Mason Journals (George Mason Univ.)
Not a member yet
3256 research outputs found
Sort by
Extracellular Vesicles Provide a Real-Time Recording of Intracellular Signaling: EGFR receptor presence and implications for personalized therapy
No full textThe epidermal growth factor receptor (EGFR) family, particularly EGFR and HER2, plays a critical role in cell signaling related to proliferation, survival, and progression of breast cancer. To better understand the causes of cancer aggressiveness, the extracellular signaling activity of EGFR was investigated in triple-negative breast cancer (TNBC) 4T1 cells, which, despite their classification, display wild-type EGFR and HER2 receptors similar to double-positive BT-474 breast cancer cells. Phosphorylation of cytoplasmic regions of EGFR and HER2 strongly correlate with patient susceptibility to anti-HER therapy. HER2 and EGFR signaling at the cell surface is followed by movement of the receptor complex into endosomes. It is unclear whether these endosomes become extracellular vesicles (EVs) that reflect the internal signaling events of the cells that release them. EVs may offer a way to eavesdrop on the ongoing internal signaling activity of the HER family receptors and could potentially be used to individualize therapy. 4T1 cells were treated with EGF, then EV isolation was performed through centrifugation, and remaining cells were lysed. EVs were purified using IZON columns before Western Blot analysis. Phosphorylation of EGFR (Y1068) and HER2 (Y877) was measured at different times after ligand stimulation and compared to the EVs released by the same cells. p-EGFR was clearly elevated in the EVs in a dose- and time-dependent manner. These results suggest that the phosphorylated state of HER family receptors in EVs can reflect internal signaling. This finding has clinical implications and may help guide future personalized therapies using EVs from tumor biopsies or blood samples
Can Intel’s Trusted Domain Extensions (TDX) support secure and reliable replication of academic papers in confidential cloud environments?
No full textAcademic replication packages often fail to run reliably across different computing environments due to software dependencies and documentation issues. This project evaluates whether Intel’s Trusted Domain Extensions (TDX) can support secure and reliable replication of academic papers in confidential cloud environments. Two recent Management Science studies from 2022 were tested using TDX-enabled virtual machines on Google Cloud. One package failed due to a missing dataset (Hotels_MW.dta) and reliance on Stata-only commands. No response was received from the authors. The other package was written in MATLAB and had to be converted to Octave. The code broke repeatedly due to nested functions and dimension mismatches. After rewriting key functions and adjusting test scripts, the model ran to completion inside the TDX VM. It used 184 GB of memory, ran for about 20 minutes, and cost under $5. Results matched those reported in the paper. This shows that TDX can be used for secure replication when code is self-contained, but packages that rely on closed data or platform-specific tools still remain a challenge
An Investigation into the use of Newspaper Data and Large Language Models for Understanding the Occurrence of Flooding in the Caribbean
No full textFlooding remains the most prevalent weather‐related disaster worldwide, causing over $10 billion in annual property damage and significant loss of life—particularly in low‐and-middle‐income countries (LMICs), which account for nearly 90 % of those affected. Traditional flood‐defense projects and early‐warning systems, while effective, often far exceed the financial capacity of LMICs. To help prioritize limited resources, we propose leveraging local newspaper archives—an abundant, underutilized data source—to identify and map flood‐prone areas dynamically. In this study, we extend our previous analysis of Trinidad and Tobago’s two leading newspapers (Trinidad Express and Trinidad Newsday) through the end of 2024, increasing our dataset to over 5500 extracted flood‐event records. We applied four large language models (LLMs) for automated information extraction: ChatGPT o3, Anthropic's Claude 4, Microsoft's Phi4, and Meta's Llama 4. Both ChatGPT and Claude achieved over 90 % accuracy in detecting flood mentions and correctly geocoding their locations, while Phi 4 and Llama 4 reliably identified flood events but struggled to assign precise geographic coordinates. By combining the high‐precision extractions of ChatGPT and Claude with GIS visualization, we generated interactive maps that reveal temporal and spatial patterns of flooding—highlighting the communities most at risk and the predominant causes (e.g., river overflow, infrastructure failure, extreme rainfall) in each area. These results demonstrate that, even with modest budgets, governments and disaster‐management agencies in LMICs can harness natural‐language processing on freely accessible local news to target interventions where they are needed most, thereby reducing both economic losses and human suffering
Gamifying Environmental Education Through Interactive Simulation and Game Design
No full textSerious games are emerging as effective tools for environmental education, allowing players to explore complex systems interactively. Topics such as urban planning and sustainability involve tradeoffs in areas such as economic and social policy, and traditional instruction often struggles to convey how interconnected ecological and urban systems are. Conversely, game-based simulations offer an intuitive solution for demonstrating these real-world challenges in a way that makes educational content more engaging and accessible. This project introduces Carbon City, a Unity-based game that challenges the player to build a sustainable metropolis. Unlike traditional city-building games, Carbon City introduces decision-making that forces the player to be climate-conscious while taking economic and social factors into account. Players must reach a population of 1 million by 2050 while maintaining public support and keeping carbon emissions low. The game includes a grid-based building system, power distribution, adjacency bonuses to reward walkable cities, green technologies to research, and a random event system that presents policy dilemmas with consequences. By simulating interconnected urban systems, Carbon City aims to promote systems thinking and understanding of sustainability tradeoffs. Preliminary data suggests Carbon City players experienced a 7.387% increase in their understanding of the carbon consequences for major industry changes after playing. In the post-playthrough survey, 88.89% of players also reported that Carbon City made them more aware of sustainability concepts and environmental issues. While formal data collection is still underway, early responses show potential for the game to be effective in environmental education
Design and Implementation of a Variable-Sweep Unmanned Aerial Vehicle for Ground Target Acquisition and Terminal Guidance Using Computer Vision Modeling
No full textOver the past several centuries, technology on the battlefield has progressed rapidly, with new capabilities able to eliminate targets from ranges never thought possible. However, as we move further into the Artificial Intelligence Era, a new type of weapon is emerging- unmanned aerial vehicle (UAV) drones. These lightweight infantry-based machines are able to target and strike with exceedingly high precision, often without fear of loss due to their low cost and maintenance. However, this area of development has been lacking, with more conventional weapons taking development time and research. This paper presents the design, implementation, and evaluation of a UAV drone capable of autonomously tracking and targeting ground locations. We explore how varied computer vision You Only Look Once (YOLO) models and Proportional Integral Derivative (PID) control values, used in conjunction with a variable-sweep wing UAV, affects targeting precision and accuracy when tracking moving ground targets under simulated aerodynamic conditions. We also explore which parameters result in improved tracking in comparison to human piloting. Testing is conducted through tracking tests aimed for single axis correction
Spatiotemporal Modelling and Prediction of California Wildfires using Machine Learning and Environmental Data
No full textIn 2024, California experienced over eight thousand wildfires in total burning around one million acres. In fact, California experiences the most destructive wildfires in the United States, driven by a complex interplay of environmental, climatic, and human factors. This project aims to develop a predictive framework for identifying high risk wildfire events using historical wildfire datasets. By leveraging machine learning models such as random forest regression, support vector machines, decision tree, gradient descent, etc., the project assesses the likelihood and severity of fire occurrences based on features like, area burned, structures destroyed and damaged. The project attempts to provide visualizations of losses due to these wildfires through an interactive dashboard created using Streamlit in python that enables users to explore spatial and temporal patterns of wildfire susceptibility. Random Forest regression model estimates area burned based on user input data, including latitude, longitude, month, and day, and structures destroyed. The dashboard allows real time predictions alongside dynamic heatmaps showing the density and severity of wildfires. Dimensionality reduction techniques such as Principal Component Analysis (PCA), are applied to uncover essential features that help to improve model efficiency and interpretability. A custom Multifeature Impact Score is created to quantify the impact of each fire by combining key metrics into one easily readable outcome prediction. This enables users to compare the severity of different fires and regions. The dashboard also includes seasonality trend analysis to visualize when wildfires are most common and most destructive, as well as key metric breakdowns by county and year. Together, these features offer a clear, user driven way to reveal patterns in wildfire behavior over time and space. The methodology and tools designed in this work is to raise awareness, support data driven climate discussions, and highlight the increasing threat of wildfires, in alignment with the UN Sustainable Development Goal 13 for Climate Action. The predictive approach presented in this work supports proactive wildfire management, resource allocation, and mitigation planning, in a rapidly changing climate. 
A Comparative Evaluation of TabNet-GPR and GNN-LSTM for Subseasonal Fire Radiative Power
No full textWildfire prediction models remain unreliable for subseasonal forecasting, contributing to 8.9 million acres burned in the U.S. in 2024. To combat this, two pipelines were compared for predicting Fire Radiative Power (FRP): a spatio-temporal Graph Neural Network (GNN) + Long Short-Term Memory (LSTM) model and a data-driven TabNet neural network model + Gaussian Process Regressor (GPR). Bayesian Optimization was applied to both pipelines to improve forecasts with minimal preprocessing, using 2-day lagged inputs from satellite, weather, and fire data. The GNN + LSTM modeled spatial and temporal patterns via graph embeddings and recursive rollout. Each node represented a fixed location using historical Latitude and Longitude data. Meanwhile, the TabNet + GPR pipeline used residual learning to refine predictions. Models were evaluated on a ~1GB sample using MAE, RMSE, and R² against three baseline models: persistence, climatology, and linear regression. The TabNet + GPR pipeline reduced MAE by 16% compared to the best baseline (0.1408 vs. 0.1668) and RMSE by 70.3% over default TabNet (0.2840 vs. 0.9758). While R² was low (max 0.0567) and unstable (NaN in the full pipeline), it consistently delivered the lowest prediction errors, proving that residual learning improves tabular forecasting with reduced preprocessing. The GNN + LSTM pipeline achieved the highest R² at 0.1332, slightly above linear regression (0.1320) and far better than persistence (–0.3344), but its MAE was 2.5636, over 15× worse than optimized TabNet. This stemmed from node sparsity and GraphSAGE’s static graph construction. TabNet + GPR achieved the lowest MAE/RMSE; GNN + LSTM had highest R². Results were limited significantly by computational constraints. Next steps include dataset expansion to improve R², quantile regression implementation for uncertainty, and replacement of GPR with LightGBM to lighten residual learning and boost efficiency
Fast Sampling Algorithms for Identifying Pathway Occlusions in Metabolism
No full textIdentifying metabolic pathways can provide valuable insights into underlying biochemical and metabolic stress, help discover novel drug targets, and enable the identification of metabolic biomarkers for early diagnosis and disease monitoring. In this project, we focus on identifying pathway occlusions in metabolism.
The problem is formulated as a mathematical model on a graph: Given a directed graph G = (V, E), where the vertex set V is known but the edge set E is governed by a probabilistic function, how can one efficiently recover the graph’s structure? At any given time, we can measure the connectivity strength of a vertex i ∈ V.
We begin with a special case where the presence of an edge between any pair of vertices is deterministic—each edge is either always present (with probability 1) or always absent (with probability 0). The objective is to reconstruct the graph’s edges based on the collected connectivity strength information.
We show that the solution to the connectivity matrix is not unique. Additionally, we demonstrate that a brute-force approach limits the graph size to four vertices on a classical computer. To overcome the limitations of brute force, we design both a Monte Carlo sampling algorithm and a randomized algorithm to reconstruct the graph’s edge set. These algorithms can handle graphs with hundreds of vertices. The probability of obtaining correct solutions is bounded using the Chernoff bound.
As future work, we aim to develop quantum sampling algorithms capable of scaling to even larger graphs
The implementation of a Gaussian Process Regressor in a geospatial LSTM model creates optimization in uncertainty quantification
No full textFor machine learning in the geospatial field, uncertainty quantification is pivotal towards evaluating model prediction accuracy. Methods of uncertainty quantification track how uncertainty evolves throughout the model training process and strive to distinguish epistemic uncertainty (model limitations). The Gaussian process is a unique method of uncertainty quantification because of its ability to propagate probability distributions over a function. However, the Gaussian process has yet to be widely studied within the geospatial field, with few applications especially in the Tensorflow program. In this study, the Gaussian process is implemented using Tensorflow in a Long Short-Term Memory(LSTM) model: a Gaussian Process Regressor trains with extracted features from the trained LSTM. Using geospatial air quality datasets, the Gaussian process model was evaluated with metrics like negative log-likelihood, expected calibration error, etc. and discovered to perform with greater value prediction accuracy than an LSTM only model, with about 20% lower mean absolute error. However, the Gaussian process has scalability limitations tied to its high computational cost, favoring smaller datasets. When training the model on a smaller dataset of Los Angeles County compared to a larger dataset of California (from OpenAQ and AirNow sensors), the smaller dataset yields better metrics, particularly a positive R squared value (as opposed to the larger dataset’s negative R squared) which indicates better quality data fitting. While the Gaussian process has the potential to optimize uncertainty quantification, its computational intensity poses problematic scalability constraints
Uncertainty Quantification with Probabilistic Deep Learning Models to Predict Out-of-distribution Events
No full textThis project investigates how probabilistic deep learning models—such as Bayesian Neural Networks (BNN) and Monte Carlo Dropout (MCD)—can quantify predictive uncertainty to detect unexpected air quality extremes. The core objective is to illustrate how uncertainty estimates can facilitate the identification of out-of-distribution events in real-time, thereby supporting more reliable calibration and decision-making for PM₂.₅ monitoring during severe wildfire episodes. Using the 2025 Los Angeles wildfire as a case study, the work focuses on calibrating PM₂.₅ measurements from low-cost Clarity sensors and demonstrating how model uncertainty increases when extreme wildfire smoke pushes concentrations beyond the conditions seen during training. Accuracy was measured using traditional values such as R² and RSME. As expected, the accuracy of the model is significantly greater for in-distribution data regardless of whether BNN or MCD is used. Understanding the uncertainty for PM₂.₅ predictions will allow government agencies to be better prepared for out-of-distribution events such as wildfires or industrial accidents, and ensure accuracy in models during rapidly evolving, high-risk environmental scenarios with limited ground truth coverage