Science Project Summaries

Akin to seismology, asteroseismology is the study of stellar pulsations to understand the interiors of stars. Most stars in the sky pulsate, and through the application of asteroseismic analysis, fundamental parameters such as stellar mass, radius, internal rotation, and stellar structure can be determined.

The selected student will work with NASA's Kepler and TESS space data to identify pulsations in a list of known heartbeat stars. Heartbeat stars are double star systems where the stars undergo gravitational distortions at their closest approach, causing their light variations to resemble a cardiogram (hence the name). When analyzing the light variations of these stars, extensive analysis is required to differentiate between pulsations and gravitational distortions. Positive detections require detailed analysis using periodograms and wavelet transforms.

The selected student will work closely with the professor to ensure clear detection for each object. The tasks involve downloading the data, modeling the double star features using pre-written code, applying periodograms, and applying filters and wavelet transforms. While much of the code is pre-written for this task, the student will collaborate with the professor to identify areas where improvements could be made as they analyze the sample.

The proteasome is a macromolecular machine that unfolds and degrades proteins in all eukaryotic cells. In the Kraut lab we study the processivity of the proteasome – that is, its ability to unfold and degrade substrates containing folding domains without falling off the substrate. Substrates to be degraded are polyubiquitinated, meaning a chain of small proteins called ubiquitin are attached to lysines within the substrate. In addition to targeting the substrate to the proteasome, we have previously shown that ubiquitin seems to activate the proteasome for better unfolding of its substrates. In this project, we will compare the degradation of substrates containing an internal ubiquitin-like-domain (UBL) with substrates that contain an internal stretch of ubiquitin domains. Our hypothesis, based on preliminary evidence, is that the UBL domain will activate the proteasome’s unfolding ability to a greater extent than ubiquitin does. Depending on the results of initial experiments, we may then generate new DNA constructs encoding new protein substrates to determine which portions of the proteasome interact with the substrate, or we may use proteasome ubiquitin receptor mutants to examine the mechanism behind any observed differences.

This collaborative and multidisciplinary project utilizes synthetic organic chemistry, physical organic chemistry, and microbiological analysis to develop novel disinfectant compounds to combat pathogenic bacteria, and shed light on why bacteria can resist the traditional disinfectants that have been used for nearly a century. Moving beyond traditional structural classes of disinfectants that have become susceptible to bacterial resistance, we will develop fundamentally different structures that, when paired with careful analyses of their physical properties and solution behavior, will shape the next generation of antimicrobials.

This project aims at characterizing the mentions to named entities by various US-based news producers. A named entity is a reference to a real-world object, such as a person, institution, or location. For example, “New York City” is a named entity that references a location. The goal is to provide a quantitative description of the entities mentioned by the news publishers with respect to the categories of entities that are referenced, as well as the sentiment of the text around such entities. We seek to answer the following questions:

1. What categories of entities are mentioned the most by each news producer?
2. What named entities are mentioned more frequently overall?
3. Which entities are mentioned in a positive context, and which are mentioned in a more negative context?

To answer these questions, we will utilize a large corpus of news publications containing over 7 million articles published by more than 500 news websites. Language models will be used to perform named-entity recognition (NER), and sentiment analysis of the target data. Once the named entities are identified, statistical analyses will be performed to determine the most frequent entities, and the correlations between entity and sentiment (positive or negative).  With these results, we will proceed to ask deeper research questions, such as whether correlations between entity, sentiment and the credibility of the news producers exist.

Now, more than ever, we experience much of the world remotely. This requires us to communicate our experiences to people through digital media. For example, to describe to a remote person what it is like to be on Villanova's campus, we might use words, send them pictures or videos, or even "walk" them around while on a video call. These modes of communication effectively communicate *our* experiences but do not tell someone what it would be like for *them* to explore the campus. It does not give them agency. The simplest way to give someone a taste of this experience would be to let them point the camera and move around campus freely in a virtual environment.

The classic way of creating these virtual environments is to collect a large amount of video data of the location, construct a 3D model, and then put that model in a rendering or game engine. This has already been done for the campus! However, extracting 3D models from video data is tricky and often gives the final product a synthetic video game feel. Furthermore, many elements of the visual world (water, glass, shiny objects, etc) are difficult to extract or even impossible to represent with a 3D surface model. Recent work in Computer Vision has addressed many of these concerns with a new paradigm for representing 3D worlds: novel-view synthesis. In novel view synthesis, one skips the step of extracting a complex 3D surface model and instead creates an implicit 3D model that can be directly queried for high-resolution renderings from any viewpoint. This project aims to apply these new techniques to the existing Villanova campus data.

The project will involve:

1. Accessing and manipulating the image data for the campus.
2. Evaluating and refining existing novel-view synthesis applied to this data.
3. Deploying the final model.

The virtual reality (VR) industry continues to experience significant growth and expansion, with approximately USD 32.64 billion market by the end of 2024. Due to its unique immersive experience and simulation ability, VR technology is increasingly being adopted in fields such as gaming, healthcare and education. Therefore, comprehensive graphic user interface (GUI) testing is crucial to validate user experience and designed functions. Unfortunately, many of the existing testing approaches are white-box testing and require the app’s source code. With over six major VR brands and over five VR engines on the market, it is challenging to design universal testing support for all VR apps.

Recent Generative AI (GenAI), such as GPT-4 and BERT, shows its ability to train, process, and generate data in mixed formats, including text and images. In this project, we want to explore the potential of using GenAI to assist VR exploration testing. In particular, we plan to conduct a case study using GPT-4o and try to answer three questions:

1. How can GenAI help in test entity selection in the user’s view?
2. How can GenAI help in suggesting test actions across multiple views?
3. What are the limitations and potential solutions?

The view is known as the field of view (FOV). The initial question in our study focuses on the fundamental operation of exploration testing: selecting the testing target. Efficient target selection involves accurately identifying objects, organizing them effectively, and correctly localizing them within the scene. Once the test targets are determined, a sequence of actions is applied accurately to complete the exploration task. Given that previous actions can influence subsequent ones, we aim to utilize GenAI's memory capabilities to explore how contextual information across multiple interactions affects the accuracy of the suggested results. The final question addresses our findings on GenAI's limitations and potential improvements for future research.

Voice Assistant (VA) in smartphones has become very popular with millions of users nowadays. A key trend is the rise of custom VA embedding, which enables users to perform the customized tasks of their favorite app through voice control. However, with such a great demand, little effort has been made to support app developers in VA development. Moreover, many user-oriented VA control approaches even increase the programming burden on developers. To reduce the workload and improve code efficiency, we will design and evaluate a novel approach that reuses the test code of an application to support its VA development.

In this project, we assume the VA function has been built and will design a mapping system that links the received user command to the functions. In particular, we will first generate the textual description (short sentences or phrases) of the pre-built functions by adopting the Large Language Model (LLM) powered code-to-text model and then using the NLP-related techniques to find the target functions for given user commands. We will compare the similarity between the sentences using different similarity measuring approaches. Moreover, we will explore the solution to identify or extract the variable values of functions from the user command. For instance, when a user requests to start a ten-minute timer, we need to invoke the setTimer() method and pass a duration variable that represents “10” minutes. After finishing the design of the approaches, we plan to evaluate the system by letting real users speak out the query and observe the task completion status.

Each year, over 20 million metric tons of road salt (e.g., sodium chloride) are applied to U.S. roadways as a de-icing agent. Much of this salt makes its way into rivers and streams where it can have harmful impacts on aquatic life and on downstream municipalities who rely on these waterways as a drinking water resource. Additionally, sodium in road salts deposited on the side of roadways can infiltrate into soils where it displaces base cations (calcium, magnesium, and potassium) and potentially harmful trace metals (copper, lead, and zinc) into shallow groundwater, which are then discharged into streams as baseflow. These increased metal concentrations can also have a negative influence on aquatic life.

For this study, we will characterize the impacts of road salt on headwater streams in the Villanova campus area. We will carry out the collection of streamwater samples and discharge measurements in several suburban watersheds. Water samples will be analyzed for major anions and cations using ion chromatography and select trace metals using inductively coupled plasma mass spectrometry. The analytical data will be subsequently compared to water quality standards for the protection of aquatic life and drinking water quality.

The Weston lab works on questions around the resilience of tidal wetland ecosystems in the face of climate change and land use change. Tidal wetlands provide many critical ecosystem services, but many wetland systems are threatened by sea-level rise, changing inputs from the watershed, and other impacts of human activities. The First Year Match students will work with the Weston lab on research in support of several externally-funded projects that examine carbon and nitrogen cycling in coastal wetland ecosystems and the response of coastal wetlands to sea-level rise and changing land use. The laboratory conducts fieldwork in the Spring, Summer, and Fall in Plum Island Sound, MA and more locally in the Delaware River estuary. Samples collected in the field are brought to the laboratory for processing.

The trade of reptiles and amphibians (i.e. herpetofauna) for exotic pets is an expansive industry that threatens biodiversity. While much of this trade is both legal and sustainable, it can simultaneously contribute to species declines by incentivizing the overexploitation of wild populations, facilitating the spread of invasive species, and acting as a vector for the transmission of zoonotic pathogens. In the United States, herpetofauna are frequently sold at trade shows, where breeders, consumers, hobbyists, and other enthusiasts convene to sell and purchase animals and supplies, often traveling long distances and across state lines. Importantly, these shows are often used by breeders to debut new species and/or morphs; however, the role these expositions play in broader US exotic herpetofauna trade network, in terms of species composition and consumer profile, has yet to be studied.

This project seeks to fill this knowledge gap by establishing a monitoring program for herpetofauna trade shows throughout the state of Pennsylvania. Pennsylvania, relative to other states in the northeastern United States, which has relatively little restrictions on which species can be sold, making it a popular destination for consumers looking for rare, venomous, and/or “risky” species; as a result, there are a relatively large number of long-term trade shows held in cities throughout the state (e.g. Morgantown, Oaks, Hershey, Scranton, Carlisle, Pittsburgh). Information on species composition, selling prices, and the geographic distribution of participants collected from trade shows will be compared to existing large datasets on herpetofauna trade collected by governmental organizations and from online retailers to characterize the contributions of trade shows to the broader exotic herpetofauna trade network.

The project is broadly in the field of extremal combinatorics, which is a field in discrete mathematics dealing with certain optimization-type problems -- i.e. maximizing or minimizing the size of a given discrete structure subject to a given constraint. The constraint is typically provided in the form of a forbidden substructure that may not be contained inside the larger structure being optimized. The project will aim to investigate generalizations and applications of a seminal extremal theorem known as the Erdos--Ko--Rado (EKR) theorem, a foundational result that has found wide-ranging applications in the fields of probability theory, discrete geometry, combinatorial designs and computer science.

The project will particularly focus on a structurally intricate graph-theoretic generalization of the EKR theorem, and its connections to a related, longstanding conjecture of Chvatal, from 1974.

X-ray Binaries (XRBs) are binary star systems in a close orbit containing a regular star and a companion that is a compact object – a black hole or neutron star. These exotic systems are given their name because they are very bright at X-ray frequencies. The basic picture of an XRB consists of a compact object surrounded by hot plasma that orbits and falls onto the compact object in the shape of a disk, called an accretion disk. At the innermost regions around the compact object is a coronal plasma producing X-ray emission at even higher frequencies, sometimes close to the gamma-ray regime. In addition to the accretion disk and corona, many XRBs also contain collimated jets that emit at radio frequencies. Each component of the XRB environment is dynamic: the inner edge of the accretion disk contracts and recedes, the corona can grow or shrink, and the radio-emitting jet may appear and disappear. The evolution of the emission over time offers a particularly powerful window into the structure of the accretion environment and the relationship between different emitting regions in the system. Astrophysicists can use techniques from time series analysis to connect the variations in brightness over time to the dominant physical mechanisms at play.

This project will focus on exploring creative approaches to presenting results from the analysis of X-ray data of a black hole XRB from the Rossi X-ray Timing Explorer. Our current preliminary results include snapshot images of the XRB at different points in time and show that the brightness variability of the source connects to the geometry and changes in the accretion environment. This project will aim to explore techniques such as sonification to make our results accessible and interesting to the diverse science community.

Dust grains in the Milky Way become aligned by interstellar magnetic fields. As a result of this alignment, the resulting far-infrared radiation emitted by these grains is polarized. By measuring the polarization of this light, we can probe the physics of these interstellar magnetic fields and by doing so increase our understanding of the role of magnetic fields in dynamical Galactic processes. This is especially true in the central region of our Milky Way where densities, velocities, and field strengths all exceed median values in the Galactic disk. Our group has completed a survey of the Galactic Center using NASA's Stratospheric Observatory for Infrared Astronomy (SOFIA). This polarimetric survey covers the inner 500 light-years of the Galaxy and provides clues as to the complexity of the magnetic field in this region. In this project, we will compare this data set with other multi-scale tracers of magnetic field structure to better understand the 3-dimensional configuration of the magnetic field. This will involve investigating trends in polarimetric quantities (magnitude and direction) with respect to properties of the central Milky Way such as temperature, density, and radiation environment.

A student working on this project will develop a machine learning algorithm to automate the categorization and analysis of optical images of van der Waals (vdW) material flakes, which are vital for flexible electronics and nanoscale devices due to their diverse properties. Currently, optical images of flakes are manually captured and tagged so that they can be used for various other research projects. The student will photograph these flakes and measure their thickness using an Atomic Force Microscope to create a dataset that correlates thickness with optical properties such as color. The student will then train a machine learning model to predict the thickness of flakes based on their optical characteristics. In addition, image processing techniques will be applied to segment regions of uniform thickness and automatically calculate their area and dimensions. This project aims to streamline what is currently a labor-intensive process used in laboratories around the world.