22689 research outputs found
Sort by
Topics in Classical and Quantum Optimization: Complexity and Algorithms
Get PDFThis thesis presents advances revolving around optimization, complexity, and applications of quantum computing.
In particular, Part 1 studies information complexity, focusing on convex mixed-integer optimization through the contents of (Basu, Jiang, Kerger, Molinaro 2023), primarily.
With new results on lower bounds, the goal thereof is to better understand how difficult these problems are to solve in different settings.
We refine the classical notion of first-order information complexity to allow for questions such as ``how many bit are needed to solve a given optimization problem?", provide results for a few settings of interest, and prove a powerful theorem to transfer lower bounds on information complexity across settings.
Notably, lower-bounds on information complexity also imply lower-bounds on algorithmic complexity.
With such lower-bounds in place for classical computational systems, with the recent emergence of quantum computing technologies, it is only natural to wonder whether existing lower-bounds on optimization problems for algorithmic complexity for classical machines can be ``beaten" by using quantum computers. We conclude the chapter with some discussion comparing the oracles we consider for the information complexity lower bounds with quantum versions of those oracles and some related directions for future work.
Part 2 then focuses on advancements in quantum computing for optimization problems. Chapter 4 presents a method for formulating the image denoising problem as a quadratic unconstrained binary problem through using Boltzmann Machines and quantum annealing, as in (Kerger & Miyazaki 2023). This is applied to real (downscaled) data using a state-of-the-art D-Wave quantum annealer. Some statistical guarantees for the method are proven, a robustness modification is proposed that performs well in practice, and promising numerical results are provided. Chapter 5 introduces quantum distributed computing models along with algorithms that are asymptotically faster than any of their known counterparts in the classical model. Though these do not yet "beat" any known lower-bounds for the classical model, these results are the first step towards such an achievement. We also spend considerable time on computing the exact complexity of the algorithms discussed, revealing that the algorithms discussed, along with existing related algorithms from the literature, are extremely impractical for realistic problem sizes. Extensions of the work to the survivable network design problem are discussed afterwards. We conclude the thesis with a general discussion on optimization, information and algorithmic complexity, and different computing technologies in relation to the content presented
Multiplexed high-resolution melt for broad-range infectious pathogens identification
No full textToward combating infectious diseases caused by pathogenic bacteria, there is a pressing demand for diagnostic tools that can quickly and accurately identify the causative even within complex and polymicrobial samples. Digital PCR integrated with high-resolution melt analysis (dPCR-HRM) emerges as a promising method for extensive and rapid bacterial identification. However, gaps in our comprehension of the design principles of dPCR-HRM, along with the need to expand the spectrum of identifiable species — including cross-kingdom detection — as well as enhance identification precision, are challenges that must be addressed to refine detection capabilities and improve outcomes in clinical diagnostics.
In this thesis, we tackle existing diagnostic challenges by systematically studying dPCR-HRM platforms. Our systematic exploration begins with the introduction of a novel design development workflow for dPCR-HRM, which utilizes computational in silico HRM analysis to pre-screen PCR primers. This approach aims to select primers that yield digital melt curves with desirable traits: high interspecies variability and low intraspecies variability, thereby enhancing the accuracy of bacterial identification. Through this new development workflow, we also report a new digital PCR-HRM assay with improved bacteria identification accuracy (Chapter 2). To further improve accuracy and address device-related temperature variances, we've implemented an oligonucleotide-based temperature calibration method that significantly refines bacterial detection accuracy (Chapter 3). Progressing further, we report a novel strategy, the duplexed dPCR-HRM for cross-kingdom detection. This innovative assay employs two sets of universal primers for the simultaneous detection of both bacterial and fungal species, effectively overcoming the limitations inherent in traditional diagnostic methods when dealing with co-infection samples (Chapter 4). Moreover, we advance our exploration by introducing a novel multiplexed dPCR-HRM assay for broad-spectrum detection. This strategy utilizes multiple sets of universal primers targeting various regions of the bacterial 16S rRNA gene, which diversifies the digital melt profiles of bacterial species. This diversification significantly enriches the assay's capability for broad-spectrum pathogen identification. Enhancing this multiplexed approach, we incorporate a Convolutional Neural Network (CNN) for deep learning analysis, ensuring precise identification of pathogens (Chapter 5). Collectively, these advancements mark a significant contribution to diagnostic methodologies, providing a comprehensive, efficient, and reliable framework for the prompt management of infectious diseases
Synthesis and Characterization of Two-Dimensional Layered Materials: A Study of Quantum Materials
Get PDFTwo-dimensional(2D) layered materials are a class of compounds which contain empty space between layers of atoms and are typically held together by van der Waals(vdW) forces. These materials can be exfoliated by both simple mechanical techniques such as the Scotch tape method made famous during the discovery of graphene, the first widely-known 2D layered material, and by more elegant liquid-based or chemical methods. Ever since the discovery of graphene in 2004, the field of 2D layered materials has grown rapidly, as many of these materials exhibit desirable quantum properties such as superconductivity and have advanced our understanding of these phenomena, to say nothing of the useful tunability of the crystal structure via exfoliation, intercalation, and other means and therefore the properties of the material.
In this dissertation, we take a close look at the synthesis and properties characterization of three different layered materials: the new misfit layered compound (EuS)1.13(NbSe2)2, the layered oxychloride TaOCl2, and the layered nitride halide family ZrNX (X = Cl, Br, I), all falling under the umbrella class of 2D layered materials, but all behaving differently and belonging to different subfamilies of materials. (EuS)1.13(NbSe2)2 possesses a complex, incommensurate structure composed of two distinct sublattices, low thermal conductivity and an antiferromagnetic phase transition at TN = 4.7 K. TaOCl2 has unusual Ta-Ta dimerization in the crystal structure that generates filled flat bands in the electronic band structure leading to significant electrically insulating behavior. ZrNX materials are typically insulating, but undergo a metal-insulator transition once intercalated with alkali metal ions and further transitions to a superconducting state with relatively high critical temperatures, starting from 13 K for the first member of the family to be discovered, Li-intercalated β-ZrNCl. Through this in-depth analysis, the vast potential for highly varied and useful properties that still exists within this intensely-studied field is emphasized. We also examine the historical background with an eye towards how these projects are expanding beyond the foundation laid by the older experiments
Vacancy clustering and void formation in energetically processed magnesium alloys: A transition interface sampling study
Get PDFSolid-state precipitation and growth of new phases within solid metals are crucial for the design and manufacturing of alloys. Precipitation, especially at the nanoscale, is one of the most effective methods for achieving high strength. In magnesium (Mg) alloys, it has been demonstrated that such microstructures can be achieved via mechanical deformation processing, although the precise physics of this process remain unclear. Notably, elevated concentrations of vacancies arise during such processing methods as well as during radiation, potentially playing a crucial role in the intermetallic nucleation process. In this study, transition interface sampling is used to statistically sample the pathways in phase space associated with void nucleation from vacancies, a likely precursor to nanoscale intermetallic precipitation. The effects of alloying are quantitatively assessed.
Simulations in Mg-Al alloys with high vacancy concentrations indicate that the presence of solute species reduces thermodynamic barriers for the clustering of vacancies and the formation of voids. The occurrence of local minima in the free energy along the reaction coordinate suggest that void formation becomes a multi-step process in the presence of solute. According to this scenario, vacancies aggregate with solute before coalescing into stable voids with well-defined internal surfaces. The emergence of vacancy-solute clusters as intermediate states suggests that classical nucleation theory may be inadequate to describe void formation in alloys with high vacancy concentrations. Furthermore, analysis of simulation data provides evidence that voids can influence the composition in their vicinity by attracting solute. This is hypothesized to promote the nucleation of precipitates
The unexpected localization of PAD4 in monocytes: implications for rheumatoid arthritis pathogenesis
Get PDFAlthough anti-citrullinated protein autoantibodies (ACPAs) are a hallmark serological feature of rheumatoid arthritis (RA), the mechanisms and cellular sources behind the generation of the RA citrullinome remain incompletely defined. Peptidylarginine deiminase IV (PAD4), one of the key enzymatic drivers of citrullination in the RA joint, is expressed by granulocytes and monocytes; however, the subcellular localization and contribution of monocyte derived PAD4 to the generation of citrullinated autoantigens remain underexplored. In this study, we demonstrate that PAD4 displays a widespread cellular distribution in monocytes, including expression on the cell surface. Surface PAD4 was enzymatically active and capable of citrullinating extracellular fibrinogen and endogenous surface proteins in a calcium dose–dependent manner, and fibrinogen citrullinated by monocyte-surface PAD4 could be recognized by ACPAs. Several novel PAD4 substrates were identified on the monocyte surface via mass spectrometry, with citrullination of the CD11b and CD18 components of the Mac-1 integrin complex being the most abundant. Citrullinated Mac-1 was found to be a target of ACPAs in 25% of RA patients, and Mac-1 ACPAs were significantly associated with HLA-DRB1 shared epitope alleles, higher C-reactive protein and IL-6 levels, and more erosive joint damage. Our findings implicate the monocyte cell surface as a previously undescribed and consequential site of extracellular and cell surface autoantigen generation in RA.
The mechanisms by which PAD4, which lacks conventional secretory signal sequences, traffics to extranuclear localizations are unknown. In this study, we also show that PAD4 was enriched in the organelle fraction of monocytes with evidence of citrullination of organelle proteins. We demonstrated that PAD4 can bind to several cytosolic, nuclear, and organelle proteins that may serve as binding partners for PAD4 to traffic intracellularly. Additionally, cell surface expression of PAD4 increased with monocyte differentiation into monocyte derived-dendritic cells and co-localized with several endocytic/autophagic and conventional secretory pathway markers, implicating the use of these pathways by PAD4 to traffic within the cell. Our results suggest that PAD4 is expressed in multiple subcellular localizations and may play previously unappreciated roles in physiologic and pathologic conditions
MY ROSS COUNTY ALMANAC
Get PDFThis collection of essays from the Master of Arts in Science Writing Program at Johns Hopkins University focuses on the author’s experiences adjusting to life on a farm in south-central Ohio, where she moved from Washington, DC, at the start of the COVID-19 pandemic. She writes of the animals, the geology, and history of Ross County, Ohio, and a few other places that she came to know over the course of the program
The fate and consequences of aneuploid cells in colon organoids
No full textAneuploidy is a large-scale genomic alteration characterized by an unbalance number of chromosomes that plays a context-dependent role in tumorigenesis. Activation of the tumor suppressor protein, p53, has been proposed to limit the proliferation of aneuploid cells in cultured cell lines. Here, we demonstrate that cells in mouse colon and human mammary organoids do not activate p53 or arrest following the induction of simple aneuploidy, which is characterized by gain or loss of a small number of chromosomes. In mouse colon organoids, loss of p53 results in chromosome mis-segregation and frequent aneuploidy suggesting p53, at least in part, maintains genome stability by promoting mitotic fidelity. Colorectal cancers, which are frequently aneuploid, arise from long-lived adult stem cells that normally self-renew and give rise to differentiated cells that make up the functional intestinal epithelium. We induced heterogeneous aneuploidy in patient-derived human colon organoids to investigate the fate of aneuploid intestinal stem cells and the consequences of aneuploidy on epithelial function. We find that cells with complex aneuploid karyotypes activate p53 and arrest while cells with simple aneuploidy continue proliferating. p53 activation and arrest in complex aneuploid cells was not suppressed by inhibiting the DNA damage response. Simple aneuploid cells expressed stem cell marker genes and maintained the capacity to form organoids from single cells indicating aneuploidy does not lead to premature differentiation of intestinal stem cells. Instead, simple aneuploidy resulted in a delay of differentiation in the absence of stem cell niche factors. We propose that aneuploidy may contribute to early cancer evolution in the context of the colon by promoting growth outside of the defined intestinal stem cell niche. Finally, we show that complex aneuploidy disrupts the normal colon epithelial architecture and compromises the epithelial barrier highlighting the importance of continuing to investigate the impacts of aneuploidy on tissue form and function
INVESTIGATING THE DOWNSTREAM REACTIONS AND EFFECTS OF DNA METHYLATION AND ONE ELECTRON OXIDATION IN NUCLEOSOME CORE PARTICLES AND CELLS
Get PDFDNA damage caused by alkylation and oxidation is harmful to cells, potentially leading to cell death or cancer. Alkylating drugs and radiotherapy exploit this cytotoxicity to kill tumor cells. While the lesions caused by alkylation and one-electron oxidation were thoroughly investigated in DNA, the chromatin environment introduces additional complexity. Interactions between DNA and histone proteins may change the reaction kinetics of certain DNA lesions and give rise to secondary products, such as DNA-protein cross-links (DPCs).
Methylating agents predominantly produce N7-methyl-deoxyguanosine (N7-M-dG). The toxicity of N7-M-dG was attributed to its depurination product, abasic sites. However, in nucleosome core particles (NCPs) – the basic units of chromatin, depurination of N7-M-dG is slower compared to free DNA, with a half-life extending up to 80 days. We utilized an ion-sensitive fluorescent probe to demonstrate that interactions with histone tails suppress the depurination of N7-M-dG by increasing the local ionic strength in NCPs. The persistence of N7-M-dG leads to DPC formation between the histone tail lysines and N7-M-dG (DPCN7MdG). Mass spectrometry (MS) analyses revealed that N7-M-dG persists in vivo. The yield and absolute amount of DPCN7MdG in MM-treated cells were quantified. Cell viability drops significantly in the absence of SPRTN-mediated DPC repair, indicating the physiological importance of DPCN7MdG.
DPC formation also occurs following one-electron oxidation of DNA, a process that predominantly generates dG radical cations (dG•+/hole). The π-system of DNA enables hole transfer to a more distant site with a lower redox potential (e.g GGG sequence). Trapping of dG•+ by water ultimately results in dG lesions. In chromatin, the presence of histone tail lysines raises the possibility of cross-link formation between histones and dG•+ (DPCHT). We investigated this in NCPs by inducing hole transfer events at various locations and sequences. Combining computational modeling results, we determined that the yield of DPCs and the site of cross-linking on DNA are influenced by DNA-histone tail interactions. GGG sequences are capable of producing DPCHT with a yield comparable to that of water-trapping dG lesions. MS and histone mutation experiments revealed that lysines and the N-terminal amine of proximal histone tails are involved in DPCHT
Modeling and Gait Control for Principally Kinematic Locomotion Systems
Get PDFGeometric mechanics provides a framework through which top-down insights permit novel motion planning approaches to a broad class of locomotion systems with symmetry, including analytical computation of optimal gaits. A core premise of this framework is that complex locomotor mechanics can be rewritten in a kinematic form, where momentum effects are negligible (e.g., dissipative forces dominate the physical interaction between the body and the environment). Here, we consider a subclass of such group-invariant dynamical systems; the equations of motion can be kinematically reduced such that the body velocity is expressed as a shape-dependent linear mapping of shape velocity. The same framework has been instrumental in understanding cyclic locomotion where a local model can be constructed in the neighborhood of the observed limit cycle, using data points from stochastically perturbed, repeated behaviors. We aim to tackle practical challenges in implementing data-driven geometric methods, including (1) enhancing the framework's real-time capabilities, (2) further improving data efficiency, (3) providing a principled gait modulation scheme for continuous steering control, and (4) extending the framework to even more general systems.
We have developed an adaptive system identification extension to the current framework that enables the real-time update of the system model that can be used to modify behaviors iteratively within a behavior optimization scheme. This capability not only enhances fundamental behaviors but also enables precise motion tracking. Additionally, we showcase its usage in refining behaviors, aiding in injury recovery, and adapting to different terrains, especially scenarios where simulations (models built from first principles) provide inadequate guidance for real-world situations.
Obtaining a locomotion model efficiently from data, as described above, is an important step that can be leveraged to guide robot behavior, as it facilitates the principled designing of control policies. Acquiring nominal gaits for various behavioral goals is an important step; in practical scenarios, the ability to steer from these nominal gaits is just as important.
We introduce principled gait modulation algorithms designed to modify a nominal gait for single-parameter steering control by constructing a continuum gait library using either global or local model information. This approach opens doors to motion planning and control for systems with complex, a priori unknown dynamics in which an intuitive `joystick' control can be provided.
As soft (and in general underactuated) robots gain popularity, efficiently acquiring useful reduced-order models for these highly underactuated systems becomes increasingly demanding. Geometric methods have great potential in describing the behaviors of many systems within this category. We developed an extension to the data-driven geometric framework that builds layers of interconnected models, which include actuator and locomotor dynamics obtained from data gathered during repeatedly stochastically perturbed behaviors. These linked models are grounded in the general formulations of Lagrangian systems with symmetry, making them suitable for a broad spectrum of robots with first-order, low-pass actuator dynamics, such as hydrogel crawlers powered by swelling-based actuators. These models effectively encapsulate the dynamics of system shape and body movements in a simplified swimming robot model. We also suggested a numerical optimization scheme for control signals via iterative model refinement, which we employed to optimize the input waveform for the hydrogel crawler
Understanding and Predicting Loss to Follow-Up from Rifampicin-Resistant Tuberculosis Treatment in South Africa
No full textTuberculosis is a leading infectious cause of death and of particular concern in countries with a high HIV burden, like South Africa. With only 63% of cases being successfully treated, rifampicin-resistant tuberculosis (RR-TB) is a substantial barrier to TB control. High loss to follow-up (LTFU) rates are a major contributor to sub-optimal treatment success. To meet the WHO’s End TB Strategy target of 90% TB treatment success, predicting and preventing LTFU must be prioritized. Accordingly, this nested retrospective cohort study among people with RR-TB treatment aimed to: (1) develop a point-of-care tool to estimate a patient’s risk for LTFU; (2) examine the relationship between the adverse treatment effects and LTFU; (3) explore the relationship between travel distance to RR-TB care and LTFU; and (4) determine the risk factors independently associated with LTFU from RR-TB care.
We ultimately developed a tool to predict LTFU from RR-TB treatment among people living with HIV based on the following characteristics available at treatment initiation: age, sex, marital status, number in household, government assistance, housing status, history of incarceration, employment status, prior TB outcome, CD4 count, viral load, and ARV regimen. However, this tool demonstrated only fair discrimination, indicating a need for improved prediction before implementation into practice. When exploring the relationships between individual variables and LTFU, adverse treatment events lowered the odds of LTFU. This finding led to the hypothesis that medication adherence may confound the relationship between adverse events and LTFU. Traveling over 60 kilometers to receive RR-TB treatment was also found to increase risk for LTFU, indicating a need for further decentralization of RR-TB services. The final analysis demonstrated that all-oral treatment regimens, higher BMI, and older age were protective against LTFU, while a history of incarceration increased risk for LTFU. Further, this analysis demonstrated a need for investigation into facility-level factors that may impact LTFU.
The results of this study demonstrate the complexity involved in understanding and predicting LTFU. Although we could not adequately predict LTFU from RR-TB care, we identified new risk factors for LTFU, which will assist the development and targeting of future patient engagement interventions