22689 research outputs found
Sort by
GENOME-WIDE ANALYSES OF CELL-FREE DNA FOR THERAPEUTIC MONITORING OF PATIENTS WITH PANCREATIC CANCER
No full textDetermining response to therapy for patients with pancreatic cancer can be challenging using imaging alone, and there is an unmet need for noninvasive assessment of tumor burden. We evaluated two liquid biopsy methods for assessing response to therapy using circulating cell-free DNA (cfDNA) in patients with metastatic pancreatic cancer treated with immune checkpoint inhibition and radiation as part of the CheckPAC trial (NCT02866383). Samples were evaluated after initiation of therapy and mutant allele fractions in cfDNA were determined using tumor-informed plasma whole-genome sequencing (WGMAF) and a mutation- and tumor-independent approach (ARTEMIS-DELFI), combining genome-wide cfDNA fragmentation profiles and repeat landscapes. Of those assessed with WGMAF, molecular responders (n=10) had a median progression-free survival (PFS) of 157 days compared to 51 days for molecular non-responders (n=10), (HR=0.23, 95% CI=0.07–0.69, p=0.0053), and a median overall survival (OS) of 319 days compared to 126 days for molecular non-responders (HR=0.29, 95% CI=0.11–0.79, p=0.011). For the ARTEMIS-DELFI approach, patients with low scores after therapy initiation (n=16) had a longer median PFS and longer median OS than patients with high scores (n=17), (PFS: 153 days versus 50 days), (HR=0.26, 95% CI=0.11-0.64, p=0.0013), (OS: 375 days versus 121 days), (HR=0.12, 95% CI=0.045-0.30, p<0.0001). By contrast, imaging at the same evaluation time point as the molecular analyses did not stratify survival (p=0.10). We validated our methodologies in a separate cohort of patients with pancreatic cancer treated with first line chemotherapy, with and without IL-6 inhibition, as part of the PACTO trial (NCT02767557). These analyses suggest that non-invasive mutation-based and fragmentation cfDNA approaches can identify individuals with pancreatic cancer who respond to therapy. Incorporation of these molecular methods to evaluate tumor burden may provide information for patient-physician decision making to improve patient care
Probing the Electronic Dynamics of Complex Materials Using Terahertz Light
Get PDFLandau theory is an extremely successful tool for describing phase transitions. Correlated electron systems are no exception, with Landau theory providing insight in a variety of kinds of ordered systems such as unconventional superconductivity and quantum magnetism. However, the description they provide is local in character. There have been recent developments of correlated electron systems, such as topological band theory, are characterized by global parameters like the Chern number, rather than by local order parameters. The most basic example of a system with a topological band structure is a Chern insulator, but these concepts also extend to more complex systems such as 3D Dirac and Weyl semimetals. In this thesis, three such materials are studied: the Weyl semimetal candidate CoTiGe, the Dirac semimetal CdAs, and the correlated ferromagnetic metal CaRuO.
The Weyl semimetal candidate CoTiGe is studied using a combination of time-domain THz spectroscopy and polarimetry. THz polarimetry with an applied magnetic field in the Faraday geometry reveals a THz-range Faraday rotation. The magnetic field dependence of this Faraday rotation, combined with \textit{ab initio} calculations, suggests that this rotation is consistent with an anomalous Hall effect arising from intrinsic Berry curvature, providing direct evidence of the topological nature of these systems.
Additionally, this thesis examines the Dirac semimetal CdAs using linear and nonlinear THz spectroscopy to explore energy and momentum relaxation in a regime relevant for low-temperature physics. The measured temperature dependence of the momentum and energy relaxation rates is consistent with electron-phonon scattering as the dominant process. However, no crossover is observed at the Bloch-Gr\"uneisen temperature, contrary to conventional theory.
Finally, the picosecond electronic dynamics of the ferromagnetic thin film CaRuO are investigated using THz spectroscopy. DC resistivity measurements and optical THz-range conductivity are combined to uncover two seperate momentum relaxation channels, consistent with other measurements on perovskite ruthenates. I find that lower frequency relaxation channel shows a sharp dependence to the onset of magnetic order. In contrast, using non-linear THz-pump THz-probe experiments, I find that the electronic energy relaxation rate is insensitive to the onset of magnetic order
The Impact of STEMscopes Science on Fifth-Grade Students’ Science Achievement in Longhorn ISD
Get PDFIn this quasi-experimental study, we examined STEMscopes program impacts on science achievement in Longhorn ISD. The report specifically analyzed STEMscopes impacts on STAAR and MAP Growth scores.
• The present study used a school-level quasi-experimental design, where 8 STEMscopes schools were selected for inclusion and 8 comparison schools (using other curriculums) were chosen for inclusion prior to data collection, based on their similarity in prior achievement and student demographics. The study followed both groups during the 2023-24 school year.
• The present study was situated in Longhorn ISD, a large school district in Texas, and one of the largest school districts in the United States. The analytic sample consisted of 990 Grade 5 students, taught by 42 teachers across 16 elementary schools.
• Data sources included Texas STAAR and MAP Growth science scores. Curriculum usage variables included the number of scopes (curriculum units) used at each treatment school.
• Impact analyses using OLS regression showed statistically significant positive impacts of STEMscopes on STAAR science scores. Treatment students outscored comparison students by approximately 94 points.
• Subgroup analyses did not show that the impacts of STEMscopes varied by student subgroups.
• Analyses of STEMscopes usage showed that median unique scope usage was significantly positively associated with STAAR science achievement scores
HOW MICROAGGRESSIONS EXPERIENCED BY BLACK WOMEN AT PREDOMINANTLY WHITE INSTITUTIONS AFFECT THEIR CHOSEN MAJOR, CAREER ASPIRATIONS, AND ECONOMIC MOBILITY
Get PDFBlack people make up 12% of the U.S. population, but they make up far less than 12% of people with college majors that lead to high-paying careers. Moreover, the economic discrepancy gap grows when gender is also at play along with race. Microaggressions based on race and gender at Predominantly White Institutions are well documented. However, the longer-term effects of microaggressions on Black women, such as impacting their self-efficacy and economic mobility after graduation, are not well documented. This research was a qualitative study to examine the effects of both racial and sexual microaggressions perpetrated against Black women at Predominantly White Institutions that affect their self-efficacy, anxiety levels, major chosen, and career trajectory. The study examined Black women who attended Predominantly White Institutions over a 45 years from 1971 to 2016.
A Black Feminist Thought methodology was used to allow the participants to tell their stories regarding their experiences while attending Predominately White Institutions. Although legislative changes were occurring within the United States around notions of educational and racial opportunities from 1971 to 2016, this study found that the overt and subtle racism and sexism the participants experienced did not change nor decrease during that same period. Moreover, the effects and frequency of the racism and sexism experienced by the participants increased over a 45-year span and, therefore, negatively impacted the anxiety levels, self-efficacy, career trajectory, and overall happiness of the participants after graduation. Furthermore, these findings conveyed that the behavioral and institutional practices by Predominantly White Institutions supported and perpetrated racist and sexist practices against Black women. The participants demonstrated resilience to the social and institutional racism and sexism they experienced. However, that resilience came at an emotional and, at times, professional cost to the Black women
Topical and injectable nanomedicine strategies for ocular drug delivery
No full textThe eye possesses physiologic and anatomic barriers that complicate ocular drug delivery. Eye drops are widely employed in the clinic due to their minimally invasive and patient-controlled nature. However, the efficacy of topical therapeutics is limited by low cornea permeability, short residence time, and tear film dilution which minimize intraocular accumulation. Conventional eye drops are prescribed for diseases of the anterior segment, including infection, hypertensive glaucoma and dry eye. For diseases affecting the posterior segment like age-related macular degeneration and retinitis pigmentosa, intravitreal or subretinal injections are required to reach the retina and surrounding tissues. These routes of delivery are invasive and have the risk of vision-threatening complications like retinal detachment or endophthalmitis. Delivery technologies that combine the minimally invasive nature of eye drops with the efficacy of local injection can improve clinical outcomes for a range of ocular diseases. Herein, we outline the current landscape for retinal neurodegeneration treatment, as well as describe two novel delivery strategies that overcome ocular barriers to improve treatment of retinitis pigmentosa and ocular infection.
Conventional eye drop formulations are currently unable to accumulate in the retina in therapeutic concentrations. Through the development of a hypotonic gelling eye drop antioxidant formulation, we were able to prolong corneal residence time, promoting accumulation in the posterior segment; this translated to the protection of photoreceptor function and retina structure in animal models of retinitis pigmentosa with once daily application, outlining a new methodology for topical delivery to the retina. Understanding the limitations of burdensome eye drop regimens for the treatment and prevention of ocular infection, we also describe a sustained-release nanocrystalline antibiotic injectable that packages a week’s worth of drug in a single, minimally invasive subconjunctival injection. By reducing frequency of application, we reduce the risk of vision-threatening complications and generation of antibiotic resistance through sublethal dosing, combined with improved treatment of bacterial keratitis and prevention of endophthalmitis in animal models of ocular infection. Collectively, this thesis outlines the development of new sustained-release nanomedicines to improve ocular delivery
An Information-Theoretic Framework for Designing Interpretable Predictors
No full textAs Machine Learning (ML) algorithms achieve state-of-the-art performance across various disciplines, there is a growing concern about the black-box nature of their decision‐making. Why does a deep neural network classify a tumour detected in an MRI scan as benign or malignant? Delineating the reasoning behind such predictions is critical for risk-sensitive applications such as healthcare and finance. Most existing approaches attempt to generate explanations post‐hoc in terms of attribution maps, e.g., by providing a subset of input features that are deemed `most important’ for the network output. However, prior work has shown that post-hoc explanations are often not faithful to the model's true decision-making process.
In this thesis, we focus on developing ML models that not only make accurate predictions but also provide an interpretable explanation of their prediction in a task- and user-dependent manner, that is, they are explainable by design. To achieve this, we propose to make predictions by asking a sequence of relevant queries about the input.
The prediction is then based on the resulting query-answer sequence, which also serves as an explanation for the model's decision. Implementing this framework poses three challenges --- (1) defining the set of relevant queries for a given task, (2) determining how to answer these queries for any given input, and (3) selecting the most relevant sequence of queries for prediction for any given input.
To address the first two challenges, we leverage user-defined queries and train classifiers on annotated datasets to answer them. If annotated datasets are unavailable, we leverage large language and vision models to obtain queries and their corresponding answers. To address the third challenge, we formulate the problem of selecting relevant queries for prediction as an optimization problem over query selection strategies such that the number of queries needed to make accurate predictions is minimized on average. This optimization problem is generally intractable, so we employ a greedy heuristic called Information Pursuit (IP). IP selects queries in order of information gain which is notoriously hard to estimate in high dimensions. Consequently, we propose efficient algorithms for IP based on generative models, variational methods and sparse coding techniques and demonstrate the efficacy of this framework on various vision, NLP and medical tasks
MECHANISMS UNDERLYING THE REGIONAL PREDISPOSITION TO AORTIC ANEURSYM IN LOEYS-DIETZ SYNDROME
No full textThoracic aortic aneurysms (TAA) are localized dilations of the aorta that predispose patients to life-threatening tears of the vessel wall. Human genetic studies have uncovered pathogenic variants that cause hereditary forms of TAA, highlighting the importance of the extracellular matrix, mechanosensing pathways, and transforming growth factor-β (TGF-β) signaling in aortic homeostasis. Loeys-Dietz syndrome (LDS) is a connective tissue disorder caused by mutations that impair TGF-β signaling, through partial inactivation of ligands, receptors, and signaling mediators.
Although LDS-causing mutations predispose individuals to aggressive aneurysms throughout the arterial tree, specific arterial regions, such as the aortic root, are especially vulnerable. Sites of severe aortic dilation are characterized by paradoxical secondary upregulation of TGF-β signaling. The work described in this dissertation investigates the dynamic responses to impaired TGF-β signaling, how it is modified by regional factors differentially expressed in specific aortic segments, and how these regional differences may explain the localized vulnerability to LDS-driven aortic pathology.
Chapter 1 provides an overview of the complex pathogenesis, genetic basis, and regional vulnerability of TAA. Chapter 2 delineates the primary and secondary transcriptional response of aortic smooth muscle cells to time-controlled, postnatal TGF-β inhibition. This analysis shows that while TGF-β inhibition causes broad downregulation of transcripts coding for extracellular matrix and focal adhesion components in smooth muscle cells, differential secondary responses, including more pronounced upregulation of these transcripts and TGF-β ligands, can be detected in specific subpopulations located in the aortic root. Chapter 3 examines the contributions of angiotensin II signaling to aortic dilation in an LDS murine model, revealing the differential effects of regional and systemic inactivation. Chapter 4 investigates the heterogeneity of smooth muscle cells in human and mouse aortas, identifying a subset of Gata4-expressing smooth muscle cells in the aortic root. Postnatal Gata4 deletion reduced aortic root dilation in LDS mice, indicating that Gata4 sensitizes the aortic root to impaired TGF-β signaling. Chapter 5 provides commentary on limitations and future directions. This dissertation highlights the importance of elucidating factors influencing regional aneurysm risk, providing insights into adaptive and maladaptive pathways for therapeutic interventions targeting specific arterial segments
BIFUNCTIONAL INHIBITORS OF SAMHD1 dNTPASE ACTIVITY AND NUCLEIC ACID BINDING OBTAINED THROUGH GUANINE TETHERING
No full textSterile Alpha Motif and Histidine-Aspartate domain-containing protein 1 (SAMHD1) is a multifunctional enzyme that regulates dNTP pools through its dNTPase activity and also facilitates DNA repair and ssRNA homeostasis through a distinct single-stranded nucleic acid (ssNA) binding activity. Both the dNTPase and ssNA binding activities require binding of guanine nucleotides to a guanine-specific A1 allosteric site on each enzyme monomer. GTP binding to the A1 site induces tetramerization of SAMHD1, which is required for the hydrolysis of all canonical dNTPs into their constituent nucleoside and tripolyphosphate group. Similarly, guanine nucleotides in ssNA bind to the A1 sites and induce a distinct tetramer form that is not competent for dNTP hydrolysis. The promiscuous dNTPase activity of SAMHD1 is important for resistance to many nucleoside drugs used to treat cancers. In addition, SAMHD1 ssNA binding at stalled replication forks downregulates an interferon response arising from release of fork associated ssDNA. Thus, inhibition of both activities would be expected to enhance the efficacy of nucleoside chemotherapeutic drugs. The aim of this dissertation is to develop novel guanine-linked small-molecule inhibitors that target the essential A1 site of SAMHD1. In one approach, a modified form of dGMP was coupled to a 376-member activated carboxylic acid library and screened for inhibition of the dNTPase and ssNA binding activities. A lead compound with low micromolar binding affinity and in vitro activity against the dNTPase and ssNA binding was identified and characterized. However, this primary hit had no efficacy in a cellular nucleoside drug potentiation assay because of the unfavorable charge characteristics of its monophosphate group. To overcome these in vitro limitations, a new A1 site targeting approach was taken using the guanine antiviral acyclovir (ACV). ACV was converted to its primary amine, which allowed facile coupling to activated carboxylic acid and sulfonyl chloride libraries to yield diverse coupled compounds with either uncharged amide or sulfonamide linkages. Several ACV-coupled compounds were identified with similar in vitro potencies as the first hit, but with more favorable pharmacologic properties. One of these compounds was tested in the cellular nucleoside drug potentiation assay, where it modestly enhanced the potency of the cytosine arabinoside chemotherapeutic. Crystal structures combined with computer modeling showed how the ACV-linked compound bound to the A1 site and how the approach can be further exploited to generate compounds with greater efficacy
Organizing millions of documents using LLMs: The case of the UCSF-JHU Opioid Industry Documents Archive
Get PDFThis paper was accepted for presentation at the LLMs for Library Info Organization (LLM4Cat) workshop held in conjunction with iConference 2025.Introduction. The UCSF-JHU Opioid Industry Documents Archive (OIDA) is a growing collection of millions of documents. The metadata necessary to differentiate these documents, as supplied, is often lacking. Both the scale and the rapid growth of the corpus preclude correction by human metadata experts. Methods. We have applied open-source AI models and multi-modal LLMs for classifying documents and images, PDF image extraction, and image captioning. Recent expansions feature LLM-powered generation and correction of descriptive metadata - including document titles, authors, and dates - along with named entity extraction to increase access through metadata and knowledge graphs. All techniques involve expert, human-in-the-loop, review. Preliminary results. Document classification has achieved 80% accuracy in testing. DocLayout-YOLO was selected for superior accuracy and scalability of document layout analysis. On image classification tasks, our zero-shot approach achieved 58% expert agreement. Rather than pursuing fine-tuning or complex prompting strategies, we have opted for expert supervision of AI-generated metadata to ensure quality. Conclusion. This work demonstrates the strengths of combining LLMs with expert review to address the challenges of metadata creation at scale. There is a significant opportunity to contribute to specialized benchmark datasets that evaluate model performance on fundamental cataloging and metadata tasks.The work of the OIDA team has been funded in part through settlements of public-interest lawsuits by U.S. states
Learning Safe Regions in High-dimensional Dynamical Systems via Recurrent Sets
No full textSafety certification in dynamical systems has relied heavily on the identification of an invariant set that strictly requires trajectories to always lie within it. This foundational approach addresses two key safety objectives: stability and avoidance. To this end, techniques such as the Lyapunov method and the barrier function method that are capable of characterizing an invariant set have been pivotal. However, the invariance requirement poses strict constraints on the learning outcome. Therefore, as the system dimension increases, directly characterizing invariant sets or identifying functions that define invariant sets typically demands considerable domain-specific knowledge or extensive computational resources.
This thesis seeks to develop new data-driven methodologies that facilitate the verification of stability and avoidance in dynamical systems, without relying on the identification of invariant sets. A key innovation of this thesis is the application of the concept of recurrence to relax the stringent constraints imposed by invariance. Specifically, a set is recurrent if trajectories originating from it return to it infinitely often. By leveraging recurrence, safety can be characterized with enhanced efficiency and accuracy.
This thesis theoretically establishes necessary and sufficient conditions for using recurrence to characterize safety, offering a deeper understanding of how recurrence can serve as a reliable proxy for invariance. Practically, it introduces practical, data-driven algorithms that utilize only a finite number of finite-length sampled trajectories to determine safe regions within a dynamical system. Optimized for computational efficiency, these algorithms can be implemented on parallel processing units, making them highly applicable in real-world scenarios where rapid and reliable safety verification is crucial