Research Projects
Open-source battery simulation
Mentor: Dr Hui-Chia Yu
Optimizing electrode microstructures can push batteries' performance to their limits. The Yu group has developed an image-based simulation toolkit, which solves ion diffusion, electrical current continuity, and Arrhenius reactions to elucidate the electrochemical processes occurring in real microstructures. An outstanding need in this project is a driver code that alters microstructures and runs our current toolkit to autonomously search for the optimal electrode structures for fast charging. The REU student will participate in the development of this driver code and simulation-based optimization design of batteries.
Biomolecular aggregation and phase separation
Mentor: Dr Michael Feig
Biological macromolecules such as proteins and nucleic acids experience highly crowded environments inside cells. Using computer simulations one can analyze the molecular interactions under such conditions. Of particular interest is the propensity of molecules to aggregate and form phase-separated condensates. In this project the student researcher will use coarse-grained modeling tools such as COCOMO together with atomistic simulations to explore the determinants of aggregation or condensation and develop hypotheses that can be tested by experimental collaborators.
Automated image annotation toolkit
Mentor: Dr Dirk Colbry
Advancements in imaging technology has led to the widespread use of image data in research across many scientific disciplines such as self-driving vehicles, medical research, and agriculture. However, manual processing and retrieval of specific information from scientific image data can be time-consuming and burdensome for researchers. In this project the student researcher will join the SEE-Insight team to research and develop a general purpose ``researcher in the loop'' toolkit that uses large scale computing to search the algorithm space for automated solutions to domain specific image annotation problems. These tools will speed up scientific discovery across disciplines and reduce the ``mean time to science''.
Precision surveillance scheduling for low-grade glioma patients
Mentor: Dr Adam Alessio
Isocitrate dehydrogenase (IDH)-mutant gliomas are a classification of brain tumors that are understudied, incurable, and often impact young adult patients; yet clinical guidelines for monitoring use a one-size-fits-all surveillance schedule that is not optimized toward an individual's disease nor established treatment paradigms. For example, patients with these gliomas receive MRI exams every 3-6 months which is then typically extended to longer intervals, and there are no established parameters to refine this approach which can be costly in the setting of delayed recurrence. This work seeks to determine if clinical information, laboratory biomarkers, and MRI images can be used to predict tumor progression in order to customize the surveillance schedule for each patient. In this project, the student researcher will use previously proposed deep-learning methods (based on 3D Convolutional Neural Networks) to develop a method to automatically segment gliomas and normal reference regions in the brain. These segmentations will be used to extract radiomic features in the 3D glioma region and in the periphery (margin) of the gliomas. This project benefits from exposure to an interdisciplinary medical research team and the ability to contribute to an ongoing pressing challenge in oncologic care.
Transport phenomena in warm dense matter mixtures
Mentor: Dr Luciano Silvestri
Warm Dense Matter (WDM) occupies a transitional state between solid condensed matter and classical plasmas. It is characterized by conditions in which electrons are partially degenerate and ions can be strongly correlated. This means that, in WDM, the electrons are hot enough that they begin to fill their quantum states above the Fermi energy, while the ions, though highly energetic, can still experience strong interactions with one another due to their close proximity. WDM conditions are found in various astrophysical scenarios, such as the interiors of giant gas planets like Jupiter and in the late stages of stars like neutron stars and white dwarfs. On Earth, these conditions can be created in laboratory experiments, often involving intense laser or pulsed power facilities. Given the challenging experimental conditions for producing and diagnosing WDM, researchers must rely on computer simulations. In particular, Molecular dynamics (MD) simulations are a crucial tool for studying Warm Dense Matter because they offer a detailed atomic-level understanding. In this project students will perform molecular dynamics (MD) simulations to study transport phenomena and dynamical properties of WDM mixtures.
Vocal communication systems in animals
Mentor: Dr Grace Smith-Vidaurre
Mammalian and avian species can learn their vocalizations in ways that are similar to humans. Learning vocalizations may be critical for animals to change the information that they communicate in different social contexts. However, addressing whether and how animals change learned vocalizations poses great challenges. Vocal learning species can be highly mobile, social animals that are difficult to mark, track, and record over time and across different social contexts. Acoustic recording datasets obtained from wild populations are often small and incomplete, with little information about individual identity, social group membership, sex, kinship, and behavioral context. We combine simulation and machine learning approaches to test how small, messy datasets can be used to make biological inferences about vocal communication systems that rely on social learning. REU students will simulate acoustic datasets, quantify patterns of acoustic variation, and test how biological inferences about vocal communication systems are robust to different sampling regimes and forms of missing data.
Directional Transform of 3D Shapes
Mentor: Dr Liz Munch
In order to measure shape, the field of Topological Data Analysis (TDA) encodes structure using tools from algebraic topology. The directional transform is a way to encode a particular shape in 3D by studying a function defined with levelsets normal to a chosen direction. Then we can use tools from TDA, such as the Euler characteristic or persistence diagram, to encode information about this particular function. In this project, students will study the directional transform of example data sets such as those arising from X-Ray CT scans of plants and realizations of attractors of dynamical systems.
Mapping emerging trends and interdisciplinary links in astrophysics research with AI
Mentor: Dr Wolfgang Kerzendorf
The explosion of academic publications, thanks to the internet, has created a treasure trove of research materials. Yet, this abundance also poses a challenge: staying updated on the most impactful publications and identifying emerging research trends is increasingly difficult. In astronomy alone, around 17,000 new papers are published annually, a number that doubles approximately every 14 years. The NASA Astrophysics Data System (ADS) serves as a comprehensive archive, storing over 16 million publications and their associated metadata (e.g., author names, keywords). For this project, the selected student will delve into the ADS dataset to analyze emerging trends and areas of focus in astrophysics research, using data science methods such as text mining, machine learning, and data visualization. By the end of this project, the student will aim to answer the following questions: What are the most frequently occurring keywords and phrases in astrophysics publications over the past decade? How has this changed over time? Can machine learning models predict which emerging topics are likely to be focal points in future research? How interconnected are various subfields within astrophysics, as indicated by co-authorship and keyword overlap?
Coarse-grained molecular dynamics simulation of an enzyme complex
Mentor: Dr Scott Calabrese Barton
The enzymatic complex hexokinase (HK) -- glucose-6-phosphate dehydrogenase (G6PDH) can accomplish a multistep reaction with strong retention (channeling) of reaction intermediates by electrostatic attraction. However, strong electrostatic charge can influence the structure of the complex over long length and time scales, which in turn can effect the channeling efficiency. In this project, the student researcher will use molecular dynamics combined with coarse graining of the model complex in order to predict the dynamics of complex structure, as well as structural effects on channeling efficiency. Coarse-grained models simplify the complex by grouping several neighboring atoms into a coarse-grained bead and enables the simulation timescales up to microseconds. These long simulation scales allow the observation of enzymatic configuration change, providing insights that will enable the design of more efficient catalytic complexes.
PHYSICS-INFORMED MODELING OF COMPLEX DYNAMIC SYSTEMS FROM DATA
Mentor: Dr Huan Lei
The far-reaching impact of the analytical frameworks of modern physics on both our understanding of complex physical systems and the rapid development of advanced technologies in sectors like aerospace, materials science, and bioengineering cannot be overstated. However, many challenges remain. For example, we struggle to accurately model the nanoscale heat transfer relevant to CPU design or the dynamics of protein folding. For such complex systems, there are no analytical models available to accurately characterize their evolution. This project is designed to introduce recent developments in machine-learning to the development of equations governing complex, multiphysics systems while ensuring that these data-driven models strictly preserve all the physical symmetries and constraints of the system.