Virginia Tech - Wake Forest University School of Biomedical Engineering & Sciences
VTech Works (Virginia Tech)Not a member yet
122049 research outputs found
Sort by
Targeting S1PR3 to mitigate flow-enhanced invasion in the glioblastoma tumor microenvironment
No full textGlioblastoma is a devastating disease with few effective treatments, in part owed to the dynamic cellular and biophysical factors that influence tumor progression and therapy response. Emerging evidence has implicated pathological interstitial fluid flow, created by high intratumoral pressure relative to the healthy parenchyma, in enhancing cancer invasion. Multiple targetable molecular pathways have been identified that drive this response, but the specific pathways employed by invasive cells differs between patient glioma cell lines. To this end, we sought to identify additional therapeutic candidates mediating flow-enhanced invasion. Our previous work established a role for the G-protein coupled receptor S1PR3 in enhancing invasion under flow. Interestingly, we found this response to be mediated by the brain parenchymal cells, astrocytes and microglia. In this work, we demonstrate clinical relevance for S1PR3 as both a biomarker and therapeutic target with efficacy across a heterogeneous patient cohort. To inform therapeutic development, we investigate the intercellular mechanisms involved in S1PR3-driven invasion. We find that S1PR3 targeting significantly alters flow dynamics in vivo. We connect astrocytic S1PR3 to flow response, finding correlations with flow in both tumor-bearing and tumor-naïve settings, suggesting redundancy across neuropathologies. We build evidence that astrocytic S1P and S1PR3 mediates the response to fluid shear stress and we imply roles for S1P and flow-sensing. This work has exciting implications suggesting a dual role for S1PR3 in flow-regulation and flow-response, thus it may be a doubly effective target for minimizing flow-enhanced glioma invasion.Doctor of PhilosophyGlioblastoma (GBM) is a devastating disease with few effective treatments, in part owed to the diverse effects of the environment surrounding the tumor on tumor progression and sensitivity to therapeutics. Emerging evidence has described how interstitial fluid flow, or the fluid that surrounds the cells in the tissue, is increased in the tumor-adjacent region and causes tumor cells to spread into the healthy tissue where they can evade surgical removal and seed recurrent tumors. The mechanism by which tumor cells respond to increased interstitial fluid flow are therapeutically targetable, but vary between patients. With the goal of identifying additional targetable mechanisms, we demonstrated an environment-driven mechanism whereby the membrane receptor S1PR3 drives GBM spread as a result of increased tumor-adjacent flow. In this dissertation, we demonstrate that S1PR3 is a clinically relevant target, as it is expressed widely in patient samples, predicts survival, and targeting significantly reduces tumor cell spread in models that replicate the tumor-adjacent environment. To better understand this mechanism to inform therapeutic development, we demonstrate that specific brain-resident cells (astrocytes) appear to be responsible for this effect, and that this is further mediated by the binding partner for S1PR3, S1P. Interestingly, we also demonstrate that S1PR3 also promotes elevated fluid flow, presumably upstream of its effects on tumor cell spread. This work has exciting implications suggesting a dual role for S1PR3 in flow-regulation and flow-response, thus it may be a doubly effective target for minimizing tumor progression due to increased fluid flow
Climate
No full textPlace-based climate adaptation workshops are designed to help communities understand their climate-related vulnerabilities and plan adaptive actions in response. Through a series of surveys and interviews with participants, we examined the immediate and long-term impacts of eight place-based climate adaptation workshops in the United States. Six took place online due to COVID-19 restrictions; two took place in-person. All workshops positively enhanced declarative, procedural, and relational knowledge of participants and, to a lesser extent, their personal commitment to work on climate adaptation, optimism about climate adaptation in their communities, and perceptions of qualities of the network of actors engaged locally in climate adaptation. In-person workshops yielded somewhat stronger positive influences on relationship-building than online workshops. Most participants who responded to surveys 6 months to a year after the workshop reported that their workshop had a “minor” to “moderate” impact on stimulating meaningful adaptation actions in their area. Reported actions attributed to the workshops included the incorporation of climate adaptation into formal planning documents, the expansion of adaptation outreach, consideration of climate adaptation in day-to-day planning and decision-making in local government departments, and both successful and unsuccessful grant applications for projects and positions associated with climate adaptation. We describe the workshops’ design, as well as participant assessments of the value of different workshop components. We conclude with lessons learned for future effective workshop planning and design.Published versio
Interpretability and Debugging for Distributed Privacy Preserving Machine Learning
No full textMachine learning systems increasingly rely on privacy-preserving distributed training to leverage sensitive data across multiple organizations without centralization. Federated Learning (FL), a distributed privacy-preserving machine learning paradigm, enables hospitals, devices, and enterprises to collaboratively train models without accessing raw client data (e.g., Siri, Alexa, and healthcare applications). Centralized machine learning benefits from rich debugging and interpretability techniques enabled by transparent access to training data. However, FL removes this transparency, rendering traditional techniques ineffective and making debugging and interpretability a challenging open problem. This thesis addresses this challenge by asking: How can we design automated debugging and interpretability methods for federated learning that effectively localize faults and attribute global model predictions without degrading performance or violating FL's core privacy principles? The central insight is that effective debugging and interpretability can be achieved by analyzing model parameters, activations, and gradients-information already shared or derivable in standard FL protocols (e.g., FedAvg). We present three contributions. First, towards fault localization, we redesign traditional differential testing to operate on neuron activations produced by auto-generated inputs, exploiting the fact that faulty clients produce models with divergent activations. Second, we introduce neuron provenance, which decouples data-influence tracking from data access. It identifies influential neurons via gradient-based weighting and decomposes them to client-specific origins, yielding ranked lists of responsible clients across CNNs and Transformers. Third, we extend neuron provenance to federated LLMs, where autoregressive generation and billion-parameter scale make naive tracking infeasible. It introduces token-level provenance at targeted transformer layers, achieving high attribution accuracy across multiple LLM architectures. In each case, the solution operates entirely on information available at the aggregator, requiring no client-side instrumentation. Collectively, these contributions culminate in practical tools that integrate seamlessly with existing distributed ML workflows, enabling real-time debugging and transparent model insights for both classification and LLMs in FL.Doctor of PhilosophyWhen multiple organizations like hospitals, smartphone companies, or banks want to train machine learning models together, they face a dilemma: combining their data would produce better AI, but sharing sensitive patient records, personal messages, or financial information is often illegal or dangerous. Federated Learning, a privacy-preserving approach to machine learning, solves this by allowing organizations to train AI collaboratively without ever sharing their private data. This technology already powers voice assistants like Siri and Alexa. However, this privacy comes with a hidden cost. When something goes wrong, such as when an AI makes a mistake or produces harmful outputs, developers cannot investigate because they cannot see the original data. Traditional debugging and interpretability approaches, designed for centralized data access, are not directly applicable in this distributed, privacy-constrained setting. This thesis develops new methods that allow developers to identify which organization caused a problem and understand why the AI behaves as it does, all without ever accessing private data. We introduce techniques that work across different types of AI, from image recognition systems to advanced language models. These contributions, now integrated into widely-used federated learning platforms, make privacy-preserving AI more trustworthy and deployable in sensitive domains like healthcare and finance
Rheological and Thermal Insights into Solidification Dynamics in Extrusion-Based Additive Manufacturing of Polymer Composites
No full textPolymer extrusion-based additive manufacturing (EB-AM) is a processing technique that allows building 3D parts by depositing liquified material layer by layer, which solidifies to form a solid structure. Its key advantages over traditional methods like injection molding include design flexibility, the ability to create complex geometries (e.g., lattices, internal channels), and faster prototyping due to the absence of mold fabrication. However, EB-AM also presents challenges, where each layer must solidify sufficiently to retain its deposited shape and support subsequent layers, requiring understanding of the viscoelastic properties of the material and the use of solidification mechanisms to ensure structural integrity during printing.
While solidification favors shape retention, it hinders interlayer chain diffusion, compromising interlayer adhesion and mechanical integrity. Achieving adequate EB-AM processing requires understanding solidification to ensure structural fidelity without hindering interlayer adhesion. This balance is influenced by printer format, with small-format EB-AM promoting rapid cooling and more anisotropy due to higher shear rates, while medium and large-format systems retain more heat, enhancing adhesion at the cost of increased likelihood of shape distortion, especially at high printing speeds (short layer times).
To address the differences in processing at different size scales and printing speeds, a literature review in Chapter 2 explores the differences between process physics in EB-AM across size (from small to large-format printers) and time scales (the effect of printing speeds and layer time) on the printing process, focusing on semicrystalline polymers as feedstock. We explore how polymer physics control process-structure-property relationships at different length- and time scales, focusing on how disparities in shear rates, rheological behavior, and heat retention affect crystallization kinetics and polymer chain mobility, relating it to the development of printed microstructure and interlayer adhesion. Strategies to minimize common issues in printing semicrystalline polymers are also included, for example through material design and processing modifications to reduce volumetric shrinkage and warpage. Lastly, fundamental research gaps in processing semicrystalline polymers in EB-AM are included, focusing on the importance of formulating the next-generation of materials and process monitoring tools to enable facilitated implementation of EB-AM across different size scales.
To complement the thorough discussion of EB-AM of semicrystalline polymers at different size/time scales in Chapter 2.1, a brief review of fundamental aspects related to EB-AM processing of hydrogels through small-format direct ink writing (DIW) is presented in Chapter 2.2. The key rheological requirements for effective processing are highlighted, and a brief explanation of chitosan hydrogels is included, as this is the only non-semicrystalline polymer used in this dissertation.
Chapter 3 continues on EB-AM processing of chitosan with DIW by exploring the rheological predictors of shape retention when printing without the immediate application of a solidification mechanism. This study investigates the rheological behavior of chitosan (CS)-based hydrogels incorporating graphene and titanium dioxide (TiO₂) as functional fillers. By correlating rheological properties with printed bead morphology, we identify tan delta as a key predictor of shape retention, where formulations exhibiting a predominant solid-like behavior (tan delta ≤ 1 at low angular frequencies) demonstrated adequate structural integrity post-deposition. The yield stress of formulations did not correlate with extrusion reliability, which was instead influenced by particle loading. Lastly, the timescale required for formulations to change from predominant solid-like to liquid-like was studied, correlating it with the ability to successfully print a multilayer part. These findings provide valuable insights into rheology-driven design strategies for hydrogel-based inks in DIW, enhancing the effectiveness of polymer composite printing.
In Chapters 4 and 5, the focus changes from chitosan hydrogels to semicrystalline polymer composites, and from small-format EB-AM through DIW to medium-format fused granular fabrication EB-AM, exploring intricacies of structure-process-properties relationships in the solidification behavior of polypropylene (PP) composites and how their properties impact high-speed printing (short layer times).
In Chapter 4, the focus is on the solidification behavior of PP filled with graphite (GR) and carbon fiber (CF), contrasting the effect of medium-format EB-AM processing and fillers on affecting crystallization and physical gelation of the system. We show that fillers aid the physical gelation process by increasing interconnectivity of crystal clusters, especially in CF due to its longer aspect ratio. However, even though fillers facilitate physical gelation, crystallization studies showed that the addition of fillers to the 3D printing grade of PP used in this work slows down the nucleation process, especially GR. Processing history was also shown to negatively affect the ability of the material to crystallize due to thermal degradation. In addition, the effect of medium-format EB-AM processing on the alignment of polymer crystals is studied, indicating less alignment than literature values for small-format EB-AM due to smaller shear rates associated with the process. The results shown in this chapter provide substantial insights into the effect of fillers on the solidification and physical gelation of PP for additive manufacturing, highlighting the importance of filler considerations when designing formulations for EB-AM.
Lastly, Chapter 5 continues to build up the current understanding of medium-format EB-AM of semicrystalline polymers by investigating material requirements necessary to optimize high-speed printing (short layer times) of PP composites. One of the drawbacks of EB-AM of polymer parts is due to long printing times caused by low printing speeds (typically 50–70 mm/s), primarily due to heat accumulation at higher speeds that compromises shape fidelity. This study investigates the influence of CF and GR fillers on the rheological properties and thermal diffusivity of a 3D printing grade of PP, correlating these with shape fidelity at high-speed printing conditions (333 mm/s, 1.1 s/layer). Enhanced shape fidelity in filled samples is attributed to improved heat dissipation caused by higher thermal diffusivities. However, the size of the deposited bead is largely dependent on filler type, being more impactful on heat retention than improvements on thermal diffusivity caused by the inclusion of fillers. In samples prone to more heat retention, improvements in shape fidelity are dependent on larger storage modulus, enabling better shape retention for parts that stay hotter for longer. The findings highlight the critical role of material composition and bead geometry in overcoming speed-related limitations in EB-AM.Doctor of Philosophy3D printing, also known as additive manufacturing, is a process that builds objects layer by layer from digital designs. In extrusion-based additive manufacturing (EB-AM), one of the most common methods, a plastic material is generally heated until it becomes soft, and it is then pushed through a nozzle to form layers. In some cases, gel-like materials are used instead of plastics, which do not require heating for them to flow through the nozzle under pressure. During EB-AM printing, the layers are deposited one on top of another, gradually forming a solid 3D object as the material cools and hardens. This technique allows for the creation of complex shapes and customized parts without the need for molds, making it ideal for rapid prototyping and low-volume production.
Compared to traditional manufacturing methods like injection molding, EB-AM offers greater design flexibility and faster turnaround times. However, it also presents unique challenges. Each layer must solidify quickly enough to support the next one, but not so fast that it prevents the layers from bonding properly. The ability of layers to bond together is directly related to how solid the previous layer is, which affects the strength of the final product. However, if the deposited bead of plastic material stays too hot for too long (or in a predominant liquid-like state in the case of gel-like materials printed at room temperature), the part may lose its shape. This balance between cooling and bonding is influenced by the material's properties, the printer's size, and the speed at which printing occurs.
This research explores how different material formulations and printing conditions affect the solidification behavior of 3D-printed parts. It begins with a broad review of how EB-AM behaves across different printer sizes, from small desktop machines to larger industrial systems, and how printing speed impacts the process. Special attention is given to semicrystalline polymers, a type of plastic that forms crystals as it cools. These materials are widely used but can be difficult to print due to issues like large volumetric shrinkage and warping. The study examines how heat retention, flow behavior, and cooling rates influence the final structure and strength of printed parts and discusses strategies to improve outcomes through material design and process adjustments.
In addition to plastics, the research includes a study on printing soft, gel-like materials called hydrogels using an EB-AM technique called direct ink writing (DIW). These materials are used in biomedical applications and require different printing conditions. The study focuses on chitosan, a natural polymer, and explores how adding particles like graphene and titanium dioxide affects its ability to hold its shape during printing. By analyzing how the material behaves under stress, the research identifies key indicators that predict whether a printed shape will remain stable. These insights help guide the design of better hydrogel formulations for 3D printing.
The final chapters return to plastics, specifically polypropylene (PP), and investigate how adding fillers like graphite and carbon fiber can improve printing performance. The study finds that the way fillers interact with the polymer affects how crystals form during cooling, which in turn influences the solidification behavior of the printed part. These fillers also affect how the material cools down and retains its shape, including during high-speed printing. The research shows that while fillers improve heat dissipation, the type of filler affects the size of the printed bead, influencing how it dissipates heat. In cases where heat builds up, materials with stronger solid-like behavior are better at maintaining shape.
Overall, this work provides a comprehensive look at how materials and printing conditions interact in EB-AM. It highlights the importance of understanding material behavior at different scales and speeds and offers practical strategies for improving print quality. These findings contribute to the development of more reliable and efficient materials for 3D printing technologies, paving the way for broader adoption in manufacturing, biomedical engineering, and other fields
Time-to-Event Prediction Using Deep Learning Models: Application to GPU Failure Data
No full textNeural network models gain significant popularity in recent years due to their ability to identify complex patterns. In the field of reliability research, efforts are made to develop neural network models for predictive reliability. However, research focused on utilizing neural networks to forecast graphics processing unit (GPU) failures remains limited. Analyzing the reliability of GPUs is crucial for effectively maintaining GPU systems and preventing issues related to GPU failures, such as safety concerns and interruptions in simulations. Most studies concentrate on predicting the remaining lifespan of GPUs, whereas our objective is to predict both lifespan and failure status. Furthermore, there is a growing trend of integrating statistical modeling with neural networks. By incorporating statistical concepts, we can create models that better represent reality and improve interpretability. Additionally, the model efficiently manages complex data structures like hierarchical clusters, which enhances its capacity to generalize. To improve our neural network model, we combine statistical concepts with neural network techniques. Building on these motivations, we outline our research approach as follows. We describe the development of deep learning models for GPU failure prediction, and explain how statistical concepts are integrated to improve model performance.
Chapter~ref{cha:genintro} provides a general overview of the research presented in this dissertation. It begins by outlining the motivation behind the study and defining the research problem. Next, it offers a brief summary of the contributions made by this research. The chapter also explains the concept of neural network models, including their training processes. Furthermore, it covers embedding layers and the use of the GPU dataset. Finally, it introduces multitask output for neural networks, which will be utilized in Chapter~ref{cha:NNspatialRE}.
Chapter~ref{cha:DL} introduces a deep learning approach for predicting GPU failure time and status. We propose two distinct neural network architectures to address these prediction tasks: TypeEmbedNet for failure type prediction and TimeEmbedNet for failure time prediction. Since our predictors are categorical variables, we employ embedding layers to transform them into lower-dimensional vectors. We utilize the Cross-Entropy Loss function to optimize TypeEmbedNet and the Mean Squared Error (MSE) for TimeEmbedNet. We develop evaluation metrics to comprehensively assess the models' performance in predicting both failure time and status, taking into account the characteristics of an imbalanced dataset. We integrate data frequency, F1 score, recall, and precision into the MSE by applying penalties based on classification outcomes.
In Chapter~ref{cha:NNspatialRE}, we introduce a deep learning model for competing risks. We develop a custom loss function for competing risks that incorporates survival theory and is specifically adapted for time prediction. We compare this model to other machine learning approaches and benchmark it against a previously introduced parametric model. Additionally, we test the integration of spatial random-effect embeddings to model GPU failure outcomes. To achieve this, we compute correlation matrices based on two different spatial structures—physical and logical distances—and apply them using Cholesky decomposition. We describe how we assign a learnable embedding to each GPU location, incorporating these correlation matrices. The neural network outperforms both the machine learning models and the parametric model across various MSE metrics. However, adding spatial random effects to the neural network does not result in a significant improvement in predictive performance. Chapter~ref{cha:conclusion} summarizes the key findings of this study and explores several potential avenues for future research.Doctor of PhilosophyGraphics Processing Units, or GPUs, are powerful computer processors used in supercomputers, scientific research, engineering tools, and everyday technologies. When a GPU fails, it can interrupt important work, slow down large systems, and increase operating costs. Being able to predict when a failure might happen and what kind of failure it will be helps organizations prevent unexpected outages and keep their systems running smoothly.
This research develops deep learning models that predict both the timing and the type of GPU failure. While many previous studies focus only on estimating how long a GPU will last, this work takes a more complete approach by predicting two outcomes at the same time. Because GPU data contains many categories and complicated patterns, deep learning provides a flexible way to capture these relationships.
To make the models more realistic and useful, this research blends ideas from statistics with modern neural networks. It incorporates concepts from survival analysis to reflect the fact that GPUs can fail for different reasons and that some GPUs do not fail during the time they are observed. The models also include information about where each GPU is located inside the system, which helps reveal whether certain areas experience failures more often.
Several types of models are tested, including common machine learning methods and a traditional statistical model. The results show that the deep learning approach can naturally represent important features of real failure data, such as multiple potential failure causes and incomplete observations. Although adding spatial information does not substantially change predictions, it offers helpful insight into how failures may cluster within the computing system.
Overall, this research demonstrates how combining statistical thinking with deep learning can support better reliability management in large-scale computing environments. The framework developed here can also be adapted to other systems where understanding and anticipating failures is essential
Disease Status, Occupancy, and Relative Activity of Little Brown Bats (Myotis lucifugus) in the Northeastern United States
No full textWhite-nose Syndrome (WNS), caused by the fungal pathogen Pseudogymnoascus destructans, has caused catastrophic declines in several North American bat species, including >90% losses in Little Brown Bats (Myotis lucifugus). Because small, remnant populations persist across the Northeast and Mid-Atlantic, there are key questions about their continued vulnerability to WNS or other pathogens, habitat associations in a post-WNS landscape, and the extent to which population stabilization or recovery may already be underway. This dissertation integrates disease surveillance, long-term acoustic monitoring and Bayesian occupancy and activity modeling to evaluate population trajectories and ecological drivers influencing Little Brown Bat persistence across multiple spatial scales. Chapter 1 of this dissertation explores the potential introduction of SARS-CoV-2 into Little Brown Bat populations throughout the Northeast and Mid-Atlantic United States. We analyzed saliva samples from 235 individual little brown bats from a total of eight maternity colonies. No bat tested positive for SARS-CoV-2 by RT-qPCR, indicating the virus was either not present or that it persists in undetectable levels in little brown bat populations in this region. The second chapter of this dissertation explores the trends of WNS affected bat species in the mid-Atlantic, including Little Brown Bats, Northern Long-eared Bats (Myotis septentrionalis), and Tricolored Bats (Perimyotis subflavus), using relative activity models using acoustic data collected from the Chesapeake and Ohio National Historical Park in Maryland along the Potomac River Corridor. Little Brown Bats and Tricolored Bats showed increasing activity over time and strong positive associations with proximity to water, and wetlands, consistent with pre-WNS habitat use and may suggest early regional stabilization. Northern Long-eared Bats, however, remained rare with no clear habitat associations, mirroring severe ongoing declines across their range. The third chapter of this dissertation explores patterns of Little Brown Bat occupancy and relative activity throughout New England. Occupancy models indicated largely stable site use with weak positive relationships with year and wetland cover. In contrast, relative activity displayed stronger and more informative patterns, including significant increases over time and positive associations with forest cover, and wetland area, providing greater insight into patterns missed by occupancy alone.Doctor of PhilosophyAn invasive fungus that causes White-nose Syndrome (WNS) in bats is responsible for the decline of several North American bat species. One of these species, the Little Brown Bat (Myotis lucifugus), may be starting to recover in some areas of its range, specifically the Northeast, where WNS was first discovered. This has led to questions about population trends over time, as well as concerns regarding the potential introduction of SARS-CoV-2, the virus responsible for the COVID-19 pandemic. It is the goal of this dissertation to address these concerns through disease monitoring, and the use of acoustic data to assess population changes and ecological associations that may support recovery for this species. After surveying 235 Little Brown Bats at eight maternity colonies throughout the Northeast and Mid-Atlantic United States, we did not find evidence of the SARS-CoV-2 virus in our sampled individuals. We used bat call data gathered from the Chesapeake and Ohio Canal National Historical Park using acoustic monitoring devices to determine changes in call numbers (relative activity) from three WNS-affected bat species including Little Brown Bats, Northern Long-eared Bats (Myotis septentrionalis), and Tricolored Bats (Perimyotis subflavus). We found increases in Little Brown Bat and Tricolored Bat relative activity over time. These species were also more active closer to water and in areas with wetlands. The number of calls for the endangered Northern Long-eared Bat were extremely low, and likely because of this, we found no strong ecological associations. We also used acoustic monitoring to analyze Little Brown Bat trends throughout New England. We used both presence/absence data (occupancy), and number of calls at each acoustic monitoring site (relative activity) to assess Little Brown Bat recovery and ecological relationships. We found that although occupancy analysis showed modest increases in occupancy probability over time, as well as a weak association with wetlands, our relative activity analysis showed much clearer patterns, including an increase in relative activity over time and strong associations between relative activity and ecological factors, such as wetlands and forest cover
Predicting the thermal performance of bio-based cold chain packaging system
No full textCold chain logistics play a critical role in ensuring the safe transport of temperature-sensitive products such as pharmaceuticals, biologics, and perishable foods. Maintaining stable internal temperatures within insulated shipping containers (ISCs) requires an in-depth understanding of how materials, design, and environmental factors influence heat transfer. This research combines experimental and computational approaches to improve the thermal efficiency and environmental sustainability of passive cold chain packaging systems.
The first phase of the study (Chapter 1) focuses on predicting the thermal performance of bio-based ISCs through finite element modeling (FEM). Material characterization was conducted using Differential Scanning Calorimetry (DSC) and Heat Flow Meter techniques to obtain the thermal properties of corrugated fiberboard, honeycomb paperboard, and phase change materials (PCMs). The FEM framework was validated through experimental data, showing strong correlation with measured results and a mean prediction deviation of less than 8% when maintaining temperatures below the critical 8 °C threshold. These findings confirm that FEM can serve as an accurate and efficient alternative to conventional performance testing while supporting the integration of renewable insulation materials in package design.
The second phase (Chapter 2) examines how environmental humidity influences the thermal behavior of ISCs. Laboratory experiments were performed across relative humidity levels from 30 % to 80 % to evaluate temperature evolution and hold-time performance. The results revealed that higher humidity significantly accelerated the warming rate, particularly in fiber-based insulation systems, due to moisture absorption that increased effective thermal conductivity. In contrast, polymer-based materials such as expanded polystyrene (EPS) and polyurethane (PU) remained relatively stable. Energy-balance modeling supported these observations, confirming humidity as a major external driver of heat transfer in porous materials. Beyond performance, the study underscores the environmental benefits of fiber-based materials, which are renewable and recyclable, while emphasizing the need for design strategies that balance thermal reliability and sustainability under real-world humidity conditions.Master of ScienceCold chain packaging helps keep medicines, vaccines, and food products safe by maintaining low temperatures during shipping. However, designing boxes that stay cold long enough while also being environmentally friendly is a major challenge. This research explores how different materials and environmental conditions affect the performance of insulated boxes used in cold-chain transport.
In the first part of the study, computer modeling and laboratory tests were used to predict how well various paper-based insulation materials keep products cold. The models closely matched real experimental results, showing that computer simulations can accurately estimate temperature performance without the need for long and expensive tests. These findings also support using bio-based materials such as corrugated fiberboard and honeycomb paperboard as sustainable alternatives to conventional foams.
The second part of the study looked at how humidity in the air changes how these boxes perform. Tests showed that when humidity is high, paper-based materials absorb moisture and lose some of their insulating power, causing the inside temperature to rise faster. Plastic foams such as EPS and polyurethane were less affected. Although fiber-based packaging is more sensitive to humidity, it offers major environmental advantages because it comes from renewable sources and can be recycled. Together, these results show that future cold chain packaging can be both efficient and eco-friendly if designed with careful attention to how materials behave under real shipping conditions, including humidity
Minerals
No full textOccupational exposure to respirable coal mine dust remains a significant health risk, especially for underground workers. Rapid dust monitoring methods are sought to support timely identification of hazards and corrective actions. Recent research has investigated how optical light microscopy (OLM) with automated image processing might meet this need. In laboratory studies, this approach has been demonstrated to classify particles into three primary classes—coal, silicates and carbonates. If the same is achievable in the field, results could support both hazard monitoring and dust source apportionment. The objective of the current study is to evaluate the performance of OLM with image processing to classify real coal mine dust particles, employing scanning electron microscopy with energy-dispersive X-ray spectroscopy (SEM-EDX) as a reference method. The results highlight two possible challenges for field implementation. First, particle agglomeration can effectively yield mixed particles that are difficult to classify, so integration of a dispersion method into the dust collection or sample preparation should be considered. Second, optical differences can exist between dust particles used for classification model development (i.e., typically generated in the lab from high-purity materials) versus real mine dust, so our results demonstrate the necessity of site-specific model calibration.Published versio
PLoS One
No full textSexual dimorphism in bats is understudied, with conflicting evidence across species and geographic regions. For Desmodus rotundus, the common vampire bat, previous reports on morphological sex differences have been inconsistent. This study aimed to assess sexual dimorphism in D. rotundus using a combination of contemporary field measurements and historical museum specimens. We analyzed six morphometric traits, including body mass, head length, body length, tibia length, ear length, and forearm length. Data were collected from 46 wild-captured individuals from five locations across Colombia in South America. Additionally, forearm length was examined in an expanded dataset of 490 specimens, including additional 444 individuals from museum vouchers collected over the past century. Principal components analysis and hierarchical clustering of the six-trait dataset showed patterns of differentiation between sexes, with partial overlap. Forearm length, analyzed independently in the full 490-specimen dataset, showed strong evidence of sexual dimorphism. Females had significantly longer forearms (mean = 61.8 mm) than males (mean = 58.5 mm), with non-overlapping 95% confidence intervals and a highly significant t-test result (t = -12.68, p < 2 × 10 ⁻ ¹⁶). Sex explained 25.7% of the variation in forearm length (R² = 0.26). Tibia length also differed significantly between sexes of the wild-catch individuals (p = 0.004), with females exhibiting greater values. Comparisons between museum specimens (historical) and wild-caught specimens (contemporary) showed no significant differences across time in either sex. Among females, the difference was not significant (t = -0.93, df = 208, p = 0.355), and the same was true for males (t = -0.01, df = 278, p = 0.992). A follow-up MANOVA on the six morphometric traits indicated a significant effect of sex (Pillai's trace = 0.389, approx. F(6,39)=4.14, p < 2.2 × 10 ⁻ ¹⁶). After correcting for multiple comparisons, significant sexual dimorphism remained for forearm and tibia lengths, with forearm showing the strongest signal. These findings provide robust support for modest but consistent female-biased dimorphism in D. rotundus. The use of both multivariate and univariate analysis, combined with long-term historical data, enhanced the reliability of signals detected regarding morphological differences. Desmodus rotundus play a role as a primary reservoir for zoonotic viruses, has potential relevance in biomedical research, and provides ecosystem services. Understanding sex-based morphological variation is critical to inform public health, ecology, and biological conservation strategies. Females were consistently larger than males, but segregation was not absolute, with some individuals falling outside the expected data range for their sex. This study contributes to a clearer understanding of morphological variation and lays the groundwork for future research into the ecological and evolutionary drivers of dimorphism in bats.Published versio
Discovering Viral Hosts, Mutations, and Diseases using Machine Learning
No full textThe discovery of a novel virus raises three important questions, namely, which host(s) can the virus infect, what mutations in the virus could affect its interaction with its hosts and enable a host-shift, and which diseases can the virus cause in humans. We propose novel machine learning (ML)-based solutions to these three different problems in computational virology.
(i) We develop a viral protein language model for predicting the host infected by a virus, given only the sequence of one of its proteins. Our approach, 'Hierarchical Attention for Viral protEin-based host iNference (HAVEN)', includes a novel architecture comprising segmentation and hierarchical self-attention to tackle the challenges posed by long sequences. Pretrained on 1.2 million viral protein sequences, the model accepts any protein sequence of any virus and predicts its host. We integrate HAVEN with a prototype-based few-shot learning (FSL) classifier to generalize it to predict rare and unseen hosts, and hosts of unseen viruses.
(ii) Structured datasets of known viral mutations and their effects are required to develop computational models that can predict potential detrimental changes in novel animal viruses. We leverage large language models (LLMs) to create these datasets from the results of experimental studies available as unstructured text in scientific literature. We design an open-ended task for 'scientific information extraction (SIE)' from publications and propose a unique two-step retrieval augmented generation (RAG) framework for the same. We curate a novel dataset of mutations in influenza A viral proteins. We use this dataset to benchmark our proposed approach, a wide range of LLMs, RAG-, and agent-based tools for SIE.
(iii) Finally, we look at the effects of viral infections in humans. Specifically, we focus on the long-term effects of SARS-CoV-2 (or long COVID) wherein patients experience the persistence of COVID-19 symptoms for a long period of time after their initial SARS-CoV-2 infection. We propose an ML-based classification pipeline to predict the diagnosis of long COVID in COVID-19 patients using their electronic health records (EHRs) in the National COVID Cohort Collaborative, which is the largest collection of clinical data across the US. Using techniques to explain our models' prediction for each patient, we uncover many features that were correlated with long COVID. We also evaluate the impact of different data sources on our long COVID prediction models using a novel a cross-site analysis.Doctor of PhilosophyViruses are one of the primary pathogens causing infectious diseases. There is a rise in the frequency of outbreaks of human infectious diseases across the globe. Several viruses originate in animals, evolve though mutations, and shift hosts to infect humans. It is important to detect the potential of animal viruses to infect humans in order to avoid, prepare for, and tackle future infectious disease epidemics through well-informed decisions. We propose novel artificial intelligence (AI)-based solutions to three important questions namely, which host(s) can a virus infect, what mutations in the virus could affect its interaction with hosts, and which diseases can the virus cause in humans.
We develop "Hierarchical Attention for Viral protEin-based host iNference (HAVEN)" based on the architecture of large language models (LLMs) such as ChatGPT. We train HAVEN to learn the properties of protein sequences of viruses and predict their hosts. HAVEN can also identify rare, unseen hosts and predict hosts of unseen viruses. Next we focus on the mutations in a virus that allow it to shift from one host to another and infect humans. Results from experimental studies analyzing the effects of viral mutations on virus-host interaction are available primarily in the form of unstructured text in scientific publications. We seek to employ LLMs to retrieve this information from the scientific literature and create these datasets. Retrieval augmented generation (RAG) is framework where an AI system first retrieves relevant information from a provided source and leverages it to generate accurate answers. We design a novel task for LLMs to perform 'scientific information extraction (SIE)' from publications and propose a unique two-step RAG framework for the same. We manually curate a novel dataset of mutations in influenza A viral proteins. We use this dataset to benchmark our proposed approach, a wide range of LLMs, and state-of-the-art RAG-based methods for SIE.
Finally, we focus on the long-term effects of SARS-CoV-2. Long COVID is a disease condition wherein patients experience the persistence of COVID-19 symptoms for a long period of time after their initial COVID-19 infection. We trained prediction models using electronic health record (EHRs) of COVID-19 patients from during their infection phase. We show that these machine learning models can effectively predict the future occurrence of long COVID, generalize to different sources of EHR data, and highlight informative indicators in EHRs for early diagnosis. The contributions in these thesis are aimed towards developing a coherent system for pandemic preparedness and prevention