Flash Talk Abstracts
Talk 1: "We just gave them part of their humanity back”: Epistemic and Distributive Justice in NYC Reentry Policy During COVID-19
Presenter: Erinn Bacchus
Mentor: Tracy Nichols, Joanna Mishtal, & Eduardo Gomez
Department/College: Community and Global Health and Population Health
Navigating reentry after incarceration is a challenging process that was further complicated by the COVID-19 pandemic. The normative basis for policies governing reentry assume that society has a moral obligation to help released individuals heal, restore human dignity, and reintegrate. Many individuals returning home struggled to find both material resources and social support. Reentry staff are often the first line of help for those reintegrating. When the pandemic forced many institutions to close their doors, reentry organizations instead adapted their services through hybrid models, creating diverse methods of interacting with clients. This study draws on 10 in-depth interviews, conducted between September and December 2023, with NYC reentry staff members, and 12 interviews, conducted between August and December 2020, with individuals formerly incarcerated in NYC jails during the pandemic. A codebook analysis approach informed by reflexive thematic analysis was used in data analyses with codes developed through a mix of inductive and deductive approaches. Findings suggest that hybrid programming and the reciprocal, supportive relationship between clients and staff were a vital part of successful reintegration at this time. Staff with a history of justice-involvement further facilitated the development of trust in these relationships. Through an epistemic and distributive justice perspective, these findings highlight the importance of (1) adaptable reentry servicing and (2) the relationships formed between reentry staff and their clients in achieving these justice-based goals. Moving forward, state and local departments of corrections must invest in the development of flexible programming that fosters staff-client relationships and prioritize hiring justice-impacted individuals.
Talk 2: Artificial Intelligence Data Centers in the United States: Beyond Electricity Demand
Presenter: Johanna Bolanos
Mentor: Alberto J. Lamadrid
Department/College: Institute for Cyber Physical Infrastructure and Energy (I-CPIE)
Artificial intelligence (AI) is driving an expansion of data centers, raising concerns about their environmental and economic impacts. This study examines how AI data centers influence electricity systems, air emissions, water resources, land use, and local communities in the United States. The results show that these impacts are determined not only by facility characteristics, but also by the electricity, water, and land-use systems in which data centers operate. In particular, emissions are largely driven by electricity consumption and depend on marginal generation, transmission constraints, and the spatial and temporal characteristics of electricity demand. The analysis also highlights regional differences in water use, noise exposure, and land transformation. The findings emphasize the importance of integrated planning and policy approaches that consider both facility-level decisions and broader infrastructure conditions to improve the environmental and economic performance of AI data centers.
Talk 3: Listening to bridges through everyday traffic: mobile sensing and AI for safer infrastructure
Presenter: Kevin Theunissen
Mentor: Shamim Pakzad
Department/College: Civil and Environmental Engineering
A bridge cannot speak, but every vehicle crossing makes it vibrate. Like a doctor listening to a heartbeat, engineers can ask: How does the bridge move? Where does it feel stress? Which parts may become more vulnerable over time? This work listens to bridges through everyday traffic by using mobile sensing, where sensors inside moving vehicles record vibrations as they cross the structure.
The study first turns these vehicle passages into a clearer picture of bridge behavior. Because each vehicle crosses at a different time, speed, and position, the measurements are asynchronous and difficult to compare directly. The proposed approach processes these non-synchronized passages, places them on a common spatial reference, identifies important vibration frequencies, and estimates signed mode shapes. These mode shapes show how different parts of the bridge move together or in opposition, helping engineers understand the structure’s dynamic behavior without relying only on fixed sensors.
The work then moves toward artificial intelligence as a diagnostic assistant. It aims to learn how acceleration signals measured by moving vehicles can be transformed into strain responses at fixed bridge locations. Strain reveals how repeated traffic loads affect the structure and is essential for fatigue assessment. By reconstructing strain histories across the bridge, AI-enhanced mobile sensing supports the identification of critical zones, fatigue analysis, and long-term life-cycle management. In this way, everyday traffic becomes a scalable tool for listening to bridges and maintaining safer infrastructure.
Talk 4: Recurrent Atlantic tropical cyclone track patterns linked to seasonal climate states
Presenter: Richard Sullivan
Mentor: Ben Felzer
Department/College: Center for Catastrophe Modelling and Resilience
Seasonal Atlantic tropical cyclone (TC) outlooks often emphasize basin-wide activity, but regional hazard depends on how storm tracks, and thus landfall probabilities, are distributed across the basin. Here, we apply machine learning to large sets (>80,000) of synthetic TCs generated from historically reconstructed ocean-atmosphere conditions and controlled climate-model simulations of pre-industrial conditions to identify recurrent patterns in Atlantic TC tracks. These patterns reflect large-scale shifts in seasonal Atlantic storm activity, highlighting how landfall risk is redistributed among coastal zones from year to year. We link these track patterns to distinct combinations of tropical Atlantic and Pacific sea surface temperatures, Atlantic sea level pressure, and vertical wind shear. The recurrence of comparable track patterns and climate states in both datasets indicates that interannual climate variability can reorganize regional TC hazards in potentially predictable ways, thus providing a framework for translating seasonal climate forecasts into regional landfall estimates.
Talk 5: Integrating Multi-Scale Surveillance to Improve Infectious Disease Forecasting in Long-Term Care Facility Populations
Presenter: Hanna Brosky
Mentor: Tom McAndrew and Gabrielle String
Department/College: Biostatistics and Health Data Science
Traditional public health surveillance captures broad, county or state-wide infectious disease data to inform public health decision making. While crucial to disease mitigation, prior work has shown that residents of long-term care facilities (LTCFs) experience different disease exposure, transmission, and health outcomes that are not fully reflected in population-wide surveillance. In this project, we will develop a Bayesian epidemic model to forecast counts of infectious diseases within a long-term care facility in Bethlehem. Novel to our work, we will augment sparse data at the facility with state-level surveillance data available from the Pennsylvania Department of Health. We hypothesize that an epidemic model trained on long-term care facility data plus state-level data will outperform a control model trained on long-term care facility data alone. If model predictions are useful, this work will provide a scalable framework for supporting infection prevention and public health decision making for long-term care facility populations.
Talk 6: A Unified Cross-Dimension and Cross-Modality Representation Learning Framework for Multimodal Tensor Data
Presenter: Jiawei Guo
Mentor: Meng Zhao
Department/College: Industrial and Systems Engineering
Modern decision analytics increasingly relies on high-dimensional multimodal data collected from heterogeneous sources, where complex dependencies exist both across tensor dimensions and across modalities. Existing tensor learning methods effectively preserve multi-way structures but often overlook cross-dimensional interactions, while multimodal fusion approaches primarily model modality-level relationships after flattening structured data. To address these limitations, we propose Relational Cross-dimensional Bridge matrix (RX-Bridge) for Multimodal Tensor Representation, a unified representation learning framework that explicitly models structural relationships through learnable bridge operators connecting dimension-specific latent representations. Unlike conventional attention mechanisms that encode interactions implicitly, the proposed bridge operators maintain relationship-level representations that enable adaptive information propagation across tensor dimensions and modalities. We further develop a bridge optimization strategy and a structured fusion mechanism to generate compact and informative representations for downstream decision tasks. The proposed framework is theoretically analyzed in terms of its structural representation capability and computational scalability. Extensive experiments on multimodal emotion recognition and brain activity classification demonstrate that RX-Bridge consistently outperforms state-of-the-art tensor decomposition, multimodal fusion, and Transformer-based methods, highlighting its effectiveness for high-dimensional decision analytics.
Talk 7: Predicting Ordered ssDNA-SWCNT Hybrid Structures with Interpretable Bayesian Optimization
Presenter: Luke Wang
Mentor: Anand Jagota
Department/College: Bioengineering
Single-stranded DNA-wrapped single-walled carbon nanotubes (ssDNA-SWCNTs) are promising nanoscale hybrids because DNA sequence can influence how the polymer organizes on the nanotube surface. However, predicting which sequences form ordered structures remains difficult. My project addresses this problem by developing an interpretable Bayesian optimization model that connects DNA sequence, nanotube geometry, and physically motivated energy terms to the experimentally observed ordering behavior of ssDNA-SWCNT hybrids.
The model evaluates candidate wrapped structures by calculating energy contributions such as base stacking, hydrogen bonding, backbone deformation, torsion, elasticity, and chirality-dependent interactions. Bayesian optimization is then used to search this parameter space and identify structures with low predicted energy. I use experimentally labeled TC-rich and TA-family sequences to test whether the optimized energy landscape can distinguish ordered from disordered hybrids. A key part of the work is not only asking whether the total energy is low, but also asking which energy terms and landscape features control predictive accuracy. For example, hydrogen-bonding and stacking terms may affect sequence selectivity differently than elastic or torsional terms.
This project combines molecular modeling, statistical thermodynamics, and machine learning to build a more transparent prediction framework for DNA-directed nanotube assembly. The long-term goal is to move from fitting individual structures toward understanding which physical interactions make a sequence likely to form a stable ordered hybrid. This would help guide the design of ssDNA sequences for nanotube separation, sensing, and programmable nanomaterials.
Talk 8: Effect of turbulence scales on the near-wake behavior of a tidal stream turbine
Presenter: Cong Han
Mentor: Arindam Banerjee
Department/College: Mechanical Engineering and Mechanics
Tidal currents are a promising source of renewable energy, harvested by underwater turbines, much like wind turbines. Real tidal channels, however, are far from smooth: the flow is highly turbulent, filled with swirling eddies of many different sizes. These eddies interact with the wake - the slower, disturbed flow left behind a turbine - and govern how quickly the wake regains energy, which in turn determines the layout of tidal farms. Prior studies have shown that the overall strength of turbulence, quantified by the turbulence intensity, accelerates wake recovery by triggering an earlier breakdown of the tip vortices [1]. What has remained largely unexplored is the role of eddy size, characterized by the integral length scale, which represents the most energetic coherent structures in the flow. We address this gap through an experimental campaign in a water tunnel equipped with an active-grid turbulence generator, a device that allows turbulence intensity and eddy size to be controlled independently. Using a 1:20 scale tidal turbine model, we generated three inflow conditions with distinct turbulence intensities and integral length scales. Comparing these cases isolates the effect of eddy size at nearly constant turbulence intensity. Wake measurements using stereoscopic particle image velocimetry reveal how turbulence scales shape tip-vortex evolution, the entrainment of energy back into the wake, and the dominant frequencies and side-to-side meandering of the near wake. These insights can inform how future tidal energy farms are laid out and designed to extract the most power from turbulent coastal waters.
Talk 9: Deployable Architectural Systems Addressing Climate Justice In Small Cities
Presenter: Lauren Scott
Mentor: Wes Hiatt and Karen Pooley
Department/College: Art, Architecture and Design
As climatic and social conditions destabilize faster than the construction timelines required for permanent infrastructure, Allentown's documented temperature increases, and high asthma rates expose the limits of conventional heat mitigation. This research treats deployable shade, water, and wind-management systems not as stopgaps but as a deliberate mode of practice: soft interventions that avoid habitat or infrastructural harm, built from accessible off-the-shelf materials, assembled and maintained by non-specialists, and designed with material life cycles in careful consideration. These systems are conceived as program-less and site-less but scale-specific, engineered for a logic of building methods and assemblies that can be adapted to a vacant lot, a city block, or a citywide network, and transferable to other small cities facing comparable conditions. Operating in the interim between immediate heat exposure and the years-long timelines of permanent infrastructure investment, these systems generate lived, community-held evidence. Stewardship of a deployed and adaptable system produces qualitative data on spatial use, comfort, and vulnerability that directly informs longer-term municipal planning. In this case, Allentown's Climate Action Plan is informed through a series of pilot projects, while the act of deploying and maintaining the intervention returns spatial knowledge and decision-making power to residents, treating spatial autonomy as a critical climate justice outcome in its own right. The resulting toolkit of joints, materials, and deployment protocols is offered as transferable infrastructure: a way of working that equips under-resourced small cities to act immediately, accumulate evidence, and use authentic community input to shape what gets built to last.
Talk 10: Looking for the genetic underpinnings of repeated trait evolution in cavefish
Presenter: Marcos da Silva
Mentor: Johanna Kowalko
Department/College: Biological Sciences
Evolution under similar environmental conditions often results in the same traits evolving repeatedly in independent lineages. While the same traits can evolve repeatedly through changes in the same genes, or through different genes, why some genes are repeatedly targeted, and the functional consequences of perturbing these genes, is poorly understood. Cave animals are striking examples of repeated trait evolution, as multiple lineages have evolved similar traits, including loss of eyes and reductions of pigmentation. Astyanax mexicanus is a powerful model for the investigation of the genetic and evolutionary basis of cave-evolved traits, as it is a single species that has extant populations of surface fish and cavefish. Further, there are multiple populations of cavefish, some of which have evolved traits independently such as eye loss, reduced pigmentation and reductions in sleep. Here, we assess if genes previously identified as harboring high frequency alleles that contain coding changes relative to surface alleles in multiple cavefish populations, including the genes rgrb, mtnr1c and rgs9a, underlie cave-evolved phenotypes. We performed a series of phenotypic assays in lines of the surface morph of A.mexicanus in which mutations in these genes were generated using CRISPR-Cas9, including quantification of eye size and morphology, optic motor response (OMR), larval feeding, and sleep. This approach allowed us to assess the role of these genes in multiple cave-evolved traits, expanding our knowledge of drivers of repeated evolution.
Poster Abstracts
Poster #1: Modeling the Impact of Antigenic Distance and Immune Imprinting of Sequential Viral Infections on Disease Severity
Presenter: Nazia Afrin
Mentor: Thomas McAndrew and Gabrielle String
Department/College: Biostatistics and Health Data Science
For many diseases, there is considerable viral diversity with multiple lineages circulating in the target populations. At various times after the primary infection, sequential infection can occur. Here we introduce a simple model that presents a balance between two distinct adaptive immune responses: cross-reactive antibodies derived from pre-existing antibodies and specific antibodies generated during secondary infection. We have analyzed outcomes of interest for the individual host and for the virus, including disease transmission outcomes and antibody portfolio outcomes. Our analytical and numerical results suggest that Cross-reactive (C) antibodies dominate the Adaptive (A) immune response when viral strains are similar, even if the initial cross-reactive response is sufficiently small; thus, the waning period does not affect the ratio A: C significantly. However, the waning period and antigenic similarity both matter to viral production and host damage (greatest production and most damage when the initial cross-reactive response is large and viruses are dissimilar). In addition, sequential infections with a circulating strain that is antigenically very similar to a previous infection may limit the host’s ability to mount a diverse portfolio of immune responses. Therefore, understanding how immune imprinting shapes overall host immune response can be critical for effective protection during an epidemic.
Poster #2: Toward Scalable and Accessible Indoor Navigation: Lessons from Real-World Deployments
Presenter: Ajay Abraham
Mentor: Vinod Namboodiri
Department/College: Biostatistics and Health Data Science
Indoor navigation remains challenging in complex environments where GPS is unavailable and spatial information can be difficult to access. These challenges are particularly important when designing for people with different abilities. We developed and deployed an indoor navigation system that combines two localization approaches: Bluetooth beacon-based localization and visual localization using a smartphone camera. Our work examines how localization, navigation, and accessibility can be addressed together to create systems that are practical for deployment across diverse indoor environments.
The system has been deployed in four settings: a university building, a public events facility, a rehabilitation hospital, and the headquarters of a nonprofit organization serving people who are blind. User studies conducted in the university and rehabilitation hospital settings have provided insights into both technical and human-centered challenges. Our deployments have revealed challenges in achieving robust visual localization across changing environmental conditions and visually ambiguous spaces. They have also highlighted that accessible indoor navigation involves more than providing turn-by-turn directions. Users may first need to explore their surroundings, understand nearby points of interest, and then determine where they want to go. Designing ways to present exploration and navigation information that are clear, intuitive, and accessible to people with diverse abilities remains an important research challenge.
These experiences have shaped our ongoing work toward a scalable deployment pipeline that reduces the effort required to configure the system for new environments. We are exploring automated approaches for generating digital maps from existing floor plans or LiDAR point clouds, collecting imagery using smartphones, 360-degree cameras, and handheld LiDAR scanners, and training localization models with minimal or crowdsourced data. Our long-term goal is to enable seamless indoor navigation across buildings, connecting points of interest within and between buildings while supporting the diverse needs of users.
Poster #3: Laser-Induced Breakdown Spectroscopy for Waste Feedstock Characterization in Gasification Systems
Presenter: Havva Hande Cebeci
Mentor: Carlos E. Romero
Department/College: Mechanical Engineering and Mechanics
Feedstocks used in gasification systems, such as waste, can pose various challenges due to their heterogeneous nature, variability in ash-forming inorganic elements, and calorific values, and the need to maintain controlled feedstock specifications. These variations can impact gasifier stability, ash-related behavior, and overall process performance, as well as impact syngas yield and quality. A machine learning (ML)-enhanced Laser-Induced Breakdown Spectroscopy (LIBS) concept was investigated for in situ, near-real-time measurement of feedstock properties, aiming to provide time-sensitive fuel data to support gasifier operators and future feedback and feedforward control schemes. While LIBS can provide elemental spectroscopic emissions from solid feedstocks, including materials moving on conveyor belts, ML algorithms can process the spectral data to provide a range of measurable parameters of interest to gasifier operators. In this study, LIBS was used to characterize waste biomass samples. For each sample, 20 replicate LIBS measurements were performed, with 800 laser shots collected per measurement. Measurement repeatability was evaluated by calculating the relative standard deviation (RSD) for selected elemental emission lines. Inorganic elements, including Al, Ca, K, Mg, Na, Fe, Si, and Ti, were analyzed, and feedstock higher heating values (HHV) were calculated and compared to laboratory data obtained from material sampling.
Poster #4: Influence of the Grain-Flow Orientation after Hot Forging Process Evaluated through Rotational Flexing Fatigue Test
Presenter: Charles Chemale Yurgel
Mentor: Wojciech Z. Misiolek
Department/College: Loewy Institute, Materials Science and Engineering
Forged mechanical components have greater strength than those produced by other manufacturing processes, offering improved mechanical properties compared with casting or machining. The metal flow during the forging process leads to crystallographic texture modifications and can be macroscopically visualized as the so-called grain-flow orientation (GFO). This work presents the effect of GFO on fatigue life by using a rotational flexing fatigue test. The tests that were performed using SAE 1045H steel material, at deformation and transverse directions, showed the influence of GFO on the specimens’ mechanical properties compared with the reference samples taken from the machined rolled bar. Numerical modeling was applied to check if the amount of material was adequate for the actual forging process and the expected fiber development for the specimen, and to demonstrate the positioning of the blank within the tooling, the flow of the material filling the tool, and the intended fiber orientation. The experimental results showed that the forged samples with GFO oriented in the main deformation direction exhibited a higher fatigue life than the other tested configurations.
Poster #5: Driving Mechanism for Tidal Turbine Power Output Fluctuations
Presenter: Mohd Hanzla
Mentor: Arindma Banerjee
Department/College: Mechanical Engineering and Mechanics
Tidal turbines capture energy from the tidal currents in much the same way that wind turbines harness air currents. Estimates suggest that tidal energy could supply approximately 3–3.5% of U.S. electricity demand while providing a highly predictable source of renewable power. However, tidal currents are inherently turbulent, containing chaotic flow structures over a wide range of scales. These turbulent motions interact with turbine blades, producing unsteady loads and power fluctuations that are transmitted to the electrical grid, increasing the need for reliable control and synchronization strategies. Of interest here is to understand the mechanics of this turbulence-turbine interaction and factors contributing to this energy conversion dynamics. The challenge lies in mimicking such flow conditions in a laboratory environment while performing controlled measurements. In fact, the question being addressed here is which characteristics of tidal flows can we recreate to study its interaction with a scaled tidal turbine. This work focuses on generating such flows in a laboratory water tunnel. An in-house design turbulence generator is used to force chaos or intermittent turbulent fluctuations in the flow. A scaled turbine of diameter 0.28 m equipped with torque sensors allows measurement of the loading imparted by the turbulent inflow. Spectral decomposition of the turbine power (torque x angular velocity) provides insights into which turbulence scales/eddies contribute to and drive the power fluctuations. The results provide a framework for modelling tools to account for turbulence scales being relevant to the energy conversion dynamics.
Poster #6: Intermittent Suppression of Galloping of a Circular Cylinder under Turbulent Inflow: Influence of Integral Length Scale
Presenter: Izhar Hussain Khan
Mentor: Arindam Banerjee
Department/College: Mechanical Engineering and Mechanics
When a fluid flows past a cylindrical object, the flow can cause it to vibrate. This flow-induced motion is investigated experimentally in a water tunnel under elevated freestream turbulence. The streamwise turbulence intensity is maintained in the range of 13-16% across the inflow cases while the length scale of the turbulent flow structures, referred to as “Integral Length Scale” (ILS) is systematically varied using an active grid, a device that produces turbulence of a specified scale and intensity. Three turbulent conditions with time-averaged ILS of approximately 2D-6D are examined alongside a quasi-laminar baseline, where D is the cylinder diameter. Both vortex-induced vibration (VIV) of a plain cylinder and galloping of a cylinder with attached surface strips are studied. Turbulent inflow reduces VIV amplitudes and shifts the response branches to lower reduced velocities relative to the quasi-laminar case. At higher reduced velocities, galloping is suppressed under turbulence, and this suppression is found to be distinctly bi-stable: the cylinder randomly switches between a full-amplitude galloping state and a strongly suppressed state with no intermediate amplitudes, ruling out chaotic behavior. Aspects of intermittency and hysteresis are observed in the suppression regime, while the switching frequency and the persistence of each stable state are found to be strong functions of ILS. Larger ILS produces a wide intermittency band with frequent state transitions, while smaller ILS yields more abrupt and persistent suppression beyond a threshold reduced velocity. These findings are characterized through amplitude statistics and time-frequency analysis revealing how turbulence structure governs galloping stability.
Poster #7: Representing Lake Snow and Ice in MOSART-Lake with an Offline LSTM Emulator
Presenter: Lingbo Li
Mentor: Hong-Yi Li
Department/College: Center for Catastrophe Modeling and Resilience
Seasonal lake ice and snow regulate heat, momentum, and water exchanges between lakes and the atmosphere, yet are difficult to represent efficiently in global lake and Earth system models (ESM). MOSART-Lake, the lake module of the Model for Scale Adaptive River Transport (MOSART), the river component of the Energy Exascale Earth System Model (E3SM), resolves lake-river water and heat exchange, storage-dependent geometry, and multilayer thermal stratification, but carries no ice or snow. We therefore introduce a snow and ice emulator built from long short-term memory (LSTM) layers. It is trained across global lakes on ERA5 reanalysis snow and ice states, from lake surface area and climate forcing predictors. We adopt a one-way offline coupling strategy that avoids embedding machine learning inference. Before each simulation, the trained emulator generates monthly ice fraction, ice thickness, snow-covered fraction, and snow depth and writes them to a driver file. At runtime, MOSART-Lake reads and screens these predictions for physically plausible ranges and cross-variable consistency, then partitions each lake into open water, bare ice, and snow-covered ice. These states further modify the existing physics: the open-water fraction scales evaporation and latent heat; area-weighted surface classes set effective albedo and shortwave transmission; snow and ice thermal resistance limits ice-bottom conduction; and ice cover suppresses wind-driven mixing, while MOSART-Lake alone advances lake temperature and its water and energy calculations. We evaluate the emulator on withheld reanalysis years and compare simulations with and without the emulator, concluding with a discussion on current limitations and future directions for lake snow-ice representation in ESMs.
Poster #8: Heat-stiffening mineral-polymer physical networks
Presenter: Peter Park
Mentor: Dimitrios Vavylonis
Department/College: Physics
Cytokinesis in fission yeast is organized by cortical "nodes", macromolecular complexes anchored to the plasma membrane that mature into the contractile ring. Cdc15, an F-BAR domain protein with a large intrinsically disordered region (IDR), is a key node-anchoring protein whose membrane binding, oligomerization, and phase separation are regulated by IDR phosphorylation. The molecular mechanisms linking phosphorylation state to Cdc15's conformation, membrane binding, and self-assembly remain elusive. We used coarse-grained Martini3 molecular dynamics to simulate full-length Cdc15 in phosphorylated and dephosphorylated states in solution. Standard Martini3 parameters overly compact the IDR relative to experimental and theoretical predictions; by introducing additional IDR-water interactions, we tuned the forcefield to reproduce IDR dimensions consistent with experimental and prior computational data for both phosphorylation states. Using the F-BAR domain alone, we then simulated membrane binding and found that residues previously implicated separately in membrane binding and in end-to-end oligomerization functionally overlap, suggesting these activities are mechanistically coupled rather than independent. We also simulated multiple F-BAR dimers on a membrane surface and propose a possible pathway by which they assemble end-to-end into higher-order oligomers. Together, these results provide a Martini3 framework suitable for simulating IDR-containing membrane proteins and offer a molecular mechanism for how phosphorylation-dependent conformational change in Cdc15 drives its membrane recruitment and self-assembly into cytokinetic nodes.
Poster #9: Evolutionary Rate and Selection in Diet-Related Genes across Mammals
Presenter: Alexander Seaver
Mentor: Wynn Meyer
Department/College: Biological Sciences
Rates of gene evolution can be tested to discern selective pressures governing adaptation. Here, we specifically examine the rates of gene evolution across mammals grouped by diet (herbivory, carnivory, and omnivory) to examine selective pressures acting upon diet phenotypes. Using the various tools within the software package HyPhy (Hypothesis testing using Phylogenies), we can test for several types of selection acting upon genes in a diet-dependent manner. For example, using the tool RELAX to detect relaxed or intensified selection, we find that the gene encoding Pancreatic Lipase (PNLIP) exhibits relaxed selection in herbivory and omnivory. This result indicates that the digestion of dietary lipids, the primary function of PNLIP, is under less selection in these diets compared to carnivory. These results open new avenues for research into the comparative enzymatic efficiencies of mammalian orthologues of PNLIP across diets, and could be used to better understand dietary diseases and disorders moving forward. In total, we find 260 genes exhibiting relaxed selection in herbivory compared to all other diets and 431 genes exhibiting intensified selection in omnivory. Similar to the insights possible for PNLIP outlined above, these additional results build upon knowledge of those genes whose function is more or less important to fitness in various diets. We also utilize BUSTED to test for branch specific selection in the mammal phylogeny and aBSREL to test for episodic selection, both tools within HyPhy. Together, these results grant new insights into the evolution of diet related genes across a wide-array of mammals.
Poster #10: Surface Deformation and Jump-To-Contact Instabilities in Soft Substrates
Presenter: Reshma Siddiquie
Mentor: Anand Jagota
Department/College: Bioengineering
Attraction between two surfaces often leads to jump-to-contact, a nucleation event that is the complement of nucleation of a defect under tensile loading. This event contains important information about the interaction forces, but measuring such mechanical instabilities in soft materials just before and after contact presents significant challenges. In this study, we utilize Newton's ring interference patterns observed through an inverted optical microscope to precisely capture the jump-to-contact and deformation behavior of a soft polydimethylsiloxane (PDMS) substrate as it approached by a glass indenter. The experimentally measured deformation profiles and jump instabilities prior to and contact are used to evaluate adhesive forces between the glass and PDMS. An integral equation is employed to convert measured surface displacements into surface tractions, and is validated using Johnson–Kendall–Roberts (JKR) adhesion theory post-contact. The investigation is further extended to PDMS substrates with controlled adhesion (via UV ozone treatment) and varying elastic moduli, achieved through different base-to-curing-agent mixing ratios.
Poster #11: Do People Perceive Flood and Drought Risks Differently?
Presenter: Meraj Sohrabi
Mentor: Ethan Yang
Department/College: Center for Catastrophe Modeling and Resilience
Floods and droughts are becoming more frequent and costly, yet communities often respond to these hazards in unexpected ways. For local governments, one of the biggest challenges is understanding what motivates people to prepare before disasters happen. Without that knowledge, it is difficult to design policies that effectively reduce risk and strengthen community resilience. This research explores a simple but important question: Do people think about flood and drought risks separately, or do they view them as part of a broader sense of environmental risk? Using survey responses from 428 homeowners across three urban areas in Pennsylvania, we examined how people perceive these two hazards and whether those perceptions influence their willingness to take protective actions. We found that people who are more concerned about drought are also more likely to be concerned about flooding, suggesting that many individuals think about water-related hazards as a whole rather than as separate threats. We also found that people who feel more threatened by these hazards are much more likely to take steps to protect themselves. However, this relationship differs from one community to another, highlighting the importance of local context. By revealing how people perceive and respond to multiple hazards, this work offers new insights for designing more effective, community-centered hazard mitigation strategies. Understanding how people think about risk can help decision-makers develop policies that encourage preparedness and build resilience in the face of a changing climate.
Poster #12: Experimental Assessment of the Influence of TiO2 Grain Size on Grain Boundary Energy Values Derived using the Mullins Analysis
Presenter: Koen Verrijt
Mentor: Helen Chan and Jeffrey Rickman
Department/College: Materials Science & Engineering
Grain boundary energies play an important role in processes such as grain growth and nucleation in ceramic materials. These energies are commonly determined based on the dihedral angle (Ψ) of groove profiles that develop as polished surfaces are heated below the sintering temperature. Values of Ψ are typically derived using the well-known Mullins analysis[1] for a single isolated grain boundary. However, as grooves and ridges grow relatively large with longer grooving time, higher grooving temperature, or smaller grain size, they can start to interact with those on the opposite side of the same grain. Although the actual Ψ remains constant at thermodynamic equilibrium, the effect of this interaction on the derived Ψ is not well established. This work addresses this question by using atomic force microscopy to track the apparent change in Ψ distribution of polycrystalline TiO2 as a function of grooving time. Hundreds of grains were examined, encompassing coarse grains with isolated grooves, fine grains where diffusion fields have coalesced, and intermediate cases. Furthermore, existing models in the literature that expanded Mullins analysis with the interaction between neighboring grooving profiles were used to predict the error in the derived Ψ. The potential for applying appropriate correction factors for different grain size/grooving conditions will be discussed. Additionally, the effect of grain orientation difference on CoTiO3-TiO2 diphase boundary energy values was determined from samples with a CoTiO3 single crystal embedded in polycrystalline TiO2.
[1] Mullins WW. Theory of thermal grooving. J Appl Phys. 1957;28(3):333-39.
Poster #13: Acrolein Synthesis from Oxidative Coupling of Biomass-Derived Methanol and Ethanol by Supported Monolayer MoOx/Fe2O3 Catalyst
Presenter: Shuting Xiang
Mentor: Israel Wachs
Department/College: Chemical & Biomolecular Engineering
The global acrolein market is substantial, with an estimated value of USD 1,178 million in 2024. Currently, acrolein is primarily produced from fossil-derived propylene. Therefore, it is highly significant to develop sustainable methods for acrolein production. Here, we report the selective formation of acrolein from biomass-derived methanol and ethanol oxidation over a model supported monolayer MoOx/Fe2O3 catalyst. In this study, a well-defined surface MoOx monolayer on Fe2O3 was synthesized, characterized, and investigated for acrolein synthesis.
This monolayer MoOx/Fe2O3 catalyst was synthesized by incipient-wetness impregnation. It was confirmed that there is a complete surface coverage of Fe2O3 by a monolayer of surface MoOx species without formation of bulk MoO3 or Fe2(MoO4)3. IR revealed that after the MoOx deposition on Fe2O3 support, the surface MoOx monolayer replaced surface Fe-OCH3/OCH2CH3 with Mo-OCH3/OCH2CH3. Temperature-programmed surface reaction (TPSR) demonstrates that methanol selectively forms only formaldehyde on Mo redox sites, indicating that acid-base reactions on Fe2O3 were completely suppressed by MoOx monolayer. During ethanol oxidation, acetaldehyde forms as the redox reaction product. Under mixed methanol/ethanol and O2 flowing conditions, the dynamic surface MoOx redox sites enabled C-C coupling leading to acrolein (CH2=CHCHO) formation. Operando UV-Vis spectroscopy showed that the presence of gas O2 keeps the sample fully oxidized, consistent with a surface Mars-van Krevelen mechanism. By establishing clear structure-function relationships for the well-defined model catalyst of supported monolayer MoOx/Fe2O3, these findings offer significant guidance for design of next-generation oxidation catalysts. Ultimately, this study enables new and sustainable pathways for producing high-value chemicals from renewable biomass-derived feedstocks.
Poster #14: Looking for the genetic underpinnings of repeated trait evolution in cavefish
Presenter: Marcos da SilvaReshma Siddiquie
Mentor: Johanna Kowalko
Department/College: Biological Sciences
Evolution under similar environmental conditions often results in the same traits evolving repeatedly in independent lineages. While the same traits can evolve repeatedly through changes in the same genes, or through different genes, why some genes are repeatedly targeted, and the functional consequences of perturbing these genes, is poorly understood. Cave animals are striking examples of repeated trait evolution, as multiple lineages have evolved similar traits, including loss of eyes and reductions of pigmentation. Astyanax mexicanus is a powerful model for the investigation of the genetic and evolutionary basis of cave-evolved traits, as it is a single species that has extant populations of surface fish and cavefish. Further, there are multiple populations of cavefish, some of which have evolved traits independently such as eye loss, reduced pigmentation and reductions in sleep. Here, we assess if genes previously identified as harboring high frequency alleles that contain coding changes relative to surface alleles in multiple cavefish populations, including the genes rgrb, mtnr1c and rgs9a, underlie cave-evolved phenotypes. We performed a series of phenotypic assays in lines of the surface morph of A.mexicanus in which mutations in these genes were generated using CRISPR-Cas9, including quantification of eye size and morphology, optic motor response (OMR), larval feeding, and sleep. This approach allowed us to assess the role of these genes in multiple cave-evolved traits, expanding our knowledge of drivers of repeated evolution.
Poster #15: Structural health monitoring of large structures using mobile sensing
Presenter: Kevin Theunissen
Mentor: Shamim N. Pakzad
Department/College: Civil and Environmental Engineering
Structural Health Monitoring (SHM) is crucial for ensuring the safety of structures. It enables early damage detection, extends service life, and reduces repair costs. Today, SHM is widely applied to various types of structures, including bridges. Non-destructive evaluation techniques are used to monitor these structures. Data such as displacements, velocities, and accelerations can be collected and processed into various indicators, such as natural frequencies and mode shapes, which can then be used to compare the current state of a structure with its reference state. If a discrepancy is detected, the structure is considered damaged. One of the main current challenges in SHM is the monitoring of large structures, such as bridges. Monitoring these structures can become costly and inefficient when using traditional sensors like accelerometers. As a result, new monitoring techniques are emerging, such as mobile sensing, a novel paradigm offering numerous advantages over conventional stationary sensor networks. With only a few sensors, mobile systems can capture comprehensive spatial information. Moreover, mobile sensing can be combined with the ubiquity of smartphones and Internet of Things (IoT) to create large-scale sensor networks capable of contributing to structural health assessment. However, a major drawback is the contamination of the collected data. To address this issue, new algorithms based on Artificial Intelligence (AI) are being developed to clean the measured signals and extract the bridge's true accelerations. My current research at Lehigh University focuses on advancing the application of mobile sensors combined with AI approaches and using efficient electric transient systems for SHM.
Poster #16: Non-monotone stochastic derivative-free optimization algorithms
Presenter: Trang H. Tran
Mentor: Luis Nunes Vicente
Department/College: Industrial & Systems Engineering
In derivative-free optimization (DFO) one minimizes functions for which the gradient is unavailable or expensive to compute. In many applications, objective function values and gradients are noisy due to simulations or system randomness. Current DFO algorithms accept trial points when a certain monotone decrease of the objection function is achieved. However, when applied to complex landscapes, such a requirement may trap the algorithm in a neighborhood of sub-optimal solutions. Non-monotone techniques allow the acceptance of trial points with temporary increases in the objective value, but have only been developed in the deterministic case. The innovation in this study is the development of non-monotone DFO techniques in the stochastic case where evaluations are noisy. We have been investigating various non-monotone strategies, including those that replace the estimated function value at the trial point with a convex combination of past and present estimates, or with the maximum of previous observed function values. In this talk, we will present a numerical comparison of the various non-monotone techniques for both linesearch and direct-search methods. The numerical results are obtained for two sets of problems from the CUTEest collection (with different levels of multimodality) and analyzed through performance and data profiles. Our results demonstrate that non-monotone techniques exhibit superior performance compared to their monotone counterparts. This empirical evidence motivates our current theoretical investigation of the rate of convergence of non-monotone stochastic DFO algorithms. This is joint work with Anjie Ding and Luis Nunes Vicente.
Poster #17: Structure of DNA-SWCNT Hybrids for SWCNT Sorting and Biosensing
Presenter: Luke Wang
Mentor: Anand Jagota
Department/College: Bioengineering
Single-walled carbon nanotubes (SWCNTs) hold great promise for optical biosensors, yet isolating single-chirality populations and maximizing signal remain major barriers. Single-stranded deoxyribonucleic acid (ssDNA) wraps SWCNTs in water and imparts strong sequence-dependent selectivity, but the structural principles underlying this recognition are only beginning to emerge. Here we combine coarse-grained molecular dynamics, single-molecule near-infrared fluorescence spectroscopy, and supervised machine learning to map how ssDNA sequence, helical handedness, and binding free energy interrelate. A new handedness-propagation metric quantifies twist transfer along the polymer, while potentials of mean force expose sequence-specific detachment barriers. Comparative analysis across twelve ssDNA sequences and three chiral nanotube species uncovers a spectrum of wrapping geometries—ranging from shallow zig-zags to tight helices—each linked to a characteristic binding-energy fingerprint and optical response. These insights provide a framework for rationally pairing ssDNA sequences with target nanotube chiralities, advancing both high-purity sorting and next-generation biosensor design.
Poster #18: Prompting Transportation Electrification with a 99.55%-Efficient GaN-Based Multilevel High-Power Wireless Power Transfer Converter
Presenter: Yue Wu
Mentor: Fei Lu
Department/College: Electrical and Computer Engineering
In the recent decade, the transportation electrification industry has undergone a booming development. To meet the surging need for efficient and user-friendly charging methods, high-power wireless power transfer (WPT) technologies have attracted increasing attention due to their merits of contactless power delivery, enhanced charging safety, and perfect electrical isolation compared to conventional plug-in charging methods. However, existing WPT technologies suffer from the bottle necks of limited power ratings, and unsatisfactory power transfer efficiency, thus driving urgent demands for better power converters and magnetic couplers. As the core of WPT systems, power converter is critical for improving voltage endurance, rated power level, power density, and conversion efficiency. To address these issues, our research team has proposed a new GaN-based multilevel converter with the straightforward diode-clamped neutral point clamp circuit topology. Meanwhile, a mixed signal driven scheme is proposed to precisely control eight GaN FETs at a high frequency up to 85kHz, enabling efficient and reliable operation without the need for complex close loop control. Experiments prove that this converter achieves a voltage endurance of 1.1kV, a power level of 18.61kW, a peak power conversion efficiency of 99.55%, and a remarkable power density of 206.78kW/m2. This converter offers a simple, high-performance, and cost-effective solution for next-generation high-power WPT systems, with the great potential contributions to the advancement of transportation electrification industry and energy-efficient future.
Poster #19: Ethylene to Propylene through Simultaneous Ethylene Dimerization and Olefin Metathesis with Dual-Site Supported 8%NiSO4-8%ReO4/γ-Al2O3 Catalyst
Presenter: Shuting Xiang
Mentor: Israel Wachs
Department/College: Chemical & Biomolecular Engineering
Propylene is a key petrochemical traditionally produced as a byproduct of steam cracking. However, the shift to lighter feedstocks due to increased shale gas production has reduced propylene yields, creating a need for new "on-purpose" production methods. Ethylene-to-propylene (ETP) conversion via ethylene dimerization and olefin metathesis is a promising alternative. In this study, a dual-site 8%NiSO4-8%ReO4/γ-Al2O3 catalyst was synthesized, where NiSO4 facilitates ethylene dimerization to 2-butene, and ReO4 catalyzes the metathesis of ethylene and 2-butene to propylene. Comprehensive characterization-using in situ Raman, IR, UV-vis, X-ray absorption spectroscopy, and High Sensitivity-Low Energy Ion Scattering (HS-LEIS)-identified surface (O=)2Re7⁺O2, O=S6⁺O3, and Ni2⁺Ox species on the alumina support. Chemical probing via temperature-programmed surface reactions (TPSR) and steady-state catalysis confirmed the distinct roles of these sites. Ethylene dimerization was only observed when both surface SO4 and NiOx were present, indicating a synergistic effect. Neither SO4 nor NiO6 sites catalyzed metathesis alone or together, while ReO4 sites exclusively facilitated olefin metathesis. Notably, incorporating NiSO4 enhanced the metathesis activity of ReO4/Al2O3, attributed to competitive adsorption on surface hydroxyls.