Mesothelin (MSLN) is a self antigen overexpressed in various cancer subtypes, such as mesothelioma, lung, ovarian, and pancreatic cancers, which makes it a viable candidate for targeted cancer immunotherapy. Because MSLN is expressed on the cell surface, flow cytometry, a method to measure cellular protein expression by laser excitation of fluorophores bound to antibodies, can be used to measure antigen expression on specific cell subsets of interest. In particular, we look to validate MSLN expression in naturally expressing cell lines and in mRNA lipid nanoparticle (mRNA-LNP) induced cells. However, there are no commercially available fluorophore conjugated antibodies for targeting MSLN, so we developed our own method of staining by directly conjugating a fluorophore to a MSLN antibody. KP4662 G12C cells—cells with a mutation associated with pancreatic cancer—were stained with various dilutions, tested for shelf stability and compared to our previous protocol of indirect staining by pairing a primary antibody that targets MSLN with a secondary antibody conjugated to a fluorophore that targets the primary antibody. Subsequently, we measured induced antigen expression in macrophages transfected with mRNA-LNPs that deliver antigen-encoding RNA to innate immune cells. We found that all the tested dilutions of both the old and new conjugations effectively identified MSLN expression, indicating the conjugation and fluorophore is stable and active at the tested conjugations. The optimization of this flow cytometry staining protocol will be applied to future assessments of MSLN protein expression in candidate mouse tumor cells and for investigating protein translational capacity following transfection with of MSLN-encoding mRNA-LNPs. 

Driven by industrialization and the rising global demand for energy, CO2 emissions are a major contributor to climate change. Limiting global warming to below 2°C and achieving net-zero emissions by 2050 requires the deployment of negative emission technologies (NETs). Capturing CO2 directly from ambient air is a particularly compelling NET approach, offering a diverse array of deployment scenarios. However, the cost of air contactors remains an obstacle to widespread deployment. This study involved the setup, calibration, and operation of a bench-scale contactor to systematically evaluate how CO2 absorption efficiency depends on solution pH. Mass flux, expressed in mol/m2/s, served as the primary performance indicator. During testing, air was introduced into the contactor at 0.4 m/s, while liquid distribution rate was held at 8.5 L/min. CO2 concentrations at the inlet and outlet were recorded at multiple pH setpoints, from pH ~12.7 to 10.4. Under these conditions, an average mass flux of 1.2 × 10⁻5 mol/m2/s was observed with no clear pH dependency, suggesting that the use of traditional packing media is limiting gas-side mass transfer for CO2 absorption due to its low concentration. This observation calls for the development of high efficiency packings tailored to CO2 absorption from ambient air. Future work will continue to explore a broader range of operating conditions, to identify optimal packing design parameters for practical carbon capture applications.

Partial Differential Equation (PDE) solvers are essential for modeling complex  physical systems but are computationally expensive for long-horizon simulations.  While recent graph neural network (GNN) and transformer-based surrogates  accelerate prediction, they often suffer from instability and error accumulation. We  present a physics-aware temporal prediction framework that achieves stable, long-  rollout predictions on unstructured meshes by combining GNN-based spatial  encoding, Graph Fourier Transform (GFT)–based spectral filtering, and transformer-  driven temporal modeling. Each PDE state is represented as a graph and encoded  into latent node embeddings, which are transformed into the spectral domain, low-  pass filtered to suppress high-frequency noise, and propagated forward in time via a  decoder-only transformer. The reconstructed states are obtained by inverse GFT and  a spatial decoder, ensuring spatial fidelity and preservation of physical invariants. Our  approach attains state-of-the-art accuracy on a plastic deformation dataset and  competitive performance on the more challenging cylinder-flow benchmark,  demonstrating reduced reconstruction error, improved stability, and physical  consistency over extended horizons.

This project explores the development of a modular, low-cost robotic system for microfluidic lab automation. Automated fluid control is achieved with ferrobots, magnetically actuated carriers designed to transport ferrofluidic droplets and reagents with high precision across microfluidic tracks. A standard 3D printer was repurposed by replacing its extruder head with an electromagnet, providing a programmable motion platform for droplet control. Complementary functional modules, including a vial decapper, liquid pump, and optical sensing unit, were also developed to extend the ferrobotic system with all functions needed for full experimental automation. Initial experiments validate the feasibility of using readily available hardware to achieve controlled droplet transport and liquid flow rate control. These results highlight a pathway toward accessible, reconfigurable, and affordable automation tools for biological research and diagnostics

High quality randomness is crucial to cryptographic security, where low entropy sequences can lead to exploitable vulnerabilities. True Random Number Generators (TRNGs) harness physical phenomena within hardware components, instead of predictable algorithms, to produce random sequences. Hardware based TRNGs are difficult to implement because of hardware's inherent deterministic nature. The Magnetic Tunnel Junction (MTJ) based TRNGs specifically are seemingly effective solutions due to their low power consumption, compact form, and high quality random streams.  This research investigates the reliability of bitstreams generated from MTJ based TRNG chips. To evaluate their randomness, we apply the National Institute of Standards and Technology (NIST) Statistical test Suite (STS), focusing on tests such as Frequency, Linear Complexity, and Approximate Entropy. These tests are designed to detect entropy bias, sequence dependencies, and flaws in post-generation that may undermine unpredictability.   Our results show that MTJ-generated bitstreams fail some of the NIST STS tests, notably the Rank and Approximate Entropy tests, which imply subtle dependencies and patterns exist. Yet, it passes the Linear Complexity test, which suggests there is structural complexity, making it difficult to reproduce. While it is not trivially predictable, these findings indicate that post processing strategies are necessary to elevate the quality of these streams.   By identifying and analyzing failures within the streams, this study contributes to the development of more secure TRNG designs that are practical for real world cryptographic applications.