2023 Research Prize Winners

1st. Prize: Christopher Callahan (Geography) Advisor: Mankin

The economic impacts of climate change are potentially severe but poorly understood. In my dissertation project, I combined climate simulations and large economic datasets to understand the costs of global warming. I showed that heat waves and El Niño, two major climate hazards, reduce economic growth in tropical countries. I then showed that the emissions of countries and fossil fuel firms in the global North have, as a result, cost these low-income, low-emitting regions billions of dollars. These findings demonstrate how computational science can support societal responses to the climate crisis and highlight the economic threat of global warming for vulnerable people.

1st Prize: Ali Massa (Molecular & Systems Biology) Advisor: Leach

Up to 8% of our DNA consists of ancient retroviral sequences known as endogenous retroviruses (ERVs). Accumulating evidence suggests that expression of ERVs in cancer – creating a state of 'viral mimicry' – may enhance anticancer immunity. However, computational mapping of ERVs has historically been challenging, limiting our understanding of their clinical and immunological impact. We utilized a recently developed bioinformatic pipeline Telescope (Bendall et al. 2019) to quantify the expression of 14,968 ERVs at specific genomic locations across 8 different cancer types using patient sequencing data. We identified both tissue-specific and commonly expressed ERVs across cancers, with a subset being significantly prognostic. Our findings suggest that expression of favorably prognostic ERVs correlates more strongly with an innate immune response, although the strength of this association is tissue specific. Using an in-silico RNA structure and stability pipeline ScanFold (Andrews et al. 2018), we found that prognostic ERVs were in fact more likely to form double-stranded RNA, providing further support for viral mimicry. This work adds to a new and growing framework for ERVs in cancer immunity.  

2nd Prize: Paxton Fitzpatrick (Psychological and Brain Science) Advisor: Manning

We developed a computational framework for tracking and characterizing conceptual knowledge and learning in real-world classroom-like settings. We used text embedding models to extract the conceptual content presented in each moment of an online course lecture. We then applied these models to simple multiple-choice quizzes to identify the specific moments in which the knowledge needed to correctly answer each question was conveyed. This allowed us to estimate how well students understood each individual moment of a lecture, predict their ability to correctly answer specific quiz questions, and recommend specific concepts to review in order to fill in their knowledge gaps and more successfully learn from future lectures.

2nd Prize:  Luyang Zhao (Computer Science) Advisor: Balkom

In nature, many creatures are capable of working collectively by joining together to form large-scale assemblies to achieve a unified goal. In robotic construction, robotic manipulators often build structures from rigid load-bearing blocks. On the other hand, soft robots can adapt their shapes to suit tasks. My research focuses on designing intelligent soft modular lattice-based blocks that bridge the gap between rigid blocks and soft robotics. These blocks exhibit versatile capabilities, transitioning between deformable states for locomotion and manipulation, while also having the ability to rigidly lock into specific shapes to support heavy loads. Their autonomous self-assembly into complex structures without external assistance unlocks opportunities for adaptive and self-reconfigurable robotic systems. To our knowledge, this is the first example of active deformable blocks that can reconfigure into different load-bearing structures on-demand [1].  

[1]Luyang Zhoa, Y. Wu, Y. Wen, W. Zhan, X. Huang, J. Booth, A. Mehta, K. Bekris, R. Kramer-Bottiglio, and D. Balkcom. StarBlocks: Soft Actuated Self-connecting Blocks for Building Deformable Lattice Structures. IEEE Robotics and Automation Letters, May. 2023 (accepted) Video: https://youtu.be/xno0FBs3ZdQ

3rd Prize: Subiga Nepal (Computer Science)  Advisor: Campbell

The university years are transformative, posing unique opportunities and challenges. Understanding students' well-being during this critical period is vital, especially given the unprecedented impact of the COVID-19 pandemic. In our study, we extensively analyze the experiences of undergraduate cohorts over four years, employing a novel combination of mobile sensing data, surveys, and qualitative interviews. We investigate patterns in students' daily activities, mental health, and resilience, from pre-pandemic years through the pandemic and once the pandemic subsides. This comprehensive longitudinal approach provides invaluable insights into the resilience and adaptability of students under extraordinary circumstances, which could guide effective strategies for supporting student mental well-being. This research represents one of the most extensive mobile sensing studies to date, promising significant implications for developing effective interventions and support systems in higher education.

3rd Prize: Natasha Kelkar (Microbiology and Immunology) Advisor: Ackerman

Insights into mechanisms of protection afforded by vaccine efficacy field trials can be complicated by both low rates of exposure and protection. However, these barriers do not preclude the discovery of correlates of reduced risk (CoR) of infection. Given the significant investment in large scale human vaccine efficacy trials, novel approaches for analyzing efficacy trials to optimally support discovery of correlates of protection (CoP) are critically needed. We investigated the value of applying Positive/Unlabeled (P/U) learning to classify study subjects using model immunogenicity data based on predicted protection status in order to support new insights into mechanisms of vaccine-mediated protection from infection. Our work demonstrated that P/U learning methods can reliably infer protection status, supporting discovery of simulated CoP that are not observed in conventional comparisons of infection status cases and controls.

Kelkar, N.S., Morrison, K.S. and Ackerman, M.E., 2023. Foundations for improved vaccine correlate of risk analysis using positive-unlabeled learning. Human Vaccines & Immunotherapeutics, p.2204020.

Honorable Mention: Martin Ying (Physics and Astronomy) Advisor: Chaboyer

The absolute age of the oldest stars in our galaxy is of fundamental interest for a wide range of applications but is difficult to measure in practice. Messier 92 is one of the oldest and most metal-poor globular clusters, and its absolute age provides a lower limit to the age of the universe which is independent of cosmological models. We conduct Monte Carlo simulation to generate 20, 000 sets of isochrones using Dartmouth Stellar Evolution Programs, and fit to the Messier 92 data from HST ACS globular cluster treasury program using an innovative 2D isochrone fitting method. We found the absolute age of M92 to be 13.80 ± 0.75 billion years.

Honorable Mention: Mijin Kwon (Psychological and Brain Science) Advisor: Wager

Predictive models of neuroimaging data ("brain signatures") have provided new insights into the neural basis of affective processes. However, a systematic evaluation of these models' performances is essential to establish their boundary conditions and potential use cases. This study utilizes a comprehensive, multi-study fMRI dataset from a collaborative data-sharing consortium to assess the predictive performance of 21 existing task-based fMRI data across various affective domains such as pain, appetitive states, aversive states, and cognitive control. The results revealed variability in the sensitivity and specificity of these signatures, suggesting specific limitations to be addressed. The project is now focused on developing a new generation of brain signatures with enhanced specificity, sensitivity, and generalizability.

1st Prize: Lewis MacMillan (Chemistry) Advisor: Read

Organic charge-transfer complexes are a chemical phenomenon that have been observed for several decades, however there are now modern computational tools that can help provide understanding and extensions for the phenomena. This project focused on developing a mixed computational/experimental methodology to generate a data set from which modern data science techniques can produce predictive models and identify chemical characteristics critical to promoting a better understanding of charge-transfer complexes.

1st Prize: Luc Cote (Computer Science) Advisor: Chakrabarty

The Monotone Submodular Maximization problem is a fundamental discrete optimization problem of maximizing a monotone submodular function subject to a cardinality constraint. This problem and its special case, the maximum coverage problem, have a wide range of applications in various fields including operations research, machine learning, and economics. In this project we design and analyze new primal-dual algorithms for these problems and show that they achieve an optimal approximation factor of (1-1/e). While other algorithms have been previously known to achieve this approximation factor, our algorithms also provide a dual certificate which upper bounds the optimum value of an instance. This certificate may be used in practice to provide much stronger guarantees than the worst-case (1-1/e) approximation factor.

2nd Prize: Christopher Picard (Geography) Advisor: Winter

The northeastern USA has experienced a dramatic increase in total and extreme precipitation over the past 30 years, yet how precipitation will evolve across the Northeast by the end of the twenty-first century remains uncertain. We examined the future of precipitation across the Northeast using a regional climate model to simulate precipitation for historical (1976–2005) and future (2070–2099) periods. We found increases in both total (10%) and extreme (52%) precipitation by the end of the twenty-first century, with winter and spring contributing most to the projected increase in extreme precipitation.

Picard, C.J., Winter, J.M., Cockburn, C. et al. Twenty-first century increases in total and extreme precipitation across the Northeastern USA. Climatic Change 176, 72 (2023). https://doi.org/10.1007/s10584-023-03545-w

3rd Prize: Matthew Timfeev (Engineering) Advisor: Scheideler

Perovskite solar cells (PSCs) are an emerging alternative to Silicon solar cells, but they can also be effective candidates for wireless power transmission (WPT) due to their tunable bandgaps, which can be matched to the wavelength of laser light, minimizing energy losses. We simulate custom PSC devices' performance under varying illumination in Sentaurus TCAD, using a finite-element model for ray-tracing of light and diffusion of charge carriers in the devices. We find an optimal efficiency when the wavelength is matched to the bandgap, which leads us to experimentally achieve a 3x efficiency improvement over Silicon cells for WPT up to 100 meters.

3rd Prize: Eric Youth (Earth Science) Advisor: Osterberg

Alaska's wildfires cause environmental devastation, economic losses, and compromised air quality. Predicting future wildfire activity is crucial, but understanding its response to climate change is limited. Using an ice core, we developed a 1000-year proxy record of Alaskan wildfires. We found higher fire activity in the Medieval era and project historically high levels by 2065 due to increased temperatures fueling destructive fires. Climate analysis suggests a transition to a hot/wet climate, resulting in a 160% increase in burned areas between 2021-2100 compared to 1950-2020. Warming temperatures will overpower increased precipitation, intensifying fire seasons despite current stability.

Honorable Mention: Brian Wang (Computer Science) Advisor: Vosoughi

With the release of ChatGPT, large language models are more relevant than ever. Because these language models are trained on data from humans, the models are prone to encode human biases and propagate stereotypes. We present a new framework for interpretability to understand how various language models are encoding stereotypes, identifying the specific internal structures in a model that are capturing these biases. We use this analysis to explore various methods to make the outputs of generative language models less biased.

Honorable Mention: You-Chi Liu (Biomedical Data Science) Advisor: Moen

This study aimed to investigate potential disparities in the receipt of multidisciplinary cancer consultation (MDCc) among early-stage non-small cell lung cancer (NSCLC) patients and use social network analysis to examine whether patient-sharing networks impact access to cancer consultations and treatment. The patient-sharing network measure of interest was the physician linchpin score, used to identify physicians whose peers lack ties to other physicians of the same specialty as the focal physician, which may arise in areas with limited oncologist supply. we assembled a patient-sharing network based on all NSCLC care. We found that patient demographic and healthcare organizational barriers contribute to disparities in the receipt of cancer care. We observed an increasing likelihood of a patient receiving MDCc as socioeconomic status increases. Furthermore, as age increases, there is a higher relative risk of exclusively consulting a radiation oncologist instead of receiving MDCc. Additionally, as the cancer stage and proportion of linchpin surgeons in a hospital referral region increase, there is a higher relative risk of exclusively consulting a radiation oncologist, surgeon, or neither versus consulting both physicians. We also found that older patients, those with lower cancer stages, more comorbidities, receiving MDCc, and residing in areas with a high proportion of linchpin surgeons are more likely to receive SBRT instead of surgery. This study expands beyond patient-level factors associated with care utilization and investigates how network structures and provider scarcity impact the facilitation or barriers to MDCc, consequently explaining disparities in care utilization. Enhancing care coordination and streamlining the integration of MDCc into the clinical workflow for patients who would benefit the most can improve the care quality.