Washington University Medical Center
Washington University St. Louis: Open ScholarshipNot a member yet
26344 research outputs found
Sort by
Explainable Decision-Making: From Formal Logic to AI Systems with Explainable Behavior
Get PDFThis thesis makes the claim that logic-based frameworks can serve as an explainability layer atop AI systems, capable of generating rigorous and flexible explanations for human users across diverse problem domains. We support this claim through a progression of novel theoretical frameworks and practical implementations, starting with a general logic-based framework for generating explanations from the knowledge bases of an AI system and a human user and showing how it can be used on a diverse set of problem domains. We then systematically extend this framework with capabilities crucial for real-world applications: probabilistic reasoning for handling uncertainty, personalization through vocabulary-based abstraction, and dynamic interaction through argumentative dialogues. Building on these foundations, we address additional challenges by developing privacy-aware explanations for multi-agent systems and exploring explanation-guided approaches to belief revision that better align with human cognitive processes. To make our methods more accessible, we demonstrate how it can be effectively combined with large language models to generate natural language explanations while maintaining formal guarantees. Our theoretical contributions are complemented by efficient computational methods that make these frameworks more practical, as demonstrated through extensive evaluations across diverse problem domains. Recognizing that the ultimate test of explanatory frameworks lies in their effectiveness with real human users, we validate our approaches through several human-subject studies that show high comprehension of the explanations as well as high overall satisfaction with the explanation process, thus providing some evidence for the effectiveness of our approaches in enhancing human-AI interaction. By showing how logic can serve as a robust explainability layer that bridges the decision-making processes of AI systems and human understanding, this work aims to contribute to the development of AI systems that are not only powerful but also understandable, trustworthy, and above all, human-aware
Advances in Simulating Global Fine-Scale Ambient Air Pollutants and Source Contributions
No full textFine-resolution chemical transport models are essential to resolve fine-scale air pollution patterns due to localized emissions, nonlinear chemical feedbacks, and complex meteorology. However, fine-resolution global simulations of air quality remain rare, especially of the Global South. Recent developments to GEOS-Chem model in its high performance configuration (GCHP) enable routinely conducting global air quality simulations at spatial resolution ~60 times finer than the coarse global models traditionally available to the research community (~25 × 25 km2 vs. ~200 × 200 km2). This dissertation is centered on utilizing and improving fine-resolution simulations to produce more accurate spatial estimates of air pollutants combined with satellite-based observations and ground-based measurements. The first section advances fine-scale estimates of population exposure and sectoral contributions from a global fine-resolution simulation with a focus on the Global South. We analyze the discrepancies of population exposure to fine particulate matter (PM2.5) and nitrogen dioxide (NO2) globally and in populous cities between fine and coarse GCHP simulations. We also examine the resolution dependence of sectoral contributions from the differences of fractional contributions of emission sectors across model resolutions with sector sensitivity tests using GCHP. The second section is aimed at investigating the model resolution effects on geophysical satellite-derived PM2.5, inferred from satellite retrieved aerosol optical depth (AOD) and simulated surface PM2.5 to AOD relationship. We compare satellite-derived PM2.5 concentrations across model spatial resolutions using GCHP and examine the overall resolution sensitivities contributed by different PM2.5 components. We further investigate the resolution effects on the simulated aerosol vertical profile, which shows vertical contrast of near-surface emissions and pollutants transported aloft, especially over isolated sources. Mineral dust exerts strong impacts on air quality as the most abundant aerosol by mass, on ecosystem health through nutrient transport and deposition, and on climate change by affecting the radiative budget globally. The third section is targeted at improving fine mineral dust representation in GEOS-Chem from the surface to the column against satellite-based and ground-based observations, leveraging recent mechanistic understanding of dust source and removal. Specifically, we implement a new dust emission scheme, revisit the size distribution of emitted dust, explicitly track dust with diameter less than 2 μm, and update the parametrization for below-cloud scavenging. In summary, these investigations highlight the capability of a global fine-resolution simulation by the GEOS-Chem model in better resolving the spatial heterogeneity of air pollution due to distinct sources, chemical nonlinearities, and complex meteorology, with implications for location-specific emission mitigation strategies. Model developments to fine mineral dust indicate the importance of consistent representation of size in models versus measurements, the spatial distribution of dust emissions, the size distribution of emitted dust, and the explicit tracking of fine bins for more accurate simulation of fine dust abundance from the surface to the column
Interpreting spatial and temporal variations in global air quality using a chemical transport modeling framework
Get PDFAir quality is a major health concern. Exposure to fine particulate matter (PM2.5) and nitrogen oxides (NOx = NO + NO2) is a leading mortality risk factor across the world. Numerous cohort studies conducted over the past two decades have identified strong associations between ambient air pollution and human mortality, highlighting the adverse health effects of PM2.5 and NO2 exposure. My thesis includes three studies that contribute to a better understanding of air quality. In many regions, including South Asia, elevated emissions from various sectors and sparse ground monitoring are significant challenges. The concurrent development and use of the high-performance configuration of GEOS-Chem (GCHP) provide unprecedented opportunities to assess the primary sources of PM2.5 as a function of emissions, meteorology, chemistry, and deposition. The application of satellite observations with high-resolution source-based information can enhance scientific understanding of air pollution patterns and their impact on human health. Our findings indicate that residential combustion (28%) and biofuels (31%) are major sources of PM2.5 linked to mortality in South Asia. NOx affects air quality and human health directly by contributing to premature mortality and asthma for children and adults and indirectly by acting as precursors for tropospheric ozone (O3) formation and nitrate aerosols. Pandora sun photometers provide hourly ground-based NO2 observations, which can be used to validate satellite observations. Pandora’s observed columns are affected by vertical variations in temperature and local solar time. We used the GCHP model to correct Pandora observations, improving accuracy by 8% through adjustments for temperature and local solar time variations. Furthermore, we used fine-scale GCHP simulations for better estimation of NO2 columns for an accurate representation of NO2 column-to-surface relationships. NOx emissions from bottom-up inventories are limited by uncertainty and latency. Satellite instruments offer NO2 column measurements that can be used to infer surface NOx emissions through inverse modeling. We simulated synthetic NO2 column densities as observed by the Tropospheric Ozone Monitoring Instrument (TROPOMI) over eastern North America to test the ability of the iterative finite difference mass balance (IFDMB) method to recover NOx emissions. Additionally, we applied a resolution-optimized mass balance approach (ROMBA) to retrieve surface NOx emissions from TROPOMI NO2 columns for June–August 2019. Our tests include the use of multiple grid resolutions of 200 km, 100 km, and 55 km, revealed that simulations with 100 km resolution most accurately recovered true emissions. Our global top-down estimates for annual land surface NOx emissions (42.1 Tg NO2) closely match the CEDS a priori estimate (38.4 Tg), with the highest agreement over the eastern United States (10–15%). We observed significant regional variations, with top-down NOx emissions 30–60% higher in regions such as northern South America, Africa, southern Europe, western Canada, the United States, and the Middle East. In contrast, we found top-down NOx emissions 50–100% lower in southern South America, Africa, India, and parts of Europe and China. Overall, my doctoral dissertation makes three key contributions to the field of atmospheric sciences: 1) identifying residential combustion and biofuel emissions as major drivers of ambient and household air pollution and mortality in South Asia; 2) correcting NO2 observation systems and interpreting hourly variations in NO2 concentrations for more accurate satellite-based analysis; and 3) enhancing satellite-based emissions estimation using inverse modeling framework
Parental Guidance Suggested: Genetic, Epigenetic, and Environmental Factors Shape Allele-Specific Gene Regulation in Metabolic Traits
No full textDeciphering the “genotype to phenotype” map is a crucial goal of biology. We will not improve our ability to predict phenotypes from DNA sequences until we better incorporate epigenetic and environmental variation into our models of human health and disease. Genes are typically presumed to express both parental alleles equally, but genetic and epigenetic factors can cause one allele to function differently than the other (e.g. expressed or silenced more). These allele-specific imbalances can alter the final composition of a gene’s transcriptional product and ultimately shape phenotypes. Allele-specific effects have widespread impacts on complex traits, but how environmental signals contribute to this phenomenon remains unclear. This thesis unravels the causes and consequences of allele-specific gene regulation on metabolic phenotypes related to obesity and diabetes. First, I explored how DNA sequence, parent-of-origin, tissue, sex, and dietary nutrition simultaneously govern allele-specific gene expression (ASE) biases. I used a simple, yet powerful, F1 reciprocal cross of the LG/J and SM/J mouse strains to distinguish between two ASE classes: parent-of-origin dependent (unequal expression based on parental origin) and sequence dependent (unequal expression based on haplotype sequence). I constructed a genome-wide map of ASE patterns in thousands of genes across three metabolic tissues and nine environmental contexts. The magnitudes and directions of these expression biases are highly sensitive to tissue type and environmental factors. Second, I probed how those same factors influence both classes of allele-specific DNA methylation (ASM) biases. Parent-of-origin ASM sites are enriched in the promoters of genes with parental expression biases, while sequence ASM sites are enriched in intergenic regions with unclear relevance to gene expression. Third, I integrated my allele-specific findings with quantitative trait loci (QTL) data from published intercrosses of these same strains. Tissue-specific ASE genes are enriched in QTLs for metabolic and musculoskeletal traits yet comprise a small number of all genes within those QTL. I demonstrate how this orthogonal approach can pinpoint actionable candidates for functional validation. Collectively, my thesis work provides novel insights into how genetic, epigenetic, and environmental variation modulate allele-specific gene regulation in metabolic phenotypes. Finally, I outlined my work with the Young Scientist Program, a K-12 STEM outreach program. I described the curriculum for a science communication course that introduces crucial science literacy skills to high school students. After completing the course, students reported significant improvements in their self-confidence to read, interpret, and communicate scientific data. This chapter also renders visible my contributions to improving how we train and support the next generation of diverse scientists
Balancing Defense and Tolerance: Complement-Mediated Pathogen Protection and Commensal Microbial Antigen-Specific Regulatory T Cell Responses in the Gut
No full textThe intestinal immune system plays a critical role in maintaining a delicate balance between immune tolerance to dietary and commensal antigens and effective defense against pathogens. This dissertation presents two independent but conceptually linked studies that investigate distinct immunological mechanisms contributing to intestinal homeostasis. The first study examines the role of the complement system, specifically complement component C3, in mucosal defense against Vibrio cholerae infection. To more accurately investigate in vivo host responses to cholera toxin, we developed the first adult (juvenile) mouse model of cholera toxin–mediated pathogenesis during V. cholerae infection. Using this model, we found that cholera toxin induces robust C3 expression in the gut, originating from multiple cell types, including epithelial cells, stromal cells, and immune cells. We further demonstrate that epithelial-derived complement activation plays a pivotal role in preserving intestinal barrier integrity during infection, without significantly affecting bacterial burden. The second study investigates how the intestinal regulatory T cell (Treg) repertoire is shaped by exposure to self, dietary, and microbial antigens. Using gnotobiotic and germ-free TCRβ-transgenic mice, combined with high-throughput TCRα sequencing, we classified Treg TCRs based on antigen origin and analyzed how colonization with defined bacterial consortia, including human-derived Clostridiales isolates and mouse Helicobacter species, differentially influences the colonic Treg landscape. In vitro validation confirmed that repertoire-based antigen predictions accurately reflect TCR specificity, revealing that distinct microbial species can imprint unique antigenic signatures on the intestinal Treg compartment. Together, these studies highlight complementary innate and adaptive mechanisms by which the immune system preserves gut integrity and immune tolerance. By dissecting the contributions of complement and regulatory T cells to intestinal immunity, this work advances our understanding of the dynamic interplay between the host and its microbial environment, with implications for infection, autoimmunity, and microbiota-targeted therapies
An Intervention to Empower People to Confront Sexism
No full textEven though sexist comments and behavior occur frequently in everyday life, people do not often confront the individuals who are responsible. However, interpersonal bias confrontations can be powerful tools for reducing expressions of bias and norm setting. The current studies tested an intervention designed to increase confrontations by harnessing theory and research on what holds people back from confronting. We designed a half-hour intervention consisting of engaging and evidence-based videos on confronting bias and self-reflection exercises to increase personal relevance. The module aimed to boost confronting by intervening at multiple points of the decision-making process by increasing recognition of sexism, perceived responsibility for confronting, and evaluations that benefits are worth potential costs. Study 1 randomly assigned participants from the general U.S. population (N = 376) on Prolific to an intervention or control condition. Relative to control, the intervention increased knowledge about gender bias, self-efficacy to identify bias and confront, and perceived benefits of confronting. When evaluating hypothetical scenarios, the intervention group was more likely to overcome barriers and report intentions to confront. As Study 1 showed promising results for intentions to confront, we sought to replicate and extend the work by using a behavioral outcome in Study 2 (N = 450). Study 2 used a similar experimental design, but participants were additionally asked to watch a video containing gender bias and leave a comment. To assess change in behavior, we examined whether participants called out gender bias in the comment they left on the video. Study 2 results replicated Study 1 on several key outcomes, including perceived benefits of confronting and identifying gender bias. Other outcomes had significant intervention effects among undergraduate men but not undergraduate women. The intervention significantly increased men’s rate of calling out gender bias in the behavioral outcome, however, there were no differences for women. The intervention, while less effective for undergraduate women, consistently led to higher identification of gender bias and increased capacity to confront sexism, which could substantially decrease norms about expressing sexism in everyday life. By making barriers seem more manageable and providing actionable pathways toward confronting, our intervention empowered individuals to confront
Simplicity from complexity: Sensing high-dimensional fluctuations via regulatory network, optimality in allocating cellular resources, and recovering functional groups via data-driven methods
Get PDFBiological systems must navigate dynamic and complex environments while operating under constraints such as limited molecular resources, biochemical noise, and structural trade-offs. This dissertation investigates three distinct cases in which biological systems allocate resources and process information efficiently, revealing emergent simplicity from underlying complexity. The first study develops a generalized end-product inhibition model with cross-talk and an excess of regulators, providing a proof-of-principle that simple biochemical circuits could, in principle, dynamically adjust their responsiveness to high-dimensional environmental variation. This result highlights how such circuits may allow adaptive filtering of both dominant and subdominant fluctuation modes. The second study, conducted in collaboration with Professor Shankar Mukherji’s group, explores cellular organelle biogenesis as a resource allocation problem. By integrating mathematical modeling and experimental imaging, this work uncovers critical scaling relationships between organelle number and size. The results suggest that cellular systems optimize organelle biogenesis under limited resource pools, leading to distinct allocation strategies for de novo synthesis and fission-derived organelles. The third study addresses functional organization in microbial ecosystems, where community-level metabolic function emerges from complex species interactions. This work compares two regression-based approaches—Ensemble Quotient Optimization (EQO) and LASSO—to infer functional groups from microbial abundance data. By evaluating these methods under increasingly realistic conditions, the analysis highlights the trade-offs between statistical regularization and prior assumptions in functional group recovery. Together, these three cases provide complementary perspectives on how biological systems transform complex processes into simpler, computationally efficient frameworks. By integrating theoretical models, statistical inference, and experimental data, this thesis contributes to our understanding of biological information processing across molecular, cellular, and ecological scales
Why Automatic Enrollment Is Essential for the Success of Trump Accounts: Lessons from SEED OK
Get PDFIn this policy brief, prominent Child Development Account (CDA) experts and researchers present the case that automatic enrollment is essential in a nationwide policy to provide CDAs and build assets for all children in the United States. The Center for Social Development (CSD) has designed and rigorously tested this policy concept, and the authors highlight key evidence on ensuring full participation, efficiency, and long-term success of a federal policy
How do self-employment rates change across the lifespan?
Get PDFPeople pursue self-employment for a variety of reasons, including autonomy, the ability to pursue their own goals, and income. Yet for many older people who are not yet eligible for Social Security retirement benefits, which can begin at age 62, or Medicare, the U.S. health insurance program that generally begins at age 65, self-employment is also associated with less health insurance and retirement savings program coverage. 1,2 3,4
This research brief asks, What is the proportion of Americans who are self-employed, and how do these rates change throughout the life course
An Enzyme-Coupled Isotope Dilution Mass Spectrometry Assay for Non-adjacent DNA Photoproducts as Intrinsic Probes for G-quadruplexes in vitro
Get PDFG-quadruplexes are noncanonical secondary structures that form in guanine-rich nucleic acid sequences and are thought to be involved in the control of gene expression.1 Unfortunately, it has been difficult to prove that these structures exist in vivo because of their dynamic nature. Rather than try to develop G-quadruplex specific binders as others have done,2 our idea is to photochemically trap certain types of G-quadruplexes by irreversible photoproduct formation3–5 and then detect the stable G-quadruplex-specific photoproducts by Isotopic Dilution Mass Spectrometry (IDMS).6 This IDMS method could then be used to unambiguously confirm the presence of specific types of G-quadruplexes in vivo and help justify more detailed studies of their role in the regulation of DNA transcription and gene regulation.7,8 Our approach is based on the observation that UV-irradiation produces DNA photoproducts that form with stereochemistry and regiochemistry that depend on the structure of the DNA and the relative positioning and orientation of the precursor bases.9 While the primary photoproduct formed in duplex DNA is an adjacent cis,syn cyclobutane pyrimidine dimer (CPD) with a head-to-head orientation,10 non-adjacent anti CPDs will form in basket and chair-type G-quadruplexes with a head-to-tail orientation.5 These adjacent and non-adjacent CPDs can be detected by enzymatic degradation with nuclease P1 and snake venom phosphodiesterase, which produces a dinucleotide CPD with an intradimer phosphodiester linkage from adjacent CPDs, and dinucleotide CPDs lacking a phosphodiester from non-adjacent CPDs.9 These enzymatically degraded DNA photoproducts can be readily distinguished by HPLC and mass spectrometry, making them suitable as intrinsic photoprobes of B and non-B DNA conformations. Herein, we report methods for preparing isotopically labeled cyclobutane thymidine dimers and show how these standards can be used to identify and quantify non-adjacent photoproducts produced in non-B DNA structures such as G-quadruplexes by IDMS. We therefore expect that this method should find application to the study of the photochemistry and photobiology of non-B DNA structures in vivo