2025 Summer Student Fellows

In the United States in 2020, over 41 million people aged 12 or older, needed substance use treatment 1. Relapse is a common challenge in substance use recovery, with approximately 40–60% of individuals returning to substance use after a period of sobriety 2. This study explores whether a machine learning model can predict impending drug relapse based on individuals’ Reddit posting behavior in the month prior to a relapse event. We introduce a methodology that leverages large language models (LLMs) to rapidly annotate and organize a massive dataset of Reddit posts for this prediction task. Filtering from over 950,000 posts across multiple addiction-related subreddits, we identified 3,086 instances of self-reported relapse events using an iterative ChatGPT labeling pipeline. This high-volume labeling task, which would have required an estimated 20 years of manual effort, was completed in under 40 computer processor hours, illustrating the efficiency and feasibility of LLMs as automated data labelers. Engineered features reflecting common relapse triggers were also derived using LLMs and used to train various classifiers, including logistic regression, random forest, gradient boosting, SVM, and k-NN, to distinguish between collections of posts made four weeks versus one week before a users’ relapse. Our results show that classical machine learning (ML) model performance remained near chance levels, suggesting that clear predictive signals were either absent or too nuanced for detection via these methods. In addition, we tested the LLM’s ability to make direct predictions based on individual authors’ posts, bypassing the classic ML models completely. This approach yielded more promising results and a level of potentially useful predictive accuracy. Our work highlights both the limitations of relapse prediction using Reddit content while also demonstrating the significant potential of LLMs for scalable behavioral data labeling, feature extraction, and as predictive classifiers.

The Fragile State Index is an annual ranking developed by the Fund for Peace of countries based on their risk of instability, conflict, and failure considering political, social, and economic pressures made up of twelve indicators. The purpose of this project is to examine how selected indicators in the FSI change over time and how they relate to each other, specifically to identify dynamic patterns—autocorrelations within indicators and correlations between different indicators—both within countries over time and across countries. The analysis also compared results based on indicator levels (scores) with results based on year-to-year changes (first differences) to determine whether they show the same patterns.

Annual FSI data was prepared in Python by keeping only countries with data throughout the whole study period and removing the first year (2006) entirely due to it having significantly countries. This data was combined into a single dataframe and then separated into country-specific dataframes. For each country and indicator, autocorrelations at multiple lags were estimated and lag-one cross-correlations between indicators were calculated. To aggregate results across countries while accounting for unequal sample sizes and distributional differences, country-level correlations were Fisher z-transformed and averaged; correlations that were not statistically significant were excluded. The country-specific dataframes were differentiated, and the same process was repeated to find correlations for the difference series.

The results show strong within-indicator autocorrelation that decays with increasing lag. Lag-one cross-indicator analysis revealed two apparent subsystems: The first including E2 (economic inequality), E3 (human flight/brain drain), P2 (public services), S1 (demographic pressures), and X1 (external intervention); the second including P1 (state legitimacy) and P3 (human rights). Analysis of the difference series produced a single notable negative correlation between C2 (fractionalized elites) and its own year-to-year difference. This contrast implies that many relationships observed reflect persistent structural conditions rather than synchronized, rapid changes in related indicators.

The findings point to stable, slow-moving dynamics in the FSI indicators, with limited evidence of large movement in annual changes within subsystems. Future work focused on the dynamics of FSI indicators could look in to these or similar subsystem’s long-term trends to determine qualitative features of the FSI.

Routine cognitive screening is critical in Parkinson’s disease (PD) to monitor progression and differentiate between normal cognition, mild cognitive impairment (PD-MCI). Mild cognitive impairment is a common feature of Parkinson’s disease (PD–MCI), even at the earliest disease stages, but there is variation in the nature and severity of cognitive involvement and in the risk of conversion to Parkinson’s disease dementia (PD-D). Therefore, they need to be distinguished from patients with normal cognition and those with dementia.

Our research develops advanced statistical and computational tools to enhance biomarker analysis for diagnostic classification. Central to this effort is a multi-class extension of the Youden Index—a key metric for evaluating diagnostic accuracy and determining optimal biomarker thresholds.

We introduce the Generalized Variable Method (GVM), a novel framework for multi-class classification that integrates Bayesian inference, bootstrap resampling, parametric and non-parametric techniques, and the Pivotal method, all implemented via parallel programming in Python. Using simulations on the Montreal Cognitive Assessment dataset, our approach extends traditional two-class diagnostics to complex, real-world medical applications. Results demonstrate that GVM yields accurate and reliable estimates of the Youden Index, coverage probability, and test size, offering a robust tool for cognitive assessment in PD.

Mutant epidermal growth factor receptor kinase (EGFR) is an oncogenic driver for varied cancers including lung, colon, head and neck cancer and leukemia. The epidermal growth factor receptor tyrosine kinase inhibitors (EGFR-TKIs) are effective against EGFR harboring exon19 deletion (del19) or leucine to arginine (L858R) mutations. The EGFR-TKIs block the RAS/RAF/ERK signaling cascade that drives uncontrolled cancer cell proliferation. The mitogen-activated protein kinase interacting kinases (MNK) have been linked to cancer progression, particularly for acute myeloid leukemia (AML), and are activated downstream of the RAS/RAF/ERK signaling pathways. Simultaneous inhibition of both EGFR and MNK could be anticipated to demonstrate synergistic anticancer effects in varied cancers.  Kurup et al. identified compound 1 as a nanomolar EGFR/MNK inhibitor. Two series of compounds have been designed and synthesized as analogs of 1 and dual EGFR/MNK inhibitors. The design, synthesis and evaluation of dual EGFR/MNK inhibitors will be presented and discussed.

This research project outlines the development of a culturally immersive, trauma-informed study away course to the Pine Ridge Reservation, home to the Oglala Lakota Nation. The initiative integrates academic inquiry, service learning, and reflective practice to foster cultural humility and competence among participating students. Grounded in literature on cultural competence, cultural humility, and trauma-informed care, the course emphasizes understanding the historical and ongoing impacts of U.S. government policies on Native American communities. It incorporates evidence-based practices aimed at facilitating transformative learning experiences, encouraging students to critically reflect on their roles and responsibilities when engaging with Indigenous populations.

A central feature of the course is an academic service-learning partnership with Re-Member, a non-profit organization collaborating with the Oglala Lakota Nation. Through this partnership, students participate in community-driven projects such as building beds, constructing outhouses, and installing mobile home skirtings, directly addressing pressing housing and infrastructure needs. These service activities are complemented by cultural immersion experiences, including presentations by Native speakers, site visits to historically significant locations such as the Wounded Knee Massacre site, and participation in community events.

As the research student, I contributed to course development by conducting comprehensive literature reviews on Native American history, culture, historical trauma, and transformative pedagogical practices. I examined peer-reviewed research on cultural competence and humility, identifying strategies that move beyond superficial cultural awareness toward sustained, ethical engagement. I also explored trauma-informed care frameworks relevant to Indigenous communities, emphasizing the importance of safety, trust, collaboration, and empowerment. Additionally, I joined the faculty member on a preliminary trip to Pine Ridge, where I engaged in service work, attended cultural presentations, and visited historically significant sites to assess the feasibility and educational value of each activity.

From this work, I found that cultural humility is best developed through sustained, community-led engagement rather than short-term exposure. I observed that trauma-informed approaches are critical for avoiding re-traumatization and fostering respectful relationships. The experience reinforced that reciprocal partnerships—where community priorities guide student projects—are essential for ethical service learning. This study away program, as designed, aims to equip students with the skills, awareness, and perspectives necessary for ethical and effective engagement with Indigenous communities, promoting mutual respect, understanding, and social responsibility.

During my project I did a deep dive into the characterization and preparation of nanoparticles. A nanoparticle is an extremely small particles 1-100 nanometers in size. I used the drug Erlotinib to be entrapped on the inside of my nanoparticles I was preparing. Erlotinib is a kinase inhibitor for cancer treatment. So, using nanoparticles we can directly target certain cancers and in a way ‘inject’ the anti-cancer drugs into the cancer. During the first few weeks of the Student Research Fellowship, I did a plethora of literature review on different methods of preparing nanoparticles and about the drug, Erlotinib. During this review I also learned how to use different machines including the sonicator and particle size analyzer. During the process of making the nanoparticles I used trial and error with the sonicator and different times figuring out what works the best for my method. After I completed my method, I used different amounts of each chemical I was working with to create a data table. Using this data table and the particle size analyzer I found out what combination was giving me the best results. During these weeks of developing a method of the nanoparticles that worked best for me derived from the literature review I also learned about the particle size analyzer and how to put it together and how to run the software while collecting data. After a few more weeks I eventually made my way to the Shimazu Core Lab located in the Arts and Science Commons building. I worked alongside Dr. Pike and a fellow student, Xhesika (Jessica) Luarasi, on the Liquid Chromatography Mass Spectroscopy (LCMS). I am still in the process of finishing the project because science is not always a linear process and there are a lot of learning curves and waiting that go into research. Dr. Pike and Xhesika have been amazing help in navigating me through the LCMS and helping me develop a method to use to characterize how much Erlotinib is in my nanoparticles. I have developed a method to use on the LCMS but am currently waiting for a new column to get delivered so I can achieve better results. I have been learning a lot through this experience about how research and science are not linear, and it is a waiting game. I have been learning a lot about the different machinery and how to use and fix it and learning how to conduct in-depth research.

The objective of this research is to explore effective methodologies for addressing the housing affordability crisis in small towns around the United States, primarily focusing on the City of Big Rapids in Michigan. Although Big Rapids isn’t facing a housing crisis, this can easily change. This study aims to identify construction methods and housing models that could provide sustainable, yet affordable solutions to meet the needs of various members of the Big Rapids community. 

An analysis of the challenges that impact housing affordability was conducted, which included research comparing rental prices to income levels, and a look at the demographics in Big Rapids. Possible housing methods were discovered and researched, like tiny houses, community living, container homes, duplexes, and modular homes. Research was done for each method to determine what would be the best approach to tackling affordable housing in Big Rapids.

Results of the research done revealed that the largest population impacted by affordable housing in Big Rapids are students. Using this information, it was determined that modular construction of apartments and dormitories offers the most affordable housing solution. Modular housing reduces overall construction costs and shortens building timelines by assembling prefabricated units off-site. This method is best for students to experience a dorm-style apartment that would be more cost effective than traditional apartment rentals. Modular living can also address the needs of the rest of the community as it can be flexible to be made in any size, while also enhancing the tight-knit community aspect that is seen in small towns across the country. 

This research concludes that modular construction is the best option for Big Rapids to address the struggles of finding affordable housing, while providing a cost-effective solution for the city. Modular construction can be made to be LEED certified by using smart energy efficient methods and sustainable approaches. While prefabricated modular construction isn’t commonplace in the United States, the market for this style of homes is growing. It’s time to get involved in the community and offer a more affordable, efficient and sustainable way of living.

This research project investigates the economic and security implications of government-mandated encryption backdoors with the objective of determining whether these policies enhance or undermine digital security and market trust. The central question asks if the perceived benefits to law enforcement and national security outweigh the long-term risks to innovation, competitiveness, and consumer confidence.

To address this question, the project employed a multidisciplinary approach grounded in economics, technology policy, and case study analysis. The research began with a review of economic theory, which frames encryption as a positive externality that fosters consumer trust in e-commerce, banking, and digital communication. Introducing backdoors, however, creates negative externalities by embedding vulnerabilities that can be exploited by malicious actors. To test these claims, several case studies were analyzed, including the Juniper Networks backdoor discovered in 2015, the NSA’s Dual EC algorithm with its suspected flaw, and the Apple versus FBI case in 2016. Each example highlighted how efforts to weaken encryption for government access either failed to achieve intended goals or directly resulted in economic losses, weakened security, or public backlash.

The findings indicate that encryption backdoors create systemic risk by acting as single points of failure for entire networks. If compromised, they open pathways for hostile actors to access sensitive systems, raising the probability of widespread breaches. Economically, such mandates threaten global competitiveness. Following revelations of U.S. surveillance programs, cloud computing firms faced projected losses of up to $35 billion due to diminished international trust, and surveys show that many businesses would cut investment or hiring if forced to comply with backdoor regulations. These results demonstrate that short-term investigative benefits are overshadowed by long-term harm to digital ecosystems.

This research concludes that government-mandated encryption backdoors are economically detrimental. They erode trust, weaken cybersecurity, distort markets, and suppress innovation. Importantly, viable alternatives such as lawful hacking, metadata analysis, and warrant-based surveillance allow law enforcement to pursue investigations without undermining security.

Ultimately, the findings suggest that preserving strong encryption is essential for protecting privacy and civil liberties while sustaining economic growth and stability. This evidence-based analysis provides policymakers with recommendations that prioritize security, competitiveness, and public trust.

This capstone presents a systematic review of research on early childhood trauma and its profound effects on infant mental health (IMH). Focusing on children aged 0–5, the review explores how trauma—rooted in factors such as abuse, neglect, systemic inequities, and caregiver mental health—can disrupt emotional regulation, attachment, and neurological development. Utilizing trauma-informed and equity-centered frameworks, the paper synthesizes findings on risk and protective factors, emphasizing the importance of early intervention, culturally grounded care, and relational support. It highlights validated tools such as the ACEs framework and DECA, while critically examining their implementation challenges in under-resourced systems. The review identifies promising psychosocial interventions—including dyadic therapy and culturally tailored programs—that strengthen caregiver-child relationships and promote resilience. Additionally, it underscores the need for policy advocacy to shift systems toward prevention and accessibility. Ultimately, this paper calls for a coordinated, socially just approach to healing that empowers caregivers, builds community trust, and fosters healthy development from the very beginning of life.