Constructing an Epistatic Network from Hepatitis C Viral Protein Sequence Data Using Algorithmic Methods

At the molecular level of life, epistasis refers to the phenomenon where the effect of one genetic mutation is modified or influenced by the presence of another mutation. This epistatic interaction between mutations can affect the resulting phenotype of a living organism. Understanding of epistatic interactions is essential for prediction of how multiple genetic mutations or variations in a virus may interact and influence disease susceptibility, treatment responses, and drug efficacy in patients. Hepatitis C viral genome possibly exhibits numerous epistatic interactions, underscoring the importance of our new computational approach, which joins combinations of specific linked viral mutations from different patients into an epistatic graph or network. We have developed an algorithmic method based on statistics and graph theory for constructing an epistatic network from hepatitis C viral protein sequence data. This approach involves several key steps: constructing a mutation binary matrix, calculating linked mutation pairs, and building an epistatic network. Additionally, we offer an optional step for identifying dense subgraphs within the network. As part of future work, we propose a hypothesis that the identification of dense subgraphs in the epistatic network may be indicative of combinations of specific mutations within the viral genome or emerging viral haplotypes. Those haplotypes can determine the virus's properties, such as its ability to infect hosts, evade the host immune system, or respond to antiviral treatments. The presence of combined specific mutations and their epistatic interactions can lead to the formation of distinct haplotypes. Our hypothesis warrants further investigation and research to explore potential associations between network density and the evolution of viral haplotypes.

Evaluating the performance of the ensemble sub-epidemic frameworks and other common statistical models in producing retrospective short-term mpox forecasts at the global and national levels.

Over the past year, an unprecedented surge in mpox cases affected multiple countries previously free of disease. While multi-model forecasts of the epidemic’s trajectory were critical in guiding the implementation of public health interventions and determining policy, there was little opportunity to assess forecasting performance and improve models amid the ongoing public health crisis. As the epidemic has declined, a retrospective evaluation of employed forecasting methodologies is vital to preparing for future public health events and advancing the growing field of epidemic forecasting. We utilized mpox case data from the Centers for Disease Control and Prevention (CDC) and Our World in Data (OWID) teams to generate (weeks of July 14th 2022 - January 26th, 2023) and evaluate retrospective sequential weekly forecasts (e.g., 1-4-weeks) for Brazil, Canada, France, Germany, Spain, the United Kingdom, the USA, and at the global scale using auto-regressive integrated moving average (ARIMA), a general additive model (GAM), simple linear regression (SLR), Facebook’s Prophet model, as well as the sub-epidemic wave (spatial-wave) and n-sub-epidemic modeling frameworks. The n-sub-epidemic framework, specifically the unweighted ensemble model, performed best across most forecasting horizons for a majority of locations regarding average mean squared error (MSE), mean absolute error (MAE), weighted interval score (WIS), and 95% prediction interval (95% PI) coverage. However, multiple models noted widespread success for 95% PI coverage. Compared to the ARIMA model, both sub-epidemic frameworks improved considerably in average MSE, MAE, and WIS and minimally (<10%) in average 95% PI coverage. Overall, the sub-epidemic frameworks performed well compared to other established modeling methodologies, highlighting their continued utility in producing short-term forecasts for epidemiologically different diseases.

Transcriptomic analyses of flavivirus-infected neurons from genetically susceptible and resistant mice

Resistance to flavivirus-induced neuropathology discovered in mice was shown to be inherited in a monogenic, dominant, autosomal manner. The gene associated with the resistant phenotype was subsequently found to be Oas1b. Resistant mice express the full-length protein while susceptible mice express a truncated protein due to a premature stop codon. A congenic resistant mouse strain was made by replacing the susceptible version of the Oas1b gene with the resistant one. Resistant mice infected intracranially with a lethal dose of a flavivirus, such as West Nile virus (WNV), show no disease signs, while susceptible mice develop lethal encephalitis. The mechanism by which Oas1b protects resistant mice is not known.

Cultures were established with cortical/hippocampal neurons obtained from 2 day old susceptible and resistant mice and infected with WNV NY99. Similar yields of virus were produced during the first 6 days of the infection from both types of cultures. However, the susceptible neurons died by 10 days post infection (dpi). The virus yields produced by the resistant neurons then progressively decreased until they were undetectable by 20 dpi. Cellular RNAs extracted at 1-, 3-, and 6 dpi from susceptible and resistant neurons, and at 18-, 20-, and 22 dpi from resistant neurons were sequenced by RNA-seq. The resulting short reads were mapped against the mouse genome using Bowtie and Tophat algorithms. Transcriptomic assembly and differential gene expression analyses were carried out using Cufflinks and Cuffdiff algorithms, respectively. Finally, gene enrichment analysis was carried out using DAVID.

We found that gene transcripts involved in the autophagy delivery pathway ERphagy (Retreg1, Rtn3, and Ccpg1), which is endoplasmic reticulum (ER) specific, were downregulated in susceptible neurons, but not in resistant neurons at early times after infection. Multiple steps of the flavivirus replication cycle occur on the ER and this result suggests that ERphagy is involved in controlling virus replication in resistant cells. Three additional cellular pathways: xenophagy (a type of selective autophagy that targets invading pathogens), immunoglobulin mediated immune response, and protection from natural killer cell mediated cytotoxicity were consistently upregulated in infected resistant neurons but not in susceptible ones. These data suggest that more than one biological process is involved in the inhibition of virus production by and continued survival of infected resistant neurons.

The study of Google Trends and Corresponding Associations and Relationship to COVID-19 Incidence and Mortality

Incidence and mortality trend prediction is useful for allocating public health resources and studying the course of the pandemic. One way to do that is to analyze search queries people use at a given time period. We analyzed a broad set of keywords related to COVID-19 incidence and mortality. We ran Granger Test and calculated Cross-correlation coefficients for those search queries against incidence and mortality time series to evaluate if there is a relationship between those time series and its magnitude. Most selected incidence-related queries correlated well with incidence, while mortality trends are less useful for mortality prediction.

Computational methods in chemistry applied to trimeric and dimeric water molecules compared with experimental ATR-FTIR

Guassian 16W computational chemistry software is used in tandem with DFT (Density Functional Theory) and the B3LYP hybrid model to explore the vibrational modes of both dimeric(1) and trimeric(2) configurations of water molecules. These computed IR spectra are compared to experiment, and various scale factors are applied to more closely match the experimental results. IEFPCM (Integral Equation Formalism Polarizable Continuum Model) is used to simulate solvation of the water dimers and trimers within an aqueous environment. The physical interpretation of the spectral scale factor is discussed.

Calculated Vibrational Frequencies for Reduced Plastoquinone in the A1 Binding Site of Photosystem I

In photosystem I (PSI) from higher plants, cyanobacteria, and algae, a phylloquinone molecule (PhQ; 2-methyl-3-phytyl-1,4 naphthoquinone) functions as the secondary electron acceptor in the so-called A1 binding site. Experimentally, other pigments could be incorporated into this protein binding site. Previously, we have incorporated plastoquinone (PQ; 2,3-dimethyl-5-prenyl-l,4-benzoquinone) into this protein binding site, and have produced flash-induced time-resolved FTIR difference spectra for PSI with PQ and PhQ incorporated.
To assess the experimentally obtained vibrational bands based on pigment-protein interactions, we developed a two-layer chemistry computational model called the ONIOM and performed vibrational frequency calculations. The ONIOM technique, uniquely, combines molecular mechanics and quantum mechanics in one calculation. The technique provides increasing accuracy while reducing calculation costs for a model consisting of several hundred atoms.
Here, we calculated the ONIOM vibrational frequencies of PhQ and PQ in the A1 binding site. The calculated spectra were used to build double difference spectra (DDS) by subtracting the spectra of PhQ from PQ. The calculated DDS was compared to the corresponding experimental DDS and the origins of the bands shifting upon the replacement of pigments in the protein were investigated.
Moreover, PQ is the native secondary electron acceptor in the QA binding site in photosystem II (PSII). In this protein environment, the PQ displays vibrational bands that are noticeably shifted in comparison to the bands of PQ in the A1 binding site of PSI. Based on our computational results, the pigment-protein interactions for PQ in the A1 and QA binding sites which may lead to different vibrational frequencies were also discussed.

Mathematical modeling of bubbles in flow streams and porous media

In many multi-phase chemical and electrochemical reaction systems, the fluid streams with dissolved gas or gas bubbles flow alongside a thin flat sheet of porous medium, composed of materials with varying surface energies. Chemical and electrochemical reactions occur at the surface and interface between the porous material and fluid. As such, the more surface is wetted, the more reaction can proceed, therefore, it is desirable to completely wet the porous material with the liquid phase.
However, the gas bubbles and their dynamics can reduce the surface area of the porous material in contact with the liquid phase. The goal of this work is to understand the dynamics of bubble interactions and dissolved gas within the porous material, in order to optimize processes and designs.

Analysis of Staphylococcus aureus biofilms with 3D image analysis software BiofilmQ

Staphylococcus aureus is a Gram-positive opportunistic pathogen that causes significant numbers of skin and sepsis infections worldwide. The ability to form biofilms is a key trait that allows many pathogens, including S. aureus, to form chronic infections. Biofilms are characterized by thick, dense aggregates of cells that are extremely resistant to antibiotics and the immune system. 4-Ethoxybenzoic acid (4EB) is a plant-derived compound that shows high anti-biofilm activity against S. aureus through an unknown mechanism. Flow cell methodology allows for the cultivation of stabilized biofilms that can be non-destructively imaged with Confocal Laser Scanning Microscopy (CLSM) and analyzed by the 3D image analysis software BiofilmQ. BiofilmQ is a MATLAB-written program developed to study biofilms and uses pseudo-cells to generate binary data from fluorescently dyed cells. BiofilmQ uses these pseudo-cells to compute diverse parameters that quantitatively characterize biofilms, including height, density, substrate area and surface roughness. 4EB treatment resulted in a statistically significant reduction in biofilm height and biovolume, both markers for the health of the biofilm. Surprisingly, many structural aspects of the biofilm like roughness were not significantly impacted. This suggests that 4EB can reduce biofilm robustness without impacting architecture. Current roadblocks in this research include maintaining biological relevance from binary data generated from fluorophores, selection of optimal graph and parameter types for comparison (E.g bin sizes, graphing method and interpretation), managing overexposed images and, ensuring proper quality control via filtering and segmentation. 4EB is among a class of chemicals with similar antibiofilm activity and discovering a novel mechanism could lead to new therapies which may subvert the rising antibiotic resistance in bacterial infections.

The Effects of Neuromodulation on the Propensity for Multistability of Bursting and Silent Regimes

Animals survival require expeditious control of rhythmic behaviors like locomotion, breathing, and heart beating. Central pattern generators (CPGs) are distinctive neuronal circuits which control motor patterns. In Medicinal leeches (Hirudo Sp.), heartbeat is controlled by a CPG that is composed of two pairs of mutually inhibitory leech heart interneurons (HNs) forming half-center oscillators (HNHCO). Endogenous neuromodulator myomodulin increases hyperpolarization-activated current (Ih) and decreases Na+/K+ pump (IpumpMax) current in these neurons. Coregulation of conductance of Ih (gh) and Ipumpmax leads to expansion of the functional bursting pattern. Synaptically isolated HN neuron exhibits expansion of seizure-like regime along the coregulation path. We investigated effects of the leak conductance on HN regimes of activity. By upregulating the leak conductance (gleak), the seizure-like pathological regime can be transformed into a high-spike frequency bursting regime and then into the usual low-spike frequency bursting regime. Thus, this upregulation could compensate the adverse seizurogenic effects of neuromodulation. We also investigate co-existence of bursting and silence. The propensity for multistability of these regimes, defined as the range of leak conductance supporting the coexistence of bursting and silent regimes, increases along with comodulation of Ih and IpumpMax current by myomodulin by a factor of 2.5. We conclude that neuromodulation notably increases propensity of multistability of silence and bursting, and the adverse effects of neuromodulation of a CPG, reducing Na+/K+ pump current, may be compensated by co-regulation of other currents such as hyperpolarization-activated current, leak current, and persistent Na+ current.

Analog Hopfield neural networks with three time delays

Purely excitatory and purely inhibitory Hopfield neural networks with three time delays were studied. Both types of neural networks demonstrated synchronous outputs. Stability threshold of the purely excitatory network did not depend on the time delays, however, it depended on the time delays for purely inhibitory network. An increase of the time delay in one or two inhibitory subnetworks resulted in multiresonances, period jumps, and period tripling bifurcations. In the neural networks that contained subnetworks with much smaller time delay, the interval until saturation of outputs dramatically decreases compared to the neural network with much larger and identical time delays.

Evolution of Visual Networks in Infancy During the Initial Six Months

The first six months of life represent a critical period for the development of the human brain's functional and structural foundation. This time frame is when early postnatal experiences can profoundly influence lifelong brain development. In this study, we examined the visual brain networks derived from the application of Independent Component Analysis (ICA) to fMRI data to investigate the spatial development of functional brain networks in 158 infants from birth to six months of age. Using ICA and spatial measurements, we explored how these networks change during this crucial early phase of life.
Our findings revealed that spatial similarity among visual brain networks significantly increased across age for all networks. However, intensity range significantly decreased for the visual networks, indicating a process of consolidation within these networks. Voxel intensity increased for visual networks, reflecting heightened activity patterns as infants matured. Voxel count and the weighted average distance from the center of mass increased, suggesting changes in network size and spatial distribution during the first six months of life.
Furthermore, we analyzed infants' fixation times on eyes, mouth, body, and objects during video clips. Our results demonstrated that infants increasingly fixated on eyes and mouths as they grew older, indicating a heightened interest in social interaction. In contrast, fixation on objects and the body decreased over time, highlighting changes in visual attention patterns.
This study provides valuable insights into the dynamic spatial development of brain networks during the first six months of life and its association with visual attention patterns. Understanding these early developmental processes is crucial for comprehending the roots of neurodevelopment disorders and enhancing our ability to support healthy brain development in infants.

Polyglot Entrainment for Higher-Dimensional Models

Entrainment is a fundamental phenomenon in the study of forced dynamical systems which occurs when the period of an intrinsic oscillator is synchronized to the period of an entraining stimuli and a stable phase relationship is maintained
between them. The modes of entrainment are represented in a V-shaped diagram called an Arnold tongue and such modes are termed as i:o, where “i” represents input cycles of external forcing and “o” denotes the output response of the underlying system. Multiple 1:1 entrainment (polyglot) has been recently explored only in two-dimensional slow-fast models in the vicinity of hopf bifurcations. In this research project, heading towards generality, we investigate the phenomenon of polyglot entrainment in the higher dimensional models including the Hodgkin-Huxley model and its reduced version, which are conductance-based mathematical models describing the initiation of action potentials in neurons. To explore the existence of polyglot in these models, dynamical system tools have been used to uncover the mechanism of entrainment and geometric structure of the null surfaces. The scientific merits of this research lie in its exploration of a previously unexplored facet of polyglot entrainment. This work not only expands the knowledge base of dynamical systems but also opens promising avenues for further research, with implications for understanding neural synchronization.

Biography
https://robinson.gsu.edu/profile/baozhong-yang/

Ultrafast generation of high harmonics in Quantum dots of Transition metal dichalcogenides

We theoretically study the generation of high harmonics in disk-shaped quantum dots of transition metal dichalcogenides (TMDC) placed in an ultrafast, ultrashort optical pulse. Parallelized algorithms are implemented on a high-performance cluster, to obtain electron dynamics of the quantum dots described within a massive Dirac-type effective model with infinite mass boundary conditions. After interaction with the pulse, the radiation spectra generated can be varied by changing either the radius of the dot, or the frequency and field amplitude of the pulse. The cutoff frequency and the intensities of low-order harmonics increase with the quantum dot radius. The harmonic cutoff frequency is also sensitive to the pulse intensity. The demonstration of high harmonic generation in the quantum dots of TMDC materials discussed holds potential in novel optoelectronic applications utilizing the nonlinear response of such finite nanoscale systems.

Mathematical modeling of deposition and erosion dynamics in a complex branching pore

Deposition and erosion are fundamental processes in fluid dynamics, and they play a crucial role in various natural phenomena and engineered systems. These processes involve the transport of particles by the fluid flow, resulting in erosion of materials from one location and their subsequent deposition at another. In this study, we propose a mathematical model to simulate the deposition and erosion processes occurring in a porous medium represented by an idealized structure composed of bifurcating cylindrical channels, featuring symmetric branching. The fluid flow within the channels is governed by the Stokes equations, while the transport, deposition and erosion of solid particles are described by an advection-diffusion equation. Furthermore, we investigated the effects of deposition and erosion processes on the evolution of the porous medium internal morphology.

Cake buildup and its influence on flow and transport within pleated membrane filters

Pleated membrane filters play a crucial role in various industrial applications due to their enhanced surface area-to-volume ratio compared to flat filters. This study presents a mathematical model for fouling mechanisms, with a particular focus on cake formation, and flow transport in pleated membrane filters. The proposed three-dimensional mathematical model considers the membrane pleated region as consisting of four distinct sub-regions: the support layer plus, the cake layer, the membrane layer, and the support layer minus. By incorporating Darcy’s law, the Stokes equation, and the advection-diffusion equation, we model the flow and transport within the pleated membrane regions as well as the filter cartridge. To overcome the complexity of the model and filter geometry, we employ asymptotic analysis techniques based on the small aspect ratios of the filter cartridge and pleated membrane. The results obtained from our study offer valuable insights into optimizing pleat packing density and filter performance while ensuring particle concentration remains below predefined thresholds.

A simplified mathematical model for cell proliferation in a tissue-engineering scaffold

Tissue engineering is a burgeoning multidisciplinary field that aims to regenerate damaged tissues and organs by cultivating artificial tissue outside the human body. Central to tissue engineering is the design and fabrication of scaffolds, which serve as templates for tissue growth and regeneration. Environmental factors such as flow rate, shear stress, and nutrient concentration within scaffold pores guide cell migration, proliferation, and differentiation. By understanding and manipulating cell- scaffold-environment interactions, tissue engineering holds the promise of providing innovative solutions for tissue replacement and regeneration in clinical therapies, with the ultimate goal of developing scaffolds that promote proper and functional cell behavior. This work presents a comprehensive continuum model for cell proliferation within tissue-engineering scaffolds. Through nondimensionalization and asymptotic analysis, the study aims to reduce computational burdens and solve mathematical models for optimal tissue growth, while considering the evolving scaffold porosity due to proliferation. Overall, this model provides valuable insights into the complex interactions driving tissue growth within scaffolds, offering a foundation for further research and advancements in tissue engineering and regenerative medicine.

CharGPT and Corporate Policies

This paper uses ChatGPT, a large language model, to extract managerial expectations of corporate policies from disclosures. We create a firm-level ChatGPT investment score, based on conference call texts, that measures managers’ anticipated changes in capital expenditures. We validate the ChatGPT investment score with interpretable textual content and its strong correlation with CFO survey responses. The investment score predicts future capital expenditure for up to nine quarters, controlling for Tobin’s q, other predictors, and fixed effects, implying the investment score provides incremental information about firms’ future investment opportunities. The investment score also separately forecasts future total, intangible, and R&D investments. High-investment-score firms experience significant negative future abnormal returns adjusted for factors, including the investment factor. We demonstrate ChatGPT’s applicability to measure other policies, such as dividends and employment. ChatGPT revolutionizes our comprehension of corporate policies, enabling the construction of managerial expectations cost-effectively for a large sample of firms over an extended period.

A data-driven approach to modeling cancer migration

In our research, we try to develop a data-driven model to identify the characteristics of migratory cancer cells vs non-migratory cancer cells. Our data set consists of close to 100,000 images extracted from the long videos taken at a 1-minute frame rate. We use dimensional reduction techniques to create a reduced representation that is easily manageable by the downstream analysis pipelines. Our goal is to develop a spatial temporal model which can be used to predict cell migration.
At the initial stage, we use autoencoders to obtain a few latent dimensions that can accurately capture the most information of the original data set. However, with the vanilla autoencoder (AE), we were not able to achieve the desired results. After training the network for 800 epochs, with a mean squared error loss function, we achieved a 10.61% test accuracy in a six-dimensional latent space. We are treating the latent dimension as a hyper-parameter here and working with 3,4,5,6,7 latent dimensions. Next, we incorporate Variational autoencoders (VAE). In theory, VAE should provide better results by learning the underlying probability distribution of the original dataset. We are seeing promising results. at 800 epoch, with a mean squared error loss function, we achieved a 42% test accuracy for VAE.

Multimodal Knowledge-infused Learning for Persuasive Marketing

Marketing is a human-centered process, as behavioral dynamics between marketers and consumers determine the success of a persuasive multimodal campaign. While online multimodal content provides an enhanced experience to the consumers, this richness of multimodality also brings the challenge of computational modeling, as the semantic contextual cues span across these modalities to make the meaning of the multimodal content. Identifying these contextual connections is crucial in retrieving the true holistic meaning. Large Language (LLM) and Vision models (LVM) show promising results on many multimodal tasks, successfully capturing holistic meaning with limited cross-modal semantic relationships. However, without explicit, common sense knowledge, Visual Language Models (VLM) only learn implicit representations, by capturing high-level patterns in vast corpora. In this project, we couple explicit external knowledge in the form of knowledge graphs with large VLMs to improve the performance of a downstream task, the classification of marketing campaigns for effectiveness. While the marketing application provides a compelling metric for assessing our methods, our computational approach enables the early detection of likely persuasive multi-modal campaigns and the assessment and augmentation of marketing theory.

Leveraging Machine Learning to Classify Applications using Wireless Network Traffic Traces

Prior work in network traffic monitoring has been limited to device or protocol identification, and there is still no mechanism or tool available to precisely identify what applications (carrying traffic in the network) are being executed. Identifying applications is important in networking, particularly for managing data traffic, security, and enabling smart systems. In this paper, we study and present the design of a technique
to identify applications from network traces by leveraging machine learning (ML). Framing the problem as an application classification problem, we set up our ML pipeline to learn key features from packet data and the behavior of the data over time. The feature generation, their training using traditional ML models, and the decision making are executed over a fourstage pipeline, to yield the name of the application. Through an in-lab environment experimentation using OpenWrt toolkit, RaspberryPi, and a set of physical devices (generating network traffic), we evaluated on average about 204K data points from the captured network packet traces for six applications. Our results show that our method is able to classify the applications with at least 90% accuracy. Through micro-benchmarking, we also show the feasibility of scaling the number of applications and running the tool in real-time.

Dr. Pejman Sanaei is currently an Assistant Professor in the Department of Mathematics and Statistics at Georgia State University. In 2019-2022, he was an Assistant Professor in the Department of Mathematics at New York Institute of Technology. Prior to that, he was (2017-2019) a Courant instructor at the Courant Institute of Mathematical Sciences (CIMS), New York University and working on mathematical modeling and simulation in the Applied Math Lab and Research and Training Group in Mathematical Modeling and Simulation at CIMS. Pejman did his PhD (August 2017), in applied mathematics at the Department of Mathematical Sciences at New Jersey Institute of Technology at Complex Flows and Soft Matter Group. He has an undergraduate degree in mechanical engineering (fluid mechanics) and a masters degree in pure mathematics, both completed before he began his PhD studies at NJIT. During his PhD studies he also completed a master's degree in applied statistics, awarded in December 2016. He has expertise in deterministic and stochastic approaches to mathematical modeling, fluid mechanics, fluid structure interaction, tissue engineering, applications of mathematics to industry and in particular his dissertation topic is mathematical modeling of membrane filtration.

Abstact

We study numerically the interaction of ultrashort pulses with low dimensional systems, such as graphene, germanene, transition-metal dichalcogenides, Weyls’ semimetals, and flakes of such materials, i.e., quantum dots. Electron dynamics in such ultrastrong pulses is highly nonlinear, which results in such effects as generation of high harmonics, nonlinear charge transfer through the system, and residual excitations of the electron systems. The graphene-like materials also have nontrivial topological properties, which can be probed by ultrashort optical pulses with high intensity. In such strong pulses, the electrons are transferred in the reciprocal space over a relatively large distance and correspondingly the electrons can accumulate, during the pulse, relatively large topological phase, . Such an accumulation of the topological phase results in unique phenomena, which can be observed in topologically nontrivial materials, and which are governed by an effect of topological resonance that is a cancellation of the topological phase by the dynamic phase. The topological resonance results, for example, in the fundamentally fastest induction of a significant population and valley polarization in a monolayer of a transition metal dichalcogenide (i.e., MoS2 and WS2). This effect may be extended to other two-dimensional materials with the same symmetry. Thus, generated valley polarization can be written and read out by a pulse consisting of just a single optical oscillation with a duration of a few femtoseconds and an amplitude of ∼0.25 V/A.

For graphene interacting with a few-fs intense optical pulse, there is unique and rich behavior dramatically different from three-dimensional solids. Quantum electron dynamics is shown to be coherent but highly nonadiabatic and effectively irreversible due to strong dephasing. Electron distribution in reciprocal space exhibits hot spots at the Dirac points and oscillations whose period is determined by nonlocality of electron response and whose number is proportional to the field amplitude.

The strong ultrafast circularly polarized optical pulses can also be used to study an interferometry in graphene's reciprocal space without a magnetic field. The reciprocal space interferograms contain information on the electronic spectra and topological properties of graphene and on the waveform and circular polarization of the excitation optical pulses. These can be measured using angle-resolved photoemission spectroscopy with attosecond ultraviolet pulses.

While in quantum dots, i.e., flakes of two dimensional topological materials, the topology is not defined, the nonlinear electron dynamics in such zero dimensional systems results in generation of high optical harmonics.  The generation of high harmonics is strongly sensitive to the electron dephasing time and to the size and the shape of a quantum dot.

Biography

Vadym Apalkov is a professor of physics and astronomy at Georgia State University. He received his masters degree in physics from the Department of General and Applied Physics at Moscow Institute of Physics and Technology in 1991. He received his PhD from the University of Utah, Salt Lake City, Utah in 1995. His PhD research area was the Fractional Quantum Hall Effect. After receiving his PhD, he worked as a research scientist at Kharkov Institute of Physics and Technology. Then he was a postdoctoral research associate at the University of Utah till 2004. From 2004, he is a faculty member at the Department of Physics and Astronomy at Georgia State University. His research interests are transport and optical properties of low dimensional systems, such as graphene-like materials, quantum dots, Weyl semimetals, and others. He is also interested in spintronics and valleytronics in applications to information storage. Part of his research is also related to plasmonics and its applications.

Ensemble method of polyp segmentation

It is important to find the polyps in a human system that helps to prevent cancer in medical diagnosis. This poster talks about the approach of ensemble U-Net based convolutional neural network that is used for polyp segmentation. This U-Net architecture has three encoders: Resnet- 50, Resnet-101 along with Efficient- NetB4. They were tre-prained in COCO dataset and transfer learning technique was used along with tversky loss. Different data augmentation techniques like rotations, flips, scaling, contrast along with varying learning rates and optimization were deployed to make a better ensemble method. We proposed a new layer that calculates the average of intermediate masks followed by the sigmoid layer. In our experiment, the performance of the proposed architecture was better than existing methods over the Kvasir-SEG dataset with accuracy 0.9692, recall 0.8872, precision 0.912, dice coefficient 0.8949, and intersection over union 0.987

Estimation of the Youden Index of a Continuous Diagnostic Test with Verification Biased Data

Youden index is a comprehensive measurement of the effectiveness of a diagnostic test. For a continuous-scale diagnostic test, its maximum diagnostic ability can be obtained by maximizing the Youden index over all possible values of the cutoff point. However, in medical diagnostic studies, verification of the true disease status might be partially missing and estimators based on partially validated subjects are usually biased. Bias-corrected estimators for the area under the ROC curve (AUC) have been developed; however, no verification bias-corrected estimators are explicitly developed for the Youden index. In this paper, various verification bias-corrected estimators of the Youden index and associated optimal cutoff point for a continuous test based on imputation (FI and MSI), reweighting (IPW) and semiparametric efficient (SPE) approaches are proposed and investigated under the assumption of missing at random (MAR) for test results. We justify that the SPE method, which is previously recommended in the AUC estimation literature, can be improved when the disease prevalence is either high or low. Simulation results show in these very common cases, our proposed hybrid estimators combining SPE approach with appropriate imputation-based approaches are generally robust and superior to the regular SPE estimator, even when the disease model is misspecified. Conclusions are also supported by observations from a real data study, in which case the underlying disease mechanism is unknown.

Simulation of cell proliferation in a tissue-engineering scaffold pore

Examining the interplay of various factors on tissue growth within a tissue-engineering scaffold channel is crucial for optimizing cell proliferation. This study delves into the combined effects of nutrient flow rate, nutrient consumption, scaffold elasticity, and cell properties. A novel mathematical model is developed to describe the dynamics of nutrient flow, concentration, scaffold elasticity, and cell proliferation. Subsequently, the model is solved and employed to simulate the cell proliferation process. The ultimate aim is to optimize the initial configuration of scaffold channels to maximize cell growth. Our findings reveal that the rate of nutrient consumption by cells, referred to as the cell hunger rate, significantly impacts tissue growth, resulting in longer incubation periods for higher cell hunger rates. Additionally, the compliance of the scaffold material slightly affects overall growth. Notably, by reducing scaffold elasticity while maintaining a constant nutrient consumption rate, an optimal funnel-shaped channel geometry emerges. This geometry, with a larger upper part compared to the narrower channel downstream, promotes improved tissue integration and functionality.

An Efficient Algorithm for Collision Avoidance Between a Solar Array Satellite and Space Debris1

Half of the risk to any satellite is from debris collision. The main body of the satellite, housing the main electronics is encapsulated by bulletproof outer layers but most satellites include solar panels as the only energy source and they cannot be covered with multiple kevlar layers or any safety material. As space junk keeps on increasing, we seek to mitigate the tragedies related to it. Every collision in turn creates many new space junk particles which drives a positive feedback chain reaction, which could ultimately lead to a phenomenon known as “Kessler Syndrome”[1], which can render whole space unusable altogether. Several private companies like “LEO-Space” and government agencies are working to help solve this issue, yet some countries perform anti-satellite operations for military purposes each of which creates more than tens of thousands of pieces greater than 0.5 centimetres (that cannot be stopped by layers of protective material) traveling at relative speeds of up to 12km/sec on an average(which usually stay in their orbits for more than 100 years, depending on their altitudes and orbit). China (in 2007), the USA (in 2009), India (in 2019), and Russia (in 2021) have performed these so-called “tests” in the orbits of the altitude of the international space station creating countless debris of various sizes that would stay as a threat in most used orbit i.e. LEO(roughly 160km to 2000km above the earth’s surface).

Samer Gozem is an Associate Professor of Chemistry and Associate Director of Graduate Studies at Georgia State University. He obtained his B.Sc. in Chemistry in 2008 from the American University of Beirut in Lebanon and his Ph.D. in Photochemical Sciences in 2013 at Bowling Green State University. He then carried out his postdoctoral training at the University of Southern California before joining Georgia State University as a faculty member in 2017. His research interests include using classical and quantum mechanics to study light-responsive chemical and biological systems.

Xiaolin Hu is a Professor of Computer Science at Georgia State University and heads the Systems Integrated Modeling and Simulation (SIMS) Lab. He obtained his Ph.D. in Computer Engineering from the University of Arizona in 2004. His research interests include modeling and simulation theory and application, dynamic data driven simulation, agent and multi-agent systems, and complex systems science. He has numerous publications in leading scientific journals and conferences, and organized many international conferences and symposiums in the field of modeling and simulation. He was a National Science Foundation (NSF) CAREER Award recipient.

Moderated by Dr. Suranga Edirisinghe | ARCTIC at Georgia State University

Konstantin Cvetanov | Sr. Solution Architect – Enterprise AI Software
Konstantin Cvetanov is a Senior Solution Architect at NVIDIA focused on advancing the adoption of Artificial Intelligence software platforms in global Enterprise, Research, and Public Sector organizations. Before his 6 years at NVIDIA Konstantin spent much of his career in consulting roles leading multi-disciplinary Professional Services and Engineering teams at prominent Systems Integrators across major industries. During his career, Konstantin has published many blogs and manuals including a highly rated technical deployment book on Datacenter Virtualization and End User Computing featured on Amazon. Over the course of his career, Konstantin has cultivated a deep passion for youth career mentoring and loves fostering relationships with academic researchers and startup communities.

Brock Davis | Principal Engineering Manager – Microsoft
Brock Davis is a Principal Engineering Manager at Microsoft within the XBOX organization leading Gameplay Systems for Minecraft which focuses on everything from graphics rendering to using AI for pathfinding and Mob behaviors. Previously he has worked at Amazon Web Services, The Walt Disney Company, WarnerBros Discovery, and Georgia State University. While at Amazon, he led the Commercial Linux team for EC2 which was responsible for the underlying Linux images for AWS. At The Walt Disney Company, he focused on the cloud infrastructure for all of Disney Streaming products such as Disney+ and Hulu. At WarnerBros Discovery, he led efforts around personalization experiences and data privacy, as well as app development for Bleacher Report and CNN. At Georgia State University, he led the Research Solutions team that focused on providing solutions for funded research ranging from custom app development to compute. Through all of his roles he has gained extensive industry knowledge on app development, cloud computing, machine learning, and artificial intelligence.

Dr. Jennifer Esposito | Chair of Educational Policy Studies, Georgia State University
Jennifer Esposito is a department chair of Educational Policy Studies and a Distinguished University Professor at Georgia State University. Her research takes an intersectional approach to educational research, centering race and gender. As an interdisciplinary scholar, she sees critical theories as tools to interrogate social life and solve problems related to the material consequences of oppression and privilege.

Scott Kent | Principal of the FCS Innovation Academy

Dr. Renata Rawlings-Goss | Executive Director - The South Big Data Innovation Hub
Dr. Renata Rawlings-Goss is the Executive Director of The South Big Data Innovation Hub, an NSF-funded 16 state center connecting industry, academia, and government around data science innovation. She is the Director of Strategic Partnerships for the Georgia Institute of Technology- Institute for Data Engineering and Science (IDEaS), and also the founder of Good with Data LLC, which runs The Data Career Academy - for professionals and faculty looking to accelerate their careers with data.Formerly, Dr. Rawlings-Goss worked with the White House Office of Science and Technology Policy founding the National Data Science Organizers and co-leading the writing team for the Federal Big Data Strategic Plan. Through her roles, she has served as an executive strategist, career mentor, and policy advisor to Fortune 500 companies, individuals, as well as over 19 federal and state government agencies around data science education, Big Data, Digital Transformation, Public-Private Partnerships, Artificial Intelligence (AI), Internet of Things (IoT), Machine Learning, Data Career Success, Professional Development and Data Innovation. Dr. Renata Rawlings-Goss is a biophysicist by training and a nationally recognized leader in Data Science.Author of “Data Careers, Training, and Hiring” published in 2019 by SpringerPress, her work across fields has been recognized in the Washington Post, the Wall Street Journal, and as one of President Obama’s Top 100 Impacts in Science and Technology.

Biography

#1 Evaluating the Performance of Machine Learning Methods for Predicting Mortality in Intensive Care Unit Patients
Alexander Ahn

Machine learning methods are increasingly being used for building diagnostic models in clinical settings to identify patients who are at a higher risk of mortality. Recent studies have shown that ensemble tree-based learning methods, provide an alternative non-parametric approach compared to traditional methods for building predictive models in high-dimensional datasets. In this study, we evaluated the performance of logistic regression, random forest, XGBoost, and LGBM (leaf-wise tree-based learning algorithm) for identifying ICU patients with a 28-day mortality risk at the time of hospital admission. The case study data originates from a subset of publicly available data from the Medical Information Mart for Intensive Care (MIMIC) II database. The performance of different methods was evaluated using prediction error curves. The results show that the XGBoost classification method achieved the best prediction accuracy for classifying survivors vs. non-survivors with (cross-validation area under the curve; AUC=0.86). The top features for predicting death at the time of ICU admission included age, simplified acute physiology score (SAPS), and serum sodium levels at admission. These results can help predict which patients are likely to die within 28 days of ICU admission so that healthcare professionals can design &amp; implement optimal treatment strategies to improve patient outcomes. All analyses were conducted using the AutoAI tool in IBM Watson Studio.

#7 Cake buildup and its influence on flow and transport within pleated membrane filters
Sima Moshafi, Pejman Sanaei, Daniel Fong

Pleated membrane filters play a crucial role in various industrial applications due to their enhanced surface area-to-volume ratio compared to flat filters. This study presents a mathematical model for fouling mechanisms, with a particular focus on cake formation, and flow transport in pleated membrane filters. The proposed three-dimensional mathematical model considers the membrane pleated region as consisting of four distinct sub-regions: the support layer plus, the cake layer, the membrane layer, and the support layer minus. By incorporating Darcy’s law, the Stokes equation, and the advection-diffusion equation, we model the flow and transport within the pleated membrane regions as well as the filter cartridge. To overcome the complexity of the model and filter geometry, we employ asymptotic analysis techniques based on the small aspect ratios of the filter cartridge and pleated membrane. The results obtained from our study offer valuable insights into optimizing pleat packing density and filter performance while ensuring particle concentration remains below predefined thresholds.

#8 Computational methods in chemistry applied to trimeric and dimeric water molecules compared with experimental ATR-FTIR
Michael Nelson

Guassian 16W computational chemistry software is used in tandem with DFT (Density Functional Theory) and the B3LYP hybrid model to explore the vibrational modes of both dimeric(1) and trimeric(2) configurations of water molecules. These computed IR spectra are compared to experiment, and various scale factors are applied to more closely match the experimental results. IEFPCM (Integral Equation Formalism Polarizable Continuum Model) is used to simulate solvation of the water dimers and trimers within an aqueous environment. The physical interpretation of the spectral scale factor is discussed.

#9 A simplified mathematical model for cell proliferation in a tissue-engineering scaffold
Mona James, Amy Sims, Paul Joseph, Sai Kunnatha, Ashok Joseph, Haniyeh Fattahpour, Pejman Sanaei

Tissue engineering is a burgeoning multidisciplinary field that aims to regenerate damaged tissues and organs by cultivating artificial tissue outside the human body. Central to tissue engineering is the design and fabrication of scaffolds, which serve as templates for tissue growth and regeneration. Environmental factors such as flow rate, shear stress, and nutrient concentration within scaffold pores guide cell migration, proliferation, and differentiation. By understanding and manipulating cell- scaffold-environment interactions, tissue engineering holds the promise of providing innovative solutions for tissue replacement and regeneration in clinical therapies, with the ultimate goal of developing scaffolds that promote proper and functional cell behavior. This work presents a comprehensive continuum model for cell proliferation within tissue-engineering scaffolds. Through nondimensionalization and asymptotic analysis, the study aims to reduce computational burdens and solve mathematical models for optimal tissue growth, while considering the evolving scaffold porosity due to proliferation. Overall, this model provides valuable insights into the complex interactions driving tissue growth within scaffolds, offering a foundation for further research and advancements in tissue engineering and regenerative medicine.

#10 Calculated Vibrational Frequencies for Reduced Plastoquinone in the A1 Binding Site of Photosystem I
Leyla Rohani, Michael Nelson, Gary Hastings

n photosystem I (PSI) from higher plants, cyanobacteria, and algae, a phylloquinone molecule (PhQ; 2-methyl-3-phytyl-1,4 naphthoquinone) functions as the secondary electron acceptor in the so-called A1 binding site. Experimentally, other pigments could be incorporated into this protein binding site. Previously, we have incorporated plastoquinone (PQ; 2,3-dimethyl-5-prenyl-l,4-benzoquinone) into this protein binding site, and have produced flash-induced time-resolved FTIR difference spectra for PSI with PQ and PhQ incorporated. To assess the experimentally obtained vibrational bands based on pigment-protein interactions, we developed a two-layer chemistry computational model called the ONIOM and performed vibrational frequency calculations. The ONIOM technique, uniquely, combines molecular mechanics and quantum mechanics in one calculation. The technique provides increasing accuracy while reducing calculation costs for a model consisting of several hundred atoms. Here, we calculated the ONIOM vibrational frequencies of PhQ and PQ in the A1 binding site. The calculated spectra were used to build double difference spectra (DDS) by subtracting the spectra of PhQ from PQ. The calculated DDS was compared to the corresponding experimental DDS and the origins of the bands shifting upon the replacement of pigments in the protein were investigated. Moreover, PQ is the native secondary electron acceptor in the QA binding site in photosystem II (PSII). In this protein environment, the PQ displays vibrational bands that are noticeably shifted in comparison to the bands of PQ in the A1 binding site of PSI. Based on our computational results, the pigment-protein interactions for PQ in the A1 and QA binding sites which may lead to different vibrational frequencies were also discussed.

#11 Evaluating the performance of the ensemble sub-epidemic frameworks and other common statistical models in producing retrospective short-term mpox forecasts at the global and national levels
Amanda Bleichrodt, Ruiyan Luo, Alexander Kirpich, Gerardo Chowell

Over the past year, an unprecedented surge in mpox cases affected multiple countries previously free of disease. While multi-model forecasts of the epidemic’s trajectory were critical in guiding the implementation of public health interventions and determining policy, there was little opportunity to assess forecasting performance and improve models amid the ongoing public health crisis. As the epidemic has declined, a retrospective evaluation of employed forecasting methodologies is vital to preparing for future public health events and advancing the growing field of epidemic forecasting. We utilized mpox case data from the Centers for Disease Control and Prevention (CDC) and Our World in Data (OWID) teams to generate (weeks of July 14th 2022 - January 26th, 2023) and evaluate retrospective sequential weekly forecasts (e.g., 1-4-weeks) for Brazil, Canada, France, Germany, Spain, the United Kingdom, the USA, and at the global scale using auto-regressive integrated moving average (ARIMA), a general additive model (GAM), simple linear regression (SLR), Facebook’s Prophet model, as well as the sub-epidemic wave (spatial-wave) and n-sub-epidemic modeling frameworks. The n-sub-epidemic framework, specifically the unweighted ensemble model, performed best across most forecasting horizons for a majority of locations regarding average mean squared error (MSE), mean absolute error (MAE), weighted interval score (WIS), and 95% prediction interval (95% PI) coverage. However, multiple models noted widespread success for 95% PI coverage. Compared to the ARIMA model, both sub-epidemic frameworks improved considerably in average MSE, MAE, and WIS and minimally (<10%) in average 95% PI coverage. Overall, the sub-epidemic frameworks performed well compared to other established modeling methodologies, highlighting their continued utility in producing short-term forecasts for epidemiologically different diseases.

#12 Estimation of the Youden Index of a Continuous Diagnostic Test with Verification Biased Data
Shirui Wang, Shuangfei Shi, Gengsheng Qin

Youden index is a comprehensive measurement of the effectiveness of a diagnostic test. For a continuous-scale diagnostic test, its maximum diagnostic ability can be obtained by maximizing the Youden index over all possible values of the cutoff point. However, in medical diagnostic studies, verification of the true disease status might be partially missing and estimators based on partially validated subjects are usually biased. Bias-corrected estimators for the area under the ROC curve (AUC) have been developed; however, no verification bias-corrected estimators are explicitly developed for the Youden index. In this paper, various verification bias-corrected estimators of the Youden index and associated optimal cutoff point for a continuous test based on imputation (FI and MSI), reweighting (IPW) and semiparametric efficient (SPE) approaches are proposed and investigated under the assumption of missing at random (MAR) for test results. We justify that the SPE method, which is previously recommended in the AUC estimation literature, can be improved when the disease prevalence is either high or low. Simulation results show in these very common cases, our proposed hybrid estimators combining SPE approach with appropriate imputation-based approaches are generally robust and superior to the regular SPE estimator, even when the disease model is misspecified. Conclusions are also supported by observations from a real data study, in which case the underlying disease mechanism is unknown.

#13 Analog Hopfield neural networks with three time delays
Vladimir E. Bondarenko

Purely excitatory and purely inhibitory Hopfield neural networks with three time delays were studied. Both types of neural networks demonstrated synchronous outputs. Stability threshold of the purely excitatory network did not depend on the time delays, however, it depended on the time delays for purely inhibitory network. An increase of the time delay in one or two inhibitory subnetworks resulted in multiresonances, period jumps, and period tripling bifurcations. In the neural networks that contained subnetworks with much smaller time delay, the interval until saturation of outputs dramatically decreases compared to the neural network with much larger and identical time delays.

#14 Mathematical modeling of deposition and erosion dynamics in a complex branching pore
Emeka Peter Mazi, Pejman Sanaei

Deposition and erosion are fundamental processes in fluid dynamics, and they play a crucial role in various natural phenomena and engineered systems. These processes involve the transport of particles by the fluid flow, resulting in erosion of materials from one location and their subsequent deposition at another. In this study, we propose a mathematical model to simulate the deposition and erosion processes occurring in a porous medium represented by an idealized structure composed of bifurcating cylindrical channels, featuring symmetric branching. The fluid flow within the channels is governed by the Stokes equations, while the transport, deposition and erosion of solid particles are described by an advection-diffusion equation. Furthermore, we investigated the effects of deposition and erosion processes on the evolution of the porous medium internal morphology.

#15 Simulation of cell proliferation in a tissue-engineering scaffold pore
Haniyeh Fattahpour, Pejman Sanaei

Examining the interplay of various factors on tissue growth within a tissue-engineering scaffold channel is crucial for optimizing cell proliferation. This study delves into the combined effects of nutrient flow rate, nutrient consumption, scaffold elasticity, and cell properties. A novel mathematical model is developed to describe the dynamics of nutrient flow, concentration, scaffold elasticity, and cell proliferation. Subsequently, the model is solved and employed to simulate the cell proliferation process. The ultimate aim is to optimize the initial configuration of scaffold channels to maximize cell growth. Our findings reveal that the rate of nutrient consumption by cells, referred to as the cell hunger rate, significantly impacts tissue growth, resulting in longer incubation periods for higher cell hunger rates. Additionally, the compliance of the scaffold material slightly affects overall growth. Notably, by reducing scaffold elasticity while maintaining a constant nutrient consumption rate, an optimal funnel-shaped channel geometry emerges. This geometry, with a larger upper part compared to the narrower channel downstream, promotes improved tissue integration and functionality.

#16 Polyglot Entrainment for Higher-Dimensional Models
Lawan Wijayasooriya, Emel Khan, Pejman Sanaei

Entrainment is a fundamental phenomenon in the study of forced dynamical systems which occurs when the period of an intrinsic oscillator is synchronized to the period of an entraining stimuli and a stable phase relationship is maintained between them. The modes of entrainment are represented in a V-shaped diagram called an Arnold tongue and such modes are termed as i:o, where “i” represents input cycles of external forcing and “o” denotes the output response of the underlying system. Multiple 1:1 entrainment (polyglot) has been recently explored only in two-dimensional slow-fast models in the vicinity of hopf bifurcations. In this research project, heading towards generality, we investigate the phenomenon of polyglot entrainment in the higher dimensional models including the Hodgkin-Huxley model and its reduced version, which are conductance-based mathematical models describing the initiation of action potentials in neurons. To explore the existence of polyglot in these models, dynamical system tools have been used to uncover the mechanism of entrainment and geometric structure of the null surfaces. The scientific merits of this research lie in its exploration of a previously unexplored facet of polyglot entrainment. This work not only expands the knowledge base of dynamical systems but also opens promising avenues for further research, with implications for understanding neural synchronization.

#17 ChatGPT and Corporate Policies
Manish Jha, Jialin Qian, Michael Weber, Baozhong Yang

This paper uses ChatGPT, a large language model, to extract managerial expectations of corporate policies from disclosures. We create a firm-level ChatGPT investment score, based on conference call texts, that measures managers’ anticipated changes in capital expenditures. We validate the ChatGPT investment score with interpretable textual content and its strong correlation with CFO survey responses. The investment score predicts future capital expenditure for up to nine quarters, controlling for Tobin’s q, other predictors, and fixed effects, implying the investment score provides incremental information about firms’ future investment opportunities. The investment score also separately forecasts future total, intangible, and R&D investments. High-investment-score firms experience significant negative future abnormal returns adjusted for factors, including the investment factor. We demonstrate ChatGPT’s applicability to measure other policies, such as dividends and employment. ChatGPT revolutionizes our comprehension of corporate policies, enabling the construction of managerial expectations cost-effectively for a large sample of firms over an extended period.

#18 Mathematical modeling of bubbles in flow streams and porous media
Hamed Karami, Pejman Sanaei

In many multi-phase chemical and electrochemical reaction systems, the fluid streams with dissolved gas or gas bubbles flow alongside a thin flat sheet of porous medium, composed of materials with varying surface energies. Chemical and electrochemical reactions occur at the surface and interface between the porous material and fluid. As such, the more surface is wetted, the more reaction can proceed, therefore, it is desirable to completely wet the porous material with the liquid phase. However, the gas bubbles and their dynamics can reduce the surface area of the porous material in contact with the liquid phase. The goal of this work is to understand the dynamics of bubble interactions and dissolved gas within the porous material, in order to optimize processes and designs.

#19 The study of Google Trends and Corresponding Associations and Relationship to COVID-19 Incidence and Mortality
Aleksandr Shishkin, Alexander Kirpich

Incidence and mortality trend prediction is useful for allocating public health resources and studying the course of the pandemic. One way to do that is to analyze search queries people use at a given time period. We analyzed a broad set of keywords related to COVID-19 incidence and mortality. We ran Granger Test and calculated Cross-correlation coefficients for those search queries against incidence and mortality time series to evaluate if there is a relationship between those time series and its magnitude. Most selected incidence-related queries correlated well with incidence, while mortality trends are less useful for mortality prediction.

#20 An Efficient Algorithm for Collision Avoidance Between a Solar Array Satellite and Space Debris1
Varun Ahlawat

Half of the risk to any satellite is from debris collision. The main body of the satellite, housing the main electronics is encapsulated by bulletproof outer layers but most satellites include solar panels as the only energy source and they cannot be covered with multiple kevlar layers or any safety material. As space junk keeps on increasing, we seek to mitigate the tragedies related to it. Every collision in turn creates many new space junk particles which drives a positive feedback chain reaction, which could ultimately lead to a phenomenon known as “Kessler Syndrome”[1], which can render whole space unusable altogether. Several private companies like “LEO-Space” and government agencies are working to help solve this issue, yet some countries perform anti-satellite operations for military purposes each of which creates more than tens of thousands of pieces greater than 0.5 centimetres (that cannot be stopped by layers of protective material) traveling at relative speeds of up to 12km/sec on an average(which usually stay in their orbits for more than 100 years, depending on their altitudes and orbit). China (in 2007), the USA (in 2009), India (in 2019), and Russia (in 2021) have performed these so-called “tests” in the orbits of the altitude of the international space station creating countless debris of various sizes that would stay as a threat in most used orbit i.e. LEO(roughly 160km to 2000km above the earth’s surface).

#21 Multimodal Knowledge-infused Learning for Persuasive Marketing
Trilok Padhi, Ugur Kursuncu, Yaman Kumar, Valerie Shalin, Lane Peterson

Marketing is a human-centered process, as behavioral dynamics between marketers and consumers determine the success of a persuasive multimodal campaign. While online multimodal content provides an enhanced experience to the consumers, this richness of multimodality also brings the challenge of computational modeling, as the semantic contextual cues span across these modalities to make the meaning of the multimodal content. Identifying these contextual connections is crucial in retrieving the true holistic meaning. Large Language (LLM) and Vision models (LVM) show promising results on many multimodal tasks, successfully capturing holistic meaning with limited cross-modal semantic relationships. However, without explicit, common sense knowledge, Visual Language Models (VLM) only learn implicit representations, by capturing high-level patterns in vast corpora. In this project, we couple explicit external knowledge in the form of knowledge graphs with large VLMs to improve the performance of a downstream task, the classification of marketing campaigns for effectiveness. While the marketing application provides a compelling metric for assessing our methods, our computational approach enables the early detection of likely persuasive multi-modal campaigns and the assessment and augmentation of marketing theory.

#22 Leveraging Machine Learning to Classify Applications using Wireless Network Traffic Traces
Alireza Marefat, Abbaas Alif Mohamed Nishar, Ashwin Ashok

Prior work in network traffic monitoring has been limited to device or protocol identification, and there is still no mechanism or tool available to precisely identify what applications (carrying traffic in the network) are being executed. Identifying applications is important in networking, particularly for managing data traffic, security, and enabling smart systems. In this paper, we study and present the design of a technique to identify applications from network traces by leveraging machine learning (ML). Framing the problem as an application classification problem, we set up our ML pipeline to learn key features from packet data and the behavior of the data over time. The feature generation, their training using traditional ML models, and the decision making are executed over a fourstage pipeline, to yield the name of the application. Through an in-lab environment experimentation using OpenWrt toolkit, RaspberryPi, and a set of physical devices (generating network traffic), we evaluated on average about 204K data points from the captured network packet traces for six applications. Our results show that our method is able to classify the applications with at least 90% accuracy. Through micro-benchmarking, we also show the feasibility of scaling the number of applications and running the tool in real-time.

#23 Ultrafast generation of high harmonics in Quantum dots of Transition metal dichalcogenides
Aranyo Mitra, Ahmal Jawad Zafar, Vadym Apalkov

We theoretically study the generation of high harmonics in disk-shaped quantum dots of transition metal dichalcogenides (TMDC) placed in an ultrafast, ultrashort optical pulse. Parallelized algorithms are implemented on a high-performance cluster, to obtain electron dynamics of the quantum dots described within a massive Dirac-type effective model with infinite mass boundary conditions. After interaction with the pulse, the radiation spectra generated can be varied by changing either the radius of the dot, or the frequency and field amplitude of the pulse. The cutoff frequency and the intensities of low-order harmonics increase with the quantum dot radius. The harmonic cutoff frequency is also sensitive to the pulse intensity. The demonstration of high harmonic generation in the quantum dots of TMDC materials discussed holds potential in novel optoelectronic applications utilizing the nonlinear response of such finite nanoscale systems.

#24 Transcriptomic analyses of flavivirus-infected neurons from genetically susceptible and resistant mice
Emilio E. Espinola, Komal Arora, Margo Brinton

Resistance to flavivirus-induced neuropathology discovered in mice was shown to be inherited in a monogenic, dominant, autosomal manner. The gene associated with the resistant phenotype was subsequently found to be Oas1b. Resistant mice express the full-length protein while susceptible mice express a truncated protein due to a premature stop codon. A congenic resistant mouse strain was made by replacing the susceptible version of the Oas1b gene with the resistant one. Resistant mice infected intracranially with a lethal dose of a flavivirus, such as West Nile virus (WNV), show no disease signs, while susceptible mice develop lethal encephalitis. The mechanism by which Oas1b protects resistant mice is not known.

Cultures were established with cortical/hippocampal neurons obtained from 2 day old susceptible and resistant mice and infected with WNV NY99. Similar yields of virus were produced during the first 6 days of the infection from both types of cultures. However, the susceptible neurons died by 10 days post infection (dpi). The virus yields produced by the resistant neurons then progressively decreased until they were undetectable by 20 dpi. Cellular RNAs extracted at 1-, 3-, and 6 dpi from susceptible and resistant neurons, and at 18-, 20-, and 22 dpi from resistant neurons were sequenced by RNA-seq. The resulting short reads were mapped against the mouse genome using Bowtie and Tophat algorithms. Transcriptomic assembly and differential gene expression analyses were carried out using Cufflinks and Cuffdiff algorithms, respectively. Finally, gene enrichment analysis was carried out using DAVID.

We found that gene transcripts involved in the autophagy delivery pathway ERphagy (Retreg1, Rtn3, and Ccpg1), which is endoplasmic reticulum (ER) specific, were downregulated in susceptible neurons, but not in resistant neurons at early times after infection. Multiple steps of the flavivirus replication cycle occur on the ER and this result suggests that ERphagy is involved in controlling virus replication in resistant cells. Three additional cellular pathways: xenophagy (a type of selective autophagy that targets invading pathogens), immunoglobulin mediated immune response, and protection from natural killer cell mediated cytotoxicity were consistently upregulated in infected resistant neurons but not in susceptible ones. These data suggest that more than one biological process is involved in the inhibition of virus production by and continued survival of infected resistant neurons.

#25 Analysis of Staphylococcus aureus biofilms with 3D image analysis software BiofilmQ
Alexander Marchesani, Eric Gilbert

Staphylococcus aureus is a Gram-positive opportunistic pathogen that causes significant numbers of skin and sepsis infections worldwide. The ability to form biofilms is a key trait that allows many pathogens, including S. aureus, to form chronic infections. Biofilms are characterized by thick, dense aggregates of cells that are extremely resistant to antibiotics and the immune system. 4-Ethoxybenzoic acid (4EB) is a plant-derived compound that shows high anti-biofilm activity against S. aureus through an unknown mechanism. Flow cell methodology allows for the cultivation of stabilized biofilms that can be non-destructively imaged with Confocal Laser Scanning Microscopy (CLSM) and analyzed by the 3D image analysis software BiofilmQ. BiofilmQ is a MATLAB-written program developed to study biofilms and uses pseudo-cells to generate binary data from fluorescently dyed cells. BiofilmQ uses these pseudo-cells to compute diverse parameters that quantitatively characterize biofilms, including height, density, substrate area and surface roughness. 4EB treatment resulted in a statistically significant reduction in biofilm height and biovolume, both markers for the health of the biofilm. Surprisingly, many structural aspects of the biofilm like roughness were not significantly impacted. This suggests that 4EB can reduce biofilm robustness without impacting architecture. Current roadblocks in this research include maintaining biological relevance from binary data generated from fluorophores, selection of optimal graph and parameter types for comparison (E.g bin sizes, graphing method and interpretation), managing overexposed images and, ensuring proper quality control via filtering and segmentation. 4EB is among a class of chemicals with similar antibiofilm activity and discovering a novel mechanism could lead to new therapies which may subvert the rising antibiotic resistance in bacterial infections.

#26 The Effects of Neuromodulation on the Propensity for Multistability of Bursting and Silent Regimes
Yousif Shams, Anna Gianella, Mykhailo Fomenko, Gennady Cymbalyuk

Animals survival require expeditious control of rhythmic behaviors like locomotion, breathing, and heart beating. Central pattern generators (CPGs) are distinctive neuronal circuits which control motor patterns. In Medicinal leeches (Hirudo Sp.), heartbeat is controlled by a CPG that is composed of two pairs of mutually inhibitory leech heart interneurons (HNs) forming half-center oscillators (HNHCO). Endogenous neuromodulator myomodulin increases hyperpolarization-activated current (Ih) and decreases Na+/K+ pump (IpumpMax) current in these neurons. Coregulation of conductance of Ih (gh) and Ipumpmax leads to expansion of the functional bursting pattern. Synaptically isolated HN neuron exhibits expansion of seizure-like regime along the coregulation path. We investigated effects of the leak conductance on HN regimes of activity. By upregulating the leak conductance (gleak), the seizure-like pathological regime can be transformed into a high-spike frequency bursting regime and then into the usual low-spike frequency bursting regime. Thus, this upregulation could compensate the adverse seizurogenic effects of neuromodulation. We also investigate co-existence of bursting and silence. The propensity for multistability of these regimes, defined as the range of leak conductance supporting the coexistence of bursting and silent regimes, increases along with comodulation of Ih and IpumpMax current by myomodulin by a factor of 2.5. We conclude that neuromodulation notably increases propensity of multistability of silence and bursting, and the adverse effects of neuromodulation of a CPG, reducing Na+/K+ pump current, may be compensated by co-regulation of other currents such as hyperpolarization-activated current, leak current, and persistent Na+ current.

#27 Constructing an Epistatic Network from Hepatitis C Viral Protein Sequence Data Using Algorithmic Methods
Alina Nemira, Pavel Skums, Alex Zelikovsky, Akshay Juyal, Pardis Sadatian

At the molecular level of life, epistasis refers to the phenomenon where the effect of one genetic mutation is modified or influenced by the presence of another mutation. This epistatic interaction between mutations can affect the resulting phenotype of a living organism. Understanding of epistatic interactions is essential for prediction of how multiple genetic mutations or variations in a virus may interact and influence disease susceptibility, treatment responses, and drug efficacy in patients. Hepatitis C viral genome possibly exhibits numerous epistatic interactions, underscoring the importance of our new computational approach, which joins combinations of specific linked viral mutations from different patients into an epistatic graph or network. We have developed an algorithmic method based on statistics and graph theory for constructing an epistatic network from hepatitis C viral protein sequence data. This approach involves several key steps: constructing a mutation binary matrix, calculating linked mutation pairs, and building an epistatic network. Additionally, we offer an optional step for identifying dense subgraphs within the network. As part of future work, we propose a hypothesis that the identification of dense subgraphs in the epistatic network may be indicative of combinations of specific mutations within the viral genome or emerging viral haplotypes. Those haplotypes can determine the virus's properties, such as its ability to infect hosts, evade the host immune system, or respond to antiviral treatments. The presence of combined specific mutations and their epistatic interactions can lead to the formation of distinct haplotypes. Our hypothesis warrants further investigation and research to explore potential associations between network density and the evolution of viral haplotypes.

#28 Evolution of Visual Networks in Infancy During the Initial Six Months
Masoud Seraji, Sarah Shultz, Armin Iraji, Vince Calhoun

The first six months of life represent a critical period for the development of the human brain's functional and structural foundation. This time frame is when early postnatal experiences can profoundly influence lifelong brain development. In this study, we examined the visual brain networks derived from the application of Independent Component Analysis (ICA) to fMRI data to investigate the spatial development of functional brain networks in 158 infants from birth to six months of age. Using ICA and spatial measurements, we explored how these networks change during this crucial early phase of life.

Our findings revealed that spatial similarity among visual brain networks significantly increased across age for all networks. However, intensity range significantly decreased for the visual networks, indicating a process of consolidation within these networks. Voxel intensity increased for visual networks, reflecting heightened activity patterns as infants matured. Voxel count and the weighted average distance from the center of mass increased, suggesting changes in network size and spatial distribution during the first six months of life.

Furthermore, we analyzed infants' fixation times on eyes, mouth, body, and objects during video clips. Our results demonstrated that infants increasingly fixated on eyes and mouths as they grew older, indicating a heightened interest in social interaction. In contrast, fixation on objects and the body decreased over time, highlighting changes in visual attention patterns.

This study provides valuable insights into the dynamic spatial development of brain networks during the first six months of life and its association with visual attention patterns. Understanding these early developmental processes is crucial for comprehending the roots of neurodevelopment disorders and enhancing our ability to support healthy brain development in infants.

#29 A data-driven approach to modeling cancer migration
Shruti Shrestha, Yi Jiang and Neranjan Suranga Edirisinghe

In our research, we try to develop a data-driven model to identify the characteristics of migratory cancer cells vs non-migratory cancer cells. Our data set consists of close to 100,000 images extracted from the long videos taken at a 1-minute frame rate. We use dimensional reduction techniques to create a reduced representation that is easily manageable by the downstream analysis pipelines. Our goal is to develop a special temporal model which can be used to predict cell migration.
At the initial stage, we use autoencoders to obtain a few latent dimensions that can accurately capture the most information of the original data set. However, with the vanilla autoencoder (AE), we were not able to achieve the desired results. After training the network for 800 epochs, with a mean squared error loss function, we achieved a 10.61% test accuracy in a six-dimensional latent space. We are treating the latent dimension as a hyper-parameter here and working with 3,4,5,6,7 latent dimensions. Next, we incorporate Variational autoencoders (VAE). In theory, VAE should provide better results by learning the underlying probability distribution of the original dataset. We are seeing promising results. at 800 epoch, with a mean squared error loss function, we achieved a 42% test accuracy for VAE.

#30 Ensemble method of polyp segmentation 
Swagat Ranjit and Jian Zhang

It is important to find the polyps in a human system that helps to prevent cancer in medical diagnosis. This poster talks about the approach of ensemble U-Net based convolutional neural network that is used for polyp segmentation. This U-Net architecture has three encoders: Resnet- 50, Resnet-101 along with Efficient- NetB4. They were tre-prained in the COCO dataset and the transfer learning technique was used along with tversky loss. Different data augmentation techniques like rotations, flips, scaling, and contrast along with varying learning rates and optimization were deployed to make a better ensemble method. We proposed a new layer that calculates the average of intermediate masks followed by the sigmoid layer. In our experiment, the performance of the proposed architecture was better than existing methods over the Kvasir-SEG dataset with accuracy 0.9692, recall 0.8872, precision 0.912, dice coefficient 0.8949, and intersection over union 0.987