Click here for the full 2021-2022 journal.
Congratulations to our 2021-2022 authors!
Kunwoo Kim - Evaluation of Plastic Materials for Football Helmet Facemasks Using Finite Element Simulation
American football reports the highest rate of head injuries including concussion in the United States. Although the use of football helmets has protected players and reduced severe head injuries, the number of concussion incidents has not decreased. A facemask mounted on the front opening of the helmet to protect the face is the second most impacted location of concussions. However, its high-rigidity material such as carbon steel or titanium makes the facemask disadvantageous in absorbing impact. This study aimed to assess the feasibility of using a plastic facemask to absorb impact in order to prevent concussion. Four different plastics, Polycarbonate (PC), Polyetherimide (PEI), Polymethyl Methacrylate (PMMA), and Acrylonitrile Butadiene Styrene (ABS), were selected to make the facemask, and a pneumatic ram impact test was conducted using numerical simulation. A PEI facemask was deformed but withstood the ram impact while preventing the ram from hitting the face. On the other hand, ABS and PMMA facemasks were cracked and failed, and a PC facemask was not cracked but allowed the ram to hit the face. This study indicated that if the plastic material was optimized, a PEI facemask would be feasible for absorbing the severe external impact, and protect players from concussions.
Click here for the full article.
ALICE FENG -A Deep Learning Model for Protein Abundance Prediction from RNA Data with Manifold-Preserving Regularization
A key challenge in single-cell multiomics study is to quantify the relationship between mRNA level and protein abundance. This relationship is complicated by the dynamic nature of mRNA and protein. In this paper, a deep learning regression model was proposed to predict protein abundance from mRNA expression data. However, overfitting was identified as a major source of error. Because different modalities of the same sample should concord to the same cell population structure, we invented a manifold-preserving regularization term to reduce overfitting induced by noise specific to training. By applying our model on CITE-seq data for 25 cell-surface proteins representing well-characterized markers, we observed an improvement of up to 30% on testing error. Thus, manifold-preserving regularization helps distill true mRNA-protein relationships from noisy data. We expect it to be generally applicable to other multiomics applications.
Click here for the full article.
SHREYA SREEKANTHAM - Evaluation of Gender Effects in Predicting Parkinson’s Disease from Voice: A Random Forest Approach
Parkinson’s Disease (PD) is the second most prevalent neurodegenerative disease in the world, affecting more than 10 million people. It has no cure, but early diagnosis of PD can help slow down the progression and improve the patient’s life quality. However, the diagnosis of PD is often subjective and inaccurate because its presentation varies widely between individuals. This study focuses on early PD diagnosis and evaluates biomedical voice parameters variation by gender using a novel Random Forest Algorithm (RFA). The study utilized a multivariate PD dataset extracted from the UCI Machine Learning data repository that consisted of 5,875 voice recordings from 42 subjects. The novel RFA introduced in this study both improves the accuracy for PD detection and consistently performs well across gender. In addition, the study identifies gender-based differences in the expression profiles of voice parameters that can be useful in future clinical applications.
Click here for the full article.
Eric Wang -Effects of Various Soil Microbiomes on Native and Invasive Plants
Many nature reserves use exclosures to preserve pockets of native plant biodiversity. These exclosures typically have a high proportion of native plants, as there are few invasive species to harm them, resulting in different plant-soil feedbacks (PSFs) within and outside of exclosures. PSFs alter the soil microbiome and have lasting effects on plant community composition; as such, they have important implications for natural ecosystem conservation. For this case study, three native and two invasive plant species were grown in soil inoculated with microbiomes collected from inside and outside of ten exclosures. Analyses of their biomass revealed that native species performed better in the soil microbiome from the exclosures, while invasive species’ growth was not significantly impacted by the different microbiomes. This research provides new insights into plants and the soil microbiome in the context of conservation and has important implications on the protection of natural ecosystems.
Click here for the full article.
RYAN PARK-X-Net: A Convolutional Neural Network for X-Ray Threat Detection
This paper proposes X-Net, a novel deep learning architecture that enhances airport security through the detection of dangerous objects in X-ray luggage scans. Scanning of luggage is a critical part of aviation safety but is alarmingly unreliable due to human error. This endangers the safety of millions of airline passengers. To eliminate this human error, several deep learning concepts engineered for the analysis of X- ray baggage scans are introduced. The different concepts are all part of one model, called X-Net. X-Net employs a network of deep convolutional lateral stacks, which combine vertical residual transpose blocks with inter-layer connections. This combination allows for multi-directional gradient flow, resulting in richer and more robust internal feature representations. These innovations enable X-Net to perform exceptionally well on real-world baggage scans, significantly enhancing public safety and potentially saving thousands of lives: X-Net detects malicious items 400% more accurately and 4200% faster than a human TSA officer. Moreover, this proposed approach provides novel and empirically useful deep learning tools that strengthen other fields of computer vision.
Click here for the full article.
LYON KIM- Genetic Association Testing and Predictive Modeling for Non- Small Cell Lung Carcinoma RNA Sequencing
We developed Python code to analyze a bulk RNA sequencing dataset consisting of lung tissues from healthy and non-small cell lung carcinoma (NSCLC) patients. Our preliminary goal was to find genes that were positively or negatively associated with cancer. To do so, we tested the null hypothesis, which stated that the average gene expression between cancer and non-cancer patients is equal. We rejected the null hypothesis since 86.8% of the genes showed a difference in expression. After finding genes associated with cancer, we built a machine learning logistic regression model from the gene expression data. To efficiently measure the performance of our method’s ability to predict cancer, we randomly split the data into training (80% of data) and testing datasets (20%) and used five-fold cross validation. By adjusting the probability threshold for classifying cancer, we created an ROC curve, representing the trade-off between the fpr (false positive rate) and the tpr (true positive rate). Ultimately, we hope that these mechanisms can help increase the chance for an early diagnosis of NSCLC, which is crucial to controlling and even preventing it.
Click here for the full article.
VAIBHAV MISHRA - DeepNeuroNet: A Novel Multiclass Model to classify Brain Tumors and
One of the most important applications of machine learning is the use of deep learning models in medical diagnosis and treatment. This application is particularly valuable in the early diagnosis of brain tumors and neurodegenerative diseases because earlier intervention leads to better prognosis and prevention of more fatal conditions. Among the most widely affected brain disorders are brain tumors, Alzheimer’s disease (AD), and Mild Cognitive Impairment (MCI). These diseases have a high incidence of 11.3% and a high mortality rate of 40% [6]. Therefore, this study aimed to help with the early diagnosis of these diseases through the development of a new multiclass convolutional neural network (CNN) model to classify glioma tumors, meningioma tumors, pituitary tumors, AD, and MCI from normal patients with an overall accuracy of 88.33%. DeepNeuroNet was the first model that used brain MRIs to classify both brain tumors and neurodegenerative diseases. This model would have many applications including brain tumor detection and possible treatment research in clinical settings and the potential to be used for the early diagnosis of brain diseases.
Click here for the full article.
ISHIKA NAG - Enhancing Efficacy and Cost Effectiveness of Air Filtration Systems by Optimized Nanoparticle Deposition
Every year, seven million people die from severe cardiovascular and respiratory diseases, caused by ambient and household air pollution. An increase in air pollution from particulate matter less than 2.5 microns in diameter (PM2.5) has shown to be a significant contributor of cardiovascular and respiratory diseases. The goal of this study was to develop an efficient and cost-effective air-filtration system by the impregnation of selected nanoparticles, utilizing their high surface-to-volume ratio to entrap PM2.5. The experimental set- up consisted of a wind tunnel with incense sticks as the particulate matter source, measured by laser particle detectors upstream and downstream of the filters. Results found that a mixture of zinc oxide, titanium dioxide & graphene improved filtration efficiency of a baseline filter by 206%. There was also a 70% improvement in the cost of the filters. The versatility and cost-effectiveness of this design makes it applicable for personal masks & filters, air-conditioning filters, car-cabin filters, and fire-fighting equipment. The correlation between air pollution and fatalities from viral infections suggests that abatement technologies with innovative filtration systems are critical in saving human lives.
Click here for the full article.
ANDY XU - Pecan: A Novel Approach to Energy Supply and Demand Forecasting in a Photovoltaic Microgrid
The adoption of renewable energy is crucial to curbing carbon emissions. Localized on-site generation methods such as microgrids are implemented due to their improved reliability and the ease of inclusion of renewable energy generation. However, current forms of renewable energy generation are unreliable. Accurate forecasting of both energy supply and demand are crucial in the transition towards a renewable energy grid and reducing reliance on fossil fuel reserves. Pecan is a novel solution that combines custom deep learning models for energy supply and demand forecasting with an artificial neural network solution. Pecan uses a novel loss function to prioritize grid stability while simultaneously decreasing carbon emissions through a lower error rate. Mean absolute percentage error (MAPE) was used to measure model performance and calculate emission reductions. The supply forecasting prediction from Pecan achieved a MAPE of 1.17%, and the demand forecasting prediction achieved a MAPE of 1.05%. The improved performance of Pecan increases the feasibility and profitability of microgrids and renewable energy solutions.
Click here for the full article.
Yaqoub Ahmad, Jessamine Qu, and Penelope Strong - Modeling the Stellar Kinematics of the Thick Disk and Halo of the Andromeda Galaxy
In studying the Andromeda Galaxy to better grasp its physical components—particularly its northeastern region— we utilized Python code to simulate its halo and disk. Using previously observed data and various formulas, such as one that calculates the density distribution of stars, we closely modeled the real dimensions of Andromeda’s halo and disk. Moreover, looking at different velocity dispersions along the height, radius, and angle axis helped us further understand Andromeda’s actual dispersions. Comparing models with different percentages of stars in the thick disk and halo, we found that both the thick disk and the halo had minimal effect on the observed dispersion. Furthermore, we had difficulty observing overarching dispersion trends brought about by changing the dispersion coordinate variables. Based on these observations, a more natural and substantial dispersion of Andromeda can be concluded. As we continue, our analysis will assist us in fine- tuning the model, more accurately simulating the Andromeda Galaxy, and eventually adapting the code to forward model any galaxy.
Click here for the full article.
KRITHIKA KARTHIK - Alzheimer’s Disease and A Prion-Like Protein: A Toxic Relationship
The role of TDP-43 and its prion-like features in Alzheimer’s disease (AD) represents a new avenue of research concerning the pathogenesis of the disorder. Research has focused on identifying proteins involved in inducing aggregation/toxicity of the illness, with the Tau and ß-amyloid proteins being primarily responsible. The TDP-43 protein was first discovered in 1995 and has attracted considerable interest in recent years. This review details the structural and functional characteristics of TDP-43. Special emphasis is given to the post-translational modifications and mutations that accompany neurotoxicity and protein aggregates found in the brain tissue of AD patients. The interface of TDP-43 with other proteins involved in AD progression is also elucidated based on studies in this regard. Investigations using animal models with the intent to identify potential therapeutic strategies to combat the disease have also been outlined in this work.
Click here for the full article.
FRANCESCA FROIO - Human Neural Stem Cell Perilesional Transplant Potential Recovery in Penetrating Traumatic Brain Injury
Traumatic brain injury (TBI) is a leading cause of death and disability in the United States [1-5] Despite exceeding the death rate of cancer by 3.5 times, there is inadequate treatment that directly targets the TBI lesion, in particular the lesions due to penetrating traumatic brain injury (pTBI)[1]; pTBI is a niche area of TBI injuries that focuses on a foreign object entering and harming the brain [2]. Since the discovery of neural stem cells in the subventricular zone (SVZ) and dentate gyrus (DG), exploration of transplantation treatment has become a topic of interest [6-8] The aim of this study was to understand how stem cell treatments could be a optimized to address penetrating traumatic brain injury. It was initially thought that neural cells were non regenerative in central nervous system (CNS) injuries and that adult neurogenesis is limited in the SVZ and DG. However, neural stem cells are still present within the subventricular cortex after the injury. This demonstrates how transplanting endogenous cells could be a better treatment option in comparison to the current treatments that only mitigate secondary injuries and symptoms[3,9,10] Indeed, a growing number of experiments and animal trials have shown that human neural stem cells (hNSCs) transplanted perilesional to the cavity have the potential to aid pTBI recovery [2,5,11,12].
Click here for the full article.
MIRIKA JAMBUDI - Glioblastoma Multiforme: A Therapeutic Review
Glioblastoma multiforme (GBM) is one of the most common forms of malignant brain cancer. Despite advancements in technology and treatment over the past century, GBM remains largely incurable. Standard approaches for treatment include surgery and combinations of radiotherapy and chemotherapy, but factors such as the highly selective blood- brain barrier have made treating GBM and other brain diseases extremely difficult. However, immunotherapy or “personalized medicine” integrated with chemotherapy or radiotherapy may become the future for targeting GBM tumors and other brain diseases. This review evaluates the mechanisms and efficacy of standard drugs such as temozolomide and bevacizumab, as well as novel advancements in the field, such as nano-mediated drug delivery systems (NDDS) and the rise of immunology as a basis for treating GBM.
Click here for the full article.