRTE) model, a new pathway for non-equilibrium wall modeling. The LaRTE model utilizes the unsteady RANS equations and a momentum integral approach to isolate quasi-equilibrium wall-stress dynamics from non-equilibrium responses to time- and spatial-varying LES inputs. Non-equilibrium physics can then be modeled separately, such as (1) the laminar Stokes layers that form in the viscous region and generate rapid wall-stress responses to fast changes in the pressure gradient or (2) turbulent velocity fluctuations and how they correlate with wall-stress fluctuations. The total modeled wall-stress thus includes contributions from various processes operating at different time scales (i.e., the LaRTE quasi-equilibrium plus laminar and turbulent non-equilibrium wall-stresses) and is called the multi-time-scale wall model (MTSWM).
The MTSWM is applied in LES of turbulent flows with both temporal and spatial non-equilibrium. Wall-parallel homogeneous flows tested include canonical stationary channel flow, channel flow with a sudden spanwise pressure gradient (SSPG), and pulsating and linearly accelerating channel flow for several forcing frequencies and acceleration rates, respectively. Streamwise developing flows tested include the canonical zero-pressure-gradient (ZPG) flat plate developing boundary layer over a wide range of Reynolds numbers and a separated boundary layer flow induced by a suction and blowing transpiration boundary condition. Results obtained with the MTSWM show improvement, relative to the EQWM, for flows with high temporal non-equilibrium while still showing good agreement with direct numerical simulation data for canonical flows (or flows with weak non-equilibrium) where the EQWM performance is also good. Both the MTSWM and EQWM show similar good performance for the separated flow case but more testing for spatial non-equilibrium flows is needed
Charity in the Crescent City: Constructing, Experiencing, and Remembering Modern Public Hospital Medicine In New Orleans 1880-1950
No full textCharity in the Crescent City traces the late 19th and early 20th century history of Charity Hospital in New Orleans, one of the oldest continually operated public hospitals in North America prior to its closure in Hurricane Katrina. This project occupies an intersection between two complementary historiographic claims; first that the history of the American public hospital should be reevaluated from the perspective of the US South, and second, that the history of medicine in the US South is incomplete without the history of the American public hospital. My project builds on the work of historians of institutional care, medicine, and New Orleans to ask: what does the Charity Hospital archive reveal about the performance and regulation of race, class, and gender in late 19th and early 20th century New Orleans?
In order to explore this question, I theorize the hospital as an archive of daily life in New Orleans. I trace lived experience at Charity through surgical case notes, administrative documents, architectural blueprints, periodical sources, audiovisual recordings, memoirs, and medical photographs.
From the 1880s to the 1950s, Charity Hospital had to answer to new standards of modern American medicine, the changing racial, spatial, and political landscape of the city, and the rise of accidents, risk, and crisis threatening to the social order. Charity’s built environment and policies both reinforced and questioned race and class based social hierarchies in Jim Crow New Orleans. Similarly, while the form of Charity Hospital’s patient records became increasingly dismissive of individual patient narratives, their content often reveals new insights into everyday life and embodied experience in turn of the century New Orleans. Read against the grain, the granular detail of the Charity Hospital archive reveals a rich history of daily life, bedside care, and the often ambiguous relationship between race, class, power, and public hospital medicine in clinical encounters of the modern era
METHODS FOR AUTOMATED ANALYSIS OF RETINAL OCT AND OCTA IMAGES
No full textOptical coherence tomography (OCT) is a non-invasive imaging technique widely used for the high-resolution, cross-sectional imaging of the human retina. These images distinctly reveal retinal layers. Optical coherence tomography angiography (OCTA) employs multiple repeat acquisitions of OCT scans to capture motion contrast, primarily from moving blood cells. It allows for a detailed visualization of the retinal vascular plexus and the foveal avascular zone without the need for dye injection. The thickness of retinal layers and the vessel density within the retinal vasculature serve as important biomarkers in various studies. Consequently, quantitative analyses using OCT and OCTA often use automated segmentation of retinal layers in OCT and vessels in OCTA images.
In this thesis, we first focus on the automated segmentation of OCT retinal layers. Despite the distinct layered topology of human retinal layers, most algorithms neglect this unique structure yielding outputs with topology defects that are immediately recognizable by human observers. We proposed layer boundary evolution, a novel OCT retinal layer segmentation algorithm that ensures topologically correct outputs with a low computational overhead.
We next focused on automated OCTA vessel segmentation. In light of the advancements in deep learning and the constraints posed by limited manual labels for OCTA images, we propose two deep learning-based algorithms for efficient segmentation with reduced dependency on manual labels. We first introduce variational intensity cross channel encoder, an unsupervised segmentation algorithm that uses repeated, unlabeled OCTA scans of the same eye to enhance robustness and accuracy of the segmentation. To further leverage any available manual labels, we present ACRROSS, a novel algorithm that uses disentangled representation learning to segments OCTA images in a semi-supervised framework.
Accurate registration of OCTA images is essential for understanding and monitoring ocular diseases. The process of aligning vessels across successive scans facilitates the computation of biomarkers within consistent regions of interest. Precisely tracking these biomarkers provides a foundation for gaining deeper insights into disease progression and the effectiveness of treatment responses. To this end, we proposed VFAttention, an innovative deep learning algorithm with differentiable feature matching and location retrieval for the efficient learning of deformable transformations. VFAttention is capable of both 2D and 3D deformable registration. We have showcased its applicability across a range of medical registration tasks, including retinal OCTA, brain MRI, and lung CT images
Influence Of Clinical-Community Relationships On Birth Outcomes: The Role Of Community Context
No full textBackground: Rooted in the delivery system framework of community health centers is a commitment to collaboration between clinical and community teams, or clinical-community relationships, to improve patient health outcomes. Maternal health outcomes are one area where the benefits of these collaborative relationships have been described to improve patient outcomes, with health centers also being cited as venues to improve birth outcomes. This research considers the influence of direct and indirect factors of clinical-community relationships on low birthweight percent in health centers by 1) dissecting the contextual components that influence these cross-sector networks and 2) exploring the impact of these environmental features on low birthweight percent.
Methods: Manuscript 1 describes themes generated from semi-structured interviews to examine the historical and social context, and policy environment, that shapes current clinical-community relationships in health centers by county. Manuscript 2 uses latent class analysis and regression to characterize the population health priorities and community context that might influence clinical-community relationships across jurisdictions. Manuscript 3 uses a multilevel linear regression is used to examine the influence of health center activities, as well as population health priorities and performance categories, on birth outcomes (i.e., low birthweight).
Results: The three manuscripts in this dissertation display how the impact of collaborative efforts can be hindered by the complex nature of organizational constraints. Interviews revealed five key themes influencing how organizations interact – community context, mission alignment, organizational behavior and collaboration structure, resources, and competition. Furthermore, local public health delivery systems with a network of support, but without community-specific plans based on priority health needs, can struggle to prioritize and perform population health activities. Results show that 0% of the variability in low birthweight percent is attributable to between-delivery system differences, despite delivery system approaches being statistically significantly associated with county-level prenatal exposures.
Conclusions: This research suggests that despite delivery-system approaches and health center efforts, prenatal exposures and possibly other mechanisms still affect low birthweight rates in health centers. Results from this dissertation reveal that there is still much work to be done to understand how policy efforts can better equip organizations to respond to the complexity of the environments that surround them
Development of Novel Inducible CRISPR tools and the Application to Study DNA Damage Responses
Get PDFCRISPR (Clustered Regularly Interspaced Palindromic Repeats) is powerful in genome and transcriptome editing, providing potential treatment for genetic diseases. However, the swift molecular responses post-CRISPR editing remain elusive. In this thesis, I applied and improved a light-responsive very fast CRISPR (vfCRISPR) technique to induce fast, efficient, and synchronized genome and transcriptome editing in human cells. I found that complementarity between guide RNA (gRNA) and target RNA within the direct repeat (DR)-distal region is required for CasRx nuclease activity, while DR-proximal region is important for PspCas13b. Using vfCRISPR/Cas9, I investigated the molecular responses to DNA double-strand break (DSB), including DNA damage repair (DDR) and transcriptional dynamics. I applied MS2 technology to monitor transcription in live cells and chromatin immunoprecipitation (ChIP) to measure the RNA polymerase II (RNAP2) occupancy after inducing DSB with vfCRISPR at a specific locus. My finding showed that a single Cas9 cleavage rapidly represses transcription of the damaged gene within minutes, which coincides with the recruitment of damage repair protein 53BP1. Notably, transcription repression propagates bi-directionally along the genome from DSB for hundreds of kilo-bases in an attenuated manner. Different from existing literature, the single DSB-induced transcription repression was independent of PRC1-mediated H2A K119 mono ubiquitination (H2AK119ub). Instead, the proteasome was evoked to remove elongating RNAP2. In summary, by developing and applying light-stimulated vfCRISPR to DNA damage response, my research revealed the rapid and propagating transcription repression kinetics after DSB and uncovered the important role of proteasome-dependent RNAP2 turnover in this process
Tropomyosin receptor kinase B signaling in the lateral septum critically regulates social behavior
Get PDFSocial behaviors are a critical component to the survival of various species. They take many forms, such as the biological extremes of species like the carpenter ant where social isolation will severely reduced life expectancy, to something as simple as sharing food with a friend, which humans do readily, but it’s a behavior seen in many species like vampire bats, crows, and naked mole rats. While the benefits of pro-social interactions are easy to understand, the neural underpinnings that enable us and drive us to engage in these behaviors are highly complex and not well understood.
The lateral septum (LS) is a basal forebrain region that is a node for social behaviors. In mice, LS regulates social saliency, social recognition, and social aggression. LS also controls social behavior in guinea pigs, prairie voles, songbirds, cats, and humans. Social dysfunction in psychiatric disorders, such as autism and schizophrenia, is prevalent. While heterogenous, these disorders have documented deficits in social saliency and recognition.
Mice are highly social and display reliable responses to novel conspecifics. While rodent models cannot capture the nuance of human behavior, they offer genetic and molecular tools to study circuitry that enables social recognition. In this dissertation, we show that basolateral amygdala (BLA) inputs to LS are necessary for social novelty recognition. Then we established that expression of brain-derived neurotrophic factor (BDNF) in BLA projections and expression of its cognate receptor tropomyosin kinase B (TrkB) in LS neurons are necessary for social novelty recognition.
Next we created a genomic atlas LS cell types using single-nucleus RNA-sequencing (snRNA-seq) and bulk RNA-sequencing data from mice with genetic knockdown of TrkB receptor expression in the LS. We examined gene networks affected by TrkB knockdown and their expression across newly identified LS cell types. These identified genes are central to neuroinflammatory and synaptic plasticity gene networks. We establish a BLA-LS circuit critical for social novelty recognition and provide evidence for BDNF-TrkB signaling as the molecular mechanism enabling LS-dependent social recognition