Turbidity is the measurement of how much light passes through a liquid. Turbidity is important to study as particles in turbid water can act as vectors to bacteria. In this study, images taken from the Sentinel-2 satellite measured the turbidity of beaches around Los Angeles. First, we compared satellite-derived turbidity from the Nechad algorithm to in situ turbidity samples, then turbidity was compared to Fecal Indicator Bacteria. This experiment was split up into three parts: first, the total suspended solids (TSS) test, measuring how many particles are in a sample. Second, the fecal indicator bacteria (FIB) test, measuring the probability of how much bacteria is in 100mL of the sample. Lastly, Google Earth Engine (GEE), a cloud-based platform for planetary-scale geospatial analysis, calculated turbidity from the Sentinel-2 satellite. After collecting water samples from Los Angeles beaches for in situ TSS and FIB tests, the correlation was found to be low (R2=0.2), suggesting that the clarity of the water does not translate to how much bacteria was in the sample. Yet, there was a correlation between in-lab TSS and the Turbidity from Sentinel-2 (R2=0.8), indicating that Sentinel-2 accurately measured turbidity. In the future, research should be done to compare TSS from different algorithms, satellites, and different beaches. This would have a major impact on society as satellites would be able to detect whether water is safe to recreate in.

Glioma patients are advised treatment and given a prognosis based on the review of Magnetic Resonance Imaging (MRI) scans. In contrast weighted scans, new contrast enhancement is largely recognized as a pertinent indicator of tumor progression. Pseudoprogression describes the phenomenon where new contrast enhancement appears but never exacerbates tumor growth and often resolves on subsequent scans. The human impact of this phenomenon is high, as patients risk unwarranted emotional, physical, and financial implications if pseudoprogression is interpreted to be and is treated as true progression. Previous studies on this topic are largely underpowered and have yielded inconsistent results. The Lai lab finds itself in a unique position with its access to decades of glioma patient data and MRI scans. In seeking answers to questions that can fundamentally alter a patient’s clinical course, a large-scale retrospective review of IDH-mutated glioma patients (n = 700) combining each patient tumor’s molecular characteristics, T1 axial MRI scans with and without contrast, and MRI reports written by board-certified neuroradiologists has been conducted. Using this data, new contrast enhancing spots can be tracked through subsequent MRI scans to be characterized as pseudoprogression versus true progression. Through this unprecedented systematic characterization of new contrast enhancement in IDH-mutated glioma patients, we expect to find distinguishing characteristics of pseudoprogression and true progression that can equip physicians to offer IDH-mutated glioma patients more accurate prognoses. This research stands to tangibly improve clinical decision making, while having the human value of alleviating patient stress and harm.

Intestinal barrier dysfunction, a prominent hallmark of aging, can be a result of increased intestinal permeability and has been associated with certain autoimmune and inflammatory diseases, as well as early death. A lesser-known consequence of untreated intestinal barrier dysfunction is its effect on the brain; namely the potential onset of neurological pathologies caused by systemic inflammation. In order to investigate the inner workings of this potentially causal relationship, we genetically induced the loss of intestinal barrier integrity using the GeneSwitch system to express RNAi targeting septate junction-associated genes, Snakeskin and Coracle, in fly enterocytes. The resulting effects on intestinal barrier function, brain pathology, healthspan, and lifespan of Drosophila were observed. Certain markers of aging health in the brain, such as actin hyperstabilization, have been found to be associated with loss of gut barrier function. Research on the potential use of the pharmacological agent, Cytochalasin D, to dismantle actin hyperstabilization in aging fly brains and its effect on lifespan and healthspan is ongoing. Overall, these experiments may be able to determine anti-aging solutions that work to preserve both intestinal and neural health, and lifespan longevity.

Federated learning, a machine learning technique, has been gaining popularity as a method of protecting user data privacy for modern-day devices while still providing a great user experience by only sending model updates to the server instead of exchanging sensitive data. However, utilizing a singular global model is extremely restricting as the data is exceedingly diverse, which limits the global model from maximizing performance for each individual client. This heterogeneously distributed data across multiple clients is the primary motivation in utilizing personalized versions of federated learning. In this research, we implement algorithms that use personalized federated learning techniques such as clustering clients and utilizing temporary models for communication. Various hyperparameters - batch size, local communication rounds, number of clients - are adjusted to maximize the algorithm’s accuracy levels. The algorithms are implemented using PyTorch (a machine learning library developed by Facebook’s AI Research Lab) and both trained and tested using the CIFAR-10 image dataset. Using random heterogeneously distributed data, the algorithms converge to higher accuracy levels through these personalization techniques when compared to traditional federated learning utilizing a singular global model. Personalized federated learning is a key component within future machine learning applications - such as connecting autonomous vehicles - as it combines the effectiveness of traditional machine learning with crucial cloud security and data privacy protection.

We explored the magneto-optic Kerr effect (MOKE), which refers to the changes in light reflected from a magnetized surface. In MOKE, the incident circular polarized light becomes elliptically polarized and its axis of polarization rotates after reflection. These changes are termed Kerr ellipticity and Kerr rotation angle, respectively. Our objective is to achieve milli-radian Kerr rotation angle readout resolution on ferro/ferrimagnetic materials, such as CoFeB and GdFeCo, with a MOKE setup at near-normal incidence. By modulating the incident light with a photoelastic modulator (PEM-100), magnetizing the sample by placing it between two solenoids, and extracting voltage signals from a photo-detector with a lock-in amplifier (SR830) and multimeter (Keithley 2000), a magnetic hysteresis loop relating applied magnetic field strength and Kerr rotation angle was generated using MATLAB. The CoFeB sample was a wedge, meaning its thickness changed linearly from 0.4 to 1.4 nm. For this CoFeB wedge, we observed many hysteresis loops along the wedge to find both quantitative Kerr rotation angle and perpendicular magnetic anisotropy at various thicknesses. The GdFeCo film was of uniform thickness, but of a nonuniform makeup as it was a composition gradient from one side being Gd-rich and the other being FeCo-rich. For the GdFeCo film, the loops generated near the Gd-rich side neared but did not reach indication of a magnetization compensation point (where the net magnetic moment is 0). Points around the middle of the film were dominated by out-of-plane anisotropy, and the FeCo-rich side displayed predominantly in-plane anisotropy.

Numerous species of animals are known to have magnetoreception, or the ability to detect the Earth's magnetic field, for orientation and navigation. However, more research is needed to confirm the underlying mechanics of magnetoreception in animals. Compelling evidence has suggested that dogs align to the Earth’s magnetic field during excremental activity if the nearby magnetic field declination, or the difference between true north and magnetic north, is stagnant. Nonetheless, this phenomenon needs a robust source of experimental data before it can be established. We are compiling a large image dataset of urinating and defecating dogs with citizen science and automating the analysis of geomagnetic metadata embedded within these images. We hope to verify whether canine alignment in urination and defecation depends on magnetic field declination. Initial results from a low sample size indicate dogs face random directions even when the percent change in magnetic field declination is less than 1%. However, the project will require more image submissions from across the world to yield more refined results. If dogs demonstrate magnetoreception in the course of this project, their potential role as experimental subjects will be pivotal in developing future magnetoreception research.

Reinforcement learning (RL) has been a growing subset of machine learning with increasing success and promise - but it has just begun to be used in complex, multiplayer environments and games. Through trial and error, an agent begins from fully random trials and finishes with sophisticated actions. We apply RL to the imperfect information game known as Liar’s Dice, which presents a challenging mix of two-player dynamics and partial information to explore. The game forces players to call bluffs and doubt opponents while reading others’ potential actions. Implementing reinforcement learning to imperfect information games allows us to find successful strategies and models in dynamic Markov Decision Process (MDP) environments that require sequential decision making. We employed the popular Q-learning method of RL to train agents that begin with random actions or to use a combination of fixed strategies against others. An agent employing Q-learning improved its win rate from 50% to only 65% within 1,000,000 episodes against a simple agent. Another agent that made decisions based on various fixed strategies available increased its win rate from 11% to 77.7% with 100,000 episodes. These rates demonstrate the variability of Q-learning in a game with partial information. In future work, we may compare counterfactual regret minimization and more state-of-the-art RL algorithms, which would expand our understanding of various methods on partially observed, dynamic environments. By studying this game, we hope to broaden our results to the education space, a similar Markov process where individuals also make decisions sequentially.

Advancing research on tooth regeneration has the potential to address various dental problems such as tooth loss due to injuries, cavities, and genetic syndromes. Human teeth have very limited regenerative potential because dental epithelial stem cells (DESCs) are naturally lost after tooth eruption. Identifying ways to recover or rederive DESCs is therefore an important first step towards regenerating human teeth. This study utilized mouse as a model as its incisors maintain DESCs and grow continuously, providing a powerful system to study mechanisms regulating DESCs. DESCs differentiate to ameloblasts, which deposit enamel, the outer hard layer of teeth. Our goal is to experiment whether it is possible to revert ameloblasts back to their progenitor state and we will do this by re-activating Yes-associated Protein (YAP), a key transcriptional factor for maintaining DESCs. In this study, we upregulated YAP in the nucleus of ameloblast by deleting YAP inhibitors, LAST1/2, in the adult mouse dental epithelium using the Cre-LoxP system. This resulted in expansion of the epithelium with increased proliferation in the region typically occupied by ameloblasts. These newly generated proliferating cells are therefore either converted from differentiated ameloblast or from expanded progenitor cells. To distinguish between these possibilities, we are currently performing BrdU/EdU double staining. Discovering the reprogramming potential of YAP in reverting ameloblasts to dental progenitors will further our understanding of the mechanism that regulates tooth homeostasis and help develop tooth regeneration strategies.

Analysis of the relative orderings of the differences between model predictions as opposed to a quantitative method is often required in cases such as customers expressing their preferences instead of giving numerical scores. The purpose of this research project is to use an algorithm based on the alternating direction method of multipliers (ADMM) to solve large-scale non-metric multidimensional scaling (NMDS) problems. The NMDS problem seeks to optimize the Gram matrix of the calculated position vectors by minimizing violations of the inequality constraints that express the ordering relations of their pairwise distances. ADMM is a method for large-scale optimization which splits variable x into two parts and performs alternating optimizations over each part. The problem is coded using Python and Matlab, allowing us to see what fraction of the ordering of the original distances is preserved. We are working with randomly generated datasets. We are also working with more interesting data, including Swiss Roll and S curve data generated using Python, and real-world data such as sets of related images. For our initial results, which did not include any code for ADMM, the fraction of the ordering of distances preserved was quite high, indicating that the relative ordering of the original distances was preserved overall.

Electron beam therapy (EBT) utilizes electrons to kill cancer cells with up to 60% less radiation affecting surrounding healthy tissue compared to photon-based radiation therapies. EBT typically uses cm-scale beams; this project focuses on using Panofsky quadrupoles to guide sub-millimeter beams in a flexible and changeable trajectory so that beam placement, and therefore treatment outcomes, are improved. Flexible Printed Circuit Boards (PCBs) were designed in a Panofsky quadrupole-like geometry, which consists of parallel copper traces that generate a quadrupolar magnetic field. The flexible material of the PCB allows for manipulation of electron beams in hard-to-reach areas for deeper tissue treatment. Joule heating of the PCBs was simulated in COMSOL Multiphysics, and the limiting current density extracted. The limiting current density was used in magnetostatic simulations to find the magnetic characteristics of these devices. Particle tracing simulations were then performed to investigate efficiency of guiding electrons at different curvatures of the flex-PCB. Flex PCBs were fabricated for testing and the thermal response of the PCBs was experimentally measured using a FLIR OnePro thermal camera.

Biological sex differences in metabolism and metabolic disease incidence have long been reported. A female-biased protection against metabolic-related disease is observed until the age surrounding the final menstrual period, when disruption of endogenous estrogen action occurs. As such, estrogen receptor α (ERα) expression has been associated with indices of metabolic health, including insulin sensitivity and reduced risk of obesity and type-2 diabetes. To test the impact of ERα expression on adipocyte function, we generated fat-specific estrogen receptor α selective knockout (FERKO) and overexpression (Adi-ERαTg) murine models. We examined the impact of ERα expression on adipocyte size, RNA transcript and protein expression profiles, and mitochondrial form and function, as mitochondria are key regulators of metabolism and cellular health. We used histochemical staining methodologies, transmission electron microscopy, immunoblotting and RNA sequencing to show that mice with adipocyte selective overexpression of ERα possessed smaller adipocytes for better adipocyte function and exhibited a gene and protein profile supporting enhancement of mitochondrial function and oxidative capacity compared with controls. Electron microscopy analyses further revealed that ERα expression is positively correlated with mitochondrial cristae density and functionality of the electron transport chain. These findings point to the promising potential of leveraging ERα to combat metabolic dysfunction and reduce the risk of metabolic-related disease.