143174 research outputs found
Sort by
Multiscale characterization of sedimentological heterogeneity in the Bunter Sandstone Formation with application to CO2 storage
No full textThis study develops a multiscale reservoir characterization framework for the Bunter Sandstone Formation, a Triassic fluvial succession widely targeted for CO2 storage across northwestern Europe, including at the Endurance CO2 storage site, offshore UK. Facies and facies association analysis of core and outcrop analogues, photomosaic analysis, minipermeameter measurements, thin section petrography, spatial statistical analysis, and numerical modeling have been integrated to characterize heterogeneity at scales ranging from lamina (mm) to channel fill storeys (100s m). Thirteen lithofacies were identified and grouped into three facies associations, and three architectural elements were defined based on the stacking patterns and geometrical configuration of lithofacies. Facies permeability varies over three orders of magnitude (0.18–5400 mD), primarily controlled by grain size, clay content, and cementation. Representative Elementary Volume (REV) analysis showed effective permeability for different lithofacies stabilizes at scales much larger than typical core plugs and is highly anisotropic. Effective permeability correlates linearly with the proportion of high-permeability (clay-poor) lithologies in each facies. The resulting values of effective permeability for different facies were populated in architectural element-scale models. Simplifying these models of architectural elements by grouping lithofacies into two or three permeability-based categories resulted in estimated values of equivalent permeability at this scale that are within 10-20% of those for models containing all the constituent lithofacies. Building on the result that a two-facies representation is sufficient to capture heterogeneity at the architectural-element scale, this approach was extended to the channel-fill storey scale by characterising 100s m scale outcrop faces as sandstone containing thin, discontinuous mudstone bodies. Spatial statistical analysis shows these mudstone lenses are randomly distributed and occur at the scale of 3rd- and 4th-order erosional elements, allowing their influence to be evaluated using a streamline-based statistical method. These mudstone lenses reduce effective vertical permeability to ~3 % of the horizontal value.Open Acces
Accurate and generalizable protein-ligand binding affinity prediction with geometric deep learning
No full textGoal: Protein-ligand binding complexes are ubiquitous and essential to life. Protein-ligand binding affinity prediction (PLA) quantifies the binding strength between ligands and proteins, providing crucial insights for discovering and designing potential candidate ligands. While recent advances have been made in predicting protein-ligand complex structures, existing algorithms for interaction and affinity prediction suffer from a sharp decline in performance when handling ligands bound with novel unseen proteins. Methods: We propose IPBind, a geometric deep learning-based computational method, enabling robust predictions by leveraging interatomic potential between complex’s bound and unbound status. Results: Experimental results on widely used binding affinity prediction benchmarks demonstrate the effectiveness and universality of IPBind. Meanwhile, it provids atom-level insights into prediction. Conclusions: This work highlight the advantage of leveraging machine learning interatomic potential for predicting protein-ligand binding affinity.
Index Terms—Deep learning, drug discovery, physics-informed neural networks, protein-ligand binding affinity prediction.
Impact Statement–This study extends state-of-the-art deep learning algorithms to applications in protein-ligand binding affinity prediction. This study has implications for enhancing the generalization capability of protein-ligand interactions prediction methods by interatomic potential modeling
Impact of Adjuvant Chemotherapy on Survival Outcomes in Intermediate- and High-Risk ER+/HER2- Breast Cancer Stratified by Genomic Profiling: A Meta-Analysis
No full text
Physics-guided impact localisation and force estimation in composite plates with uncertainty quantification
No full textPhysics-guided learning offers a promising pathway for accurate impact identification in composite structures under sparse experimental data. This paper presents a hybrid framework for impact localisation and force reconstruction that integrates a data-driven First-Order Shear Deformation Theory (FSDT) model with probabilistic machine learning and uncertainty quantification. Structural material properties and boundary conditions are identified from dispersion relations and modal characteristics, enabling construction of a low-fidelity but physically consistent FSDT model directly from measured responses. Physics-augmented low-fidelity time-difference-of-arrival data are fused with sparse experimental measurements using multi-fidelity Gaussian Process Regression to achieve accurate, uncertainty-aware impact localisation. Impact force reconstruction is performed via transfer-function-based deconvolution with an adaptive regularisation scheme informed by FSDT interpolation errors. Experimental validation on a composite flat plate demonstrates accurate localisation and force reconstruction under sparse training data, while additional localisation results on stiffened and sandwich composite panels confirm the generalisability of the localisation framework to more complex structures. The proposed approach provides a computationally efficient and physically interpretable solution for uncertainty-aware impact monitoring in composite aerostructures
Exact solution for the electric field associated with charge caps on a leaky dielectric droplet at high electric Reynolds number
No full textIn a recent paper [Peng et al., Phys. Rev. Fluids 9, 083701 (2024)], the phenomenon of electric charge cap formation on the surface of a leaky dielectric droplet in an electric
field was identified. The surface charge caps, which form at the droplet poles when charge relaxation is faster inside the droplet phase than in the ambient, are stagnant and perfectly conducting. Using asymptotic methods, a nonstandard, inhomogeneous, mixed boundary value problem satisfied by the droplet field potential in this high electric Reynolds number limit was derived by Peng et al. and then solved numerically. The present article gives
an exact solution to that mixed boundary value problem. This solution resolves fully the square-root singularities in the potential at the edges of the electric caps that can cause convergence difficulties in numerical schemes. Knowledge of the solution in analytical form is valuable because it feeds into a nonlinear problem for the coupled hydrodynamics. Moreover, the theoretical approach here is potentially extendible to finding the fields associated with electric charge cap formation in other setting
Are language models intelligent enough for entrepreneurial work? A language-centered perspective
No full textLarge Language Models (LLMs) are poised to fundamentally reshape entrepreneurial work, but it remains unclear whether this technology can support judgment-intensive entrepreneurial tasks. Prevailing skepticism holds that LLMs are inherently unreliable for such deep augmentation because, despite their language competence, they do not think. In contrast, we draw on Ludwig Wittgenstein and Alan Turing to advance a language-centered perspective on entrepreneurial work. Wittgenstein demystifies thought as linguistic activity and treats reasoning and understanding as linguistic abilities exercised in thinking. Extending this stance to the domain of machine intelligence, Turing grounds claims about intelligence in testable performances of language use. Together, they enable us to (1) conceptualize LLMs as an epistemic technology whose linguistic competence may suffice for the deep augmentation of entrepreneurship and (2) reorient research from skepticism toward fine-grained Turing tests of entrepreneurial work. We illustrate and support the language-centered perspective through two studies on crafting effective entrepreneurial narratives, a judgment-intensive task. Initially, we document that the LLM competently blends expert rhetorical strategies to create and refine narratives that effectively align with stakeholder needs. We then experimentally demonstrate that, when coupled with stakeholder-guided iterations, LLMs produce measurable improvements in narratives tailored to distinct stakeholder priorities. More broadly, our rethinking of entrepreneurial work through language-centered lenses helps theoretically support bold predictions about what entrepreneurs can accomplish with a nonhuman intelligence that has “only” mastered human language
Antepartum exposure to greenness, air pollution, and temperature and outcomes of preterm infants
No full textImportance Existing evidence shows that the environment is a risk factor in preterm birth. However, more evidence is needed about the association between environmental exposures and outcomes of preterm birth.
Objective To examine the association of antepartum exposure to individual and combined indices of greenness, air pollution, and extreme temperatures with outcomes of preterm infants.
Design, Setting, and Participants This cohort study used linked national databases of the Canadian Neonatal Network and Canadian Urban Environmental Health Research Consortium. The cohort consisted of infants born at 22 to 28 weeks plus 6 days’ gestation between January 1, 2010, and December 31, 2020, and treated in tertiary neonatal intensive care units (NICUs) across Canada. Data were analyzed between May 1 and September 15, 2024.
Exposure Indices summarizing levels of greenness, air pollutants, and ambient temperature at mothers’ residential postal code at birth.
Main Outcome and Measure Infant survival without major morbidity (SWMM) assessed at death or discharge from the NICU. Logistic regression models estimated associations between individual and combination of indices with SWMM.
Results A total of 14 748 infants (7965 males [54.0%]; mean [SD] gestational age, 26.1 [1.6] weeks; median [IQR] birth weight, 890 [720-1090] g) were included. The rate of SWMM was 32.1% (4737 of 14 748). Infants born to mothers who were exposed to high ozone levels had lower odds of SWMM vs infants with mothers exposed to low ozone levels (adjusted odds ratio [AOR], 0.83; 95% CI, 0.74-0.95). Infants with antepartum exposure to low temperature combined with either high levels of ozone (AOR, 0.76; 95% CI, 0.60-0.95) or low levels of greenness (AOR, 0.77; 95% CI, 0.60-0.99) or both low levels of greenness and high levels of ozone (AOR, 0.58; 95% CI, 0.43-0.77) had lower odds of SWMM than infants with mothers who were not exposed to these risk factors.
Conclusions and Relevance In this cohort study, antepartum exposure to high ozone and combined low temperatures and low levels of greenness or high ozone were associated with lower odds of SWMM in preterm infants. These findings suggest that the health outcomes of antepartum environmental exposure extend to neonatal outcomes of preterm infants
Development of nanoporous packaging for improving stability and pulmonary delivery of vaccine formulations
No full textNucleic acid vaccines carried by lipid nanoparticles (LNPs) are a desirable pathway to address worldwide outbreaks, but the required cold-chain storage causes high cost. Functionalised lipid nanoparticles (fLNPs) paired with silica-based nanoporous packaging were explored to improve their thermostability at ambient temperature.
Silica substrates were used to load LNPs coated with amphiphilic, anionic polymers PP75 that enable pH-triggered endosomal escape. Quartz Crystal Microbalance with Dissipation (QCM-D) revealed that a higher temperature was favourable for the PP75 adsorption. There was a micelle-to-unimer change with increasing temperature, and the polymer behaviour also depended on the charge and hydrophobicity of the surface. Similarly, the LNPs with the PP75 coating had the same response to the temperature change as compared to free PP75.
Nanoporous silica with alumina or gold coatings to adjust surface properties was engineered by controlling pore diameter, depth, and distribution to capture nanocarriers during the loading process. On alumina surfaces, by altering the pH from 5.5 to 7.0, over 30% of the fLNPs were unloaded measured by QCM-D, displaying controllable storage at the nanoscale.
Based on fLNP delivery insights, lung-endogenous sugars were evaluated as alternative carriers for pulmonary administration. The aerosol performance of dry powders was optimised by ball milling and blending lactose of various sizes, achieving 70% in fine particle fraction (FPF). The effects of particle size, contact surface and relative humidity (RH) on the triboelectrification behaviour were studied. Pure crystalline lactose acquired a negative charge on stainless steel (SS), polyethylene (PE) and polypropylene (PP), while the selected dry powder of multiple components obtained a positive charge.
Overall, silica-based nanopackaging offers a viable solution for the instability of fLNPs at room temperature and their agglomeration. QCM-D effectively guides substrate selection and monitors interfacial interactions in a changeable environment. Furthermore, pulmonary administration would be a promising route to carry LNP vaccines.Open Acces
Multi-agent distributed optimisation in adversarial environments
No full textThe extensive computational and communication capabilities of Cyber–Physical Systems introduce new challenges in designing distributed algorithms for cooperative and competitive smart devices. A major challenge in distributed systems is ensuring security, particularly in the presence of malicious or self-interested agents. Malicious agents may intentionally inject noise into critical information, such as gradient updates in distributed optimisation processes. Self-interested agents can strategically manipulate the transmitted information to optimise their individual benefits at the expense of global objectives. These behaviours disrupt coordinated actions, thus addressing such issues is crucial for maintaining system efficiency and achieving reliable and resilient optimisation in adversarial environments.
To address these challenges, this thesis is structured around two topics. The first one focuses on designing a resilient filter for convex distributed optimisation algorithms to mitigate the impact of adversarial gradients. The second topic aims at aligning individual and societal interests using incentive mechanisms. We first investigate the cyclical monotonicity property of convex functions, based on which a gradient filter is designed. We discuss that this filter can either be used for detecting adversarial behaviour or for recovering convexity from the attacked data and therefore recovering convergence. We argue that to avoid being detected by the filter, adversarial agents alter their local objective functions and pretend to be regular agents. We then propose an incentive mechanism with an induced game, in which an ε-dominant strategy equilibrium of the system is obtained when all agents truthfully use their local objective functions. We further improve the computational efficiency of the distributed implementation of the mechanism by designing cutting plane-based algorithms. A case study is provided to validate the performance of the proposed filter, mechanism and algorithms.Open Acces