Scientific Computing Day Poster Presenters
The Scientific Computing Day poster presentation features more than 20 accepted posters, by 50 authors, from multiple disciplines across the university – College of Arts and Sciences, Robinson College of Business, Neuroscience Institute, Center for Nano Optics and many more.
Congratulations to our 2018 Poster Presenters!
Comparing parameter estimates generated by R and SAS through simulations
Abstract: Software used to analyze survey data can generate estimates that are similar but not equal.The estimates may have a large standard error due to the nature of the survey design, especially when subgroups of the population are studied. The true values of the parameters in question are unknown, so it can be challenging to determine if the inflated standard error is an artifact of the survey design, the software, or reflects the true nature of the parameter. This project is the creation of a simulated population based on the 2016 NSCH complex survey. This simulated population will allow the researcher to study the relationship between the survey design, the inflated variance, and the software that is estimating the variance. The parameters will be known because the population is known. The behavior of the simulated population parameters should mimic the behavior of those in the complex survey it is based on. This will be the basis of comparison of the accuracy of estimates produced by R and SAS, and of the behavior of the standard error as the population is sub-setted.
|2.||Global motions and dynamic interactions facilitate MCM-Cdt1 heptamer loading onto origin DNA
Thomas Dodd and Ivaylo Ivanov
Abstract: DNA replication is essential to all living organisms. In eukaryotes, replication licensing begins with the binding of origin DNA by the origin recognition complex (ORC) and Cdc6. Minichromosome maintenance complex (MCM), in complex with Cdt1, is then subsequently loaded onto origin DNA to form an ORC-Cdc6-Cdt1-MCM (OCCM) complex. Our current understanding is that MCM-Cdt1 loading proceeds through DNA insertion between the Mcm2 and Mcm5 subunits (termed gate). However, details on the coordination of the gate opening with the passage of DNA have yet to be elucidated. We have used molecular modeling to delineate the loading mechanism of MCM-Cdt1 onto origin DNA. First, we employed novel path sampling methodologies to construct an optimized transition path for the loading of the MCM-Cdt1 heptamer. We then sampled the conformational space along the transition path using classical molecular dynamics (MD). Principal component analysis (PCA) of the MD trajectories revealed that the global motions of the ORC-Cdc6 influence the conformational dynamics of the MCM gate, suggesting ORC-Cdc6 plays a passive role in the opening of the gate. We also identified key loading intermediates in which MCM secondary structure elements play pivotal roles in stabilizing the DNA as it undergoes passage into the MCM inner cavity. Our results show that global motions of the ORC-Cdc6, along with dynamic MCM-nucleic acid interactions collectively provide an elegant mechanism for the loading of MCM-Cdt1 onto origin DNA.
|3.||Gastrointestinal Motility: A Computational Model Exploring the Role of Subcellular Calcium Oscillators in Pattern Generation
Parker Ellingson, Taylor Kahl, Siddharth Natham, Natalia Maksymchuk, Sarah Johnson, Chun Jiang and Gennady Cymbalyuk
Abstract: Proper digestive functioning requires a variety of coordinated activities in the gastrointestinal tract. During the starved state, peristaltic waves are the dominant pattern of motility in the small intestine. After a meal, the intestine switches to a mixing pattern reminiscent of the beating pattern produced by interacting oscillators with different frequencies. This helps to break up food particles and increase nutrient absorption. These patterns are also modulated by temperature and a variety of pharmacological agents. Rhythm-generating cells electrically connected to the smooth muscle known as interstitial Cells of Cajal (ICC) drive the patterns of motility. The mixing pattern is particularly dynamically interesting and is generally explained as an interaction of two classes of ICC, oscillating at different frequencies (1). We suggest that intrinsic dynamics of a single ICC can explain the mixing pattern as well. We developed models of ICC containing intracellular calcium dynamics, and Hodgkin-Huxley representations of key ionic currents. Both the endoplasmic reticulum and mitochondria are intracellular calcium stores capable of producing calcium oscillations with different periods in the cytosol. We used a mathematical model of subcellular dynamics (4) to observe interactions between calcium oscillations from the ER and mitochondria. A combination of two of the subcellular models, representing the interaction of two spatially unique intracellular calcium oscillators, produced a beating pattern. We compared the results of this model to our experimental recordings of muscle contractions from murine small intestines. Our model suggests a mechanism for this mixing pattern: interactions between two oscillatory calcium subsystems in a single ICC. We also investigated the effects of temperature on motility patterns by adjusting Q10 values and incorporating the dynamics of TRPA1 channels into our model. These results explain how temperature can affect the frequency of oscillations, which is consistent with experimental data (3). In keeping with the coupled oscillator theory of intestinal activity, we investigated the model behavior in a long chain of electrically coupled cells. In conclusion, we demonstrate that ICC are capable of producing a variety of basic regimes of activity corresponding to key motility patterns. Our model shows that factors affecting the internal calcium dynamics impact the period of oscillations, while factors which affect membrane based currents primarily affect amplitude.
Mechanisms modulating speed of traveling wave in neural networks
Ricardo Erazo, Remus Osan and Gennady Cymbalyuk
Abstract: The present study investigates wave propagation in neural networks under multiple conditions and provides a link between wave phenomena observed in simulations with abstract and biophysical neural models. First, how intrinsic properties of neurons such as: number of spikes in a burst, excitatory and inhibitory synapses, and coupling strength; modulate the speed of wave propagation. The research objective is to characterize how spikes influence wave propagation, and contrast the behavior of single-spike waves, multiple-spike waves. Further, we demonstrate that the number of spikes in a burst influences the speed of the propagating wave and thus provides a mechanism by which neural networks modulate wave speed. Second, consider traveling waves in Integrate-and-Fire (I-F) neural networks with excitatory coupling, exponentially decaying synapses, and space-dependent coupling strength. We take the integral form of the I-F model with finite support connectivity and use it to derive the wave speed and acceleration. Numerical simulations were in excellent agreement with the analytical prediction.
Combinatorial Analysis of Local Configurations in Molecular Dynamics Simulations
Ka Chun Ho and Donald Hamelberg
Abstract: An automatic, multi-scale, and three-dimensional (3D) summary of local configurations of the dynamics of proteins can help to discover and describe the relationships between different parts of proteins across spatial scales, including the overall conformation and 3D configurations of side chains and domains. These discoveries can improve our understanding of the function and allosteric mechanism of proteins and could potentially provide an avenue to test and improve the molecular mechanics force fields at different spatial resolutions. Many of the current methods are unable to effectively summarize shapes of 3D local configurations across all spatial scales. Here, we propose Frequent Substructure Clustering (FSC) to fill this gap. Frequent substructure clustering of the Cβ atoms of the GB3 protein identifies six clusters of co-occurring local configurations. The clusters that are localized at different regions contribute to the overall conformation, and form two anti-correlating groups. The results suggest that FSC could describe dynamical relationships between different parts of proteins by providing a 3D description of the frequently occurring local configurations at different spatial resolutions.
Ets-1 ETS domain binding dynamics in variable affinity DNA sequences
Kenneth Huang and Gregory Poon
Abstract: The ETS family of transcription factor proteins is characterized by a winged helix-turn-helix motif present in the namesake (ETS) domain in all members of the family. In addition to having a highly conserved secondary structure, the ETS domain also specifically recognizes a quartet sequence of nucleotides (5’-GGA[A/T]-3’). Despite significant deviations in both functional role and primary sequence over the course of evolution, the ETS domain remains remarkably conserved in its secondary structure. As such, ETS present a model system for the intersection between homologous structure and disparate function. Ets-1, one of the earliest members of the family, has been well characterized for its bias against specific and nonspecific DNA, but understanding of how ETS proteins discriminate against the variations in flanking regions that determine affinity remain unclear. Using molecular dynamics, we determined the molecular basis for selectivity in Ets-1/DNA complexes- several residues (Q336, E343, and R378) are involved in a correlated mechanism to differentiate high/low affinity DNA. Upon binding to low affinity DNA, Q336 suffers a significant loss in contact to DNA, thereupon triggering the salt bridge between E343/R378 to break and expose the hydrophobic core of the protein. This illustrates a mechanism by which minimal length Ets-1 detects and responds to variations in the flanking regions of specific site bearing DNA.
Deep Convolutional Neural Network with Transfer Learning for Biomedical Image Classification
A K M Kamrul Islam
Abstract: In this paper, a deep transfer learning with convolutional neural network(CNN) is considered to identify objects from biomedical images to ensure the better accuracy in terms of image segmentation and classification. We considered several experiments along with SVM and random forests to select appropriate features from an image. Our experiments considered 10 classification datasets that connected residual networks showed better performances comparing with state-of-the art method. Convolutional neural network is
Multimodal Fusion Analysis of Grey and White Matter Changes in Schizophrenia
Abstract:Parallel independent component analysis (pICA) is a data-driven method that identifies the maximally independent components of multiple modalities while simultaneously investigating the strength of their correlations. As yet underutilized in neuroimaging, we applied it to a well-researched brain disorder, schizophrenia, to test the sensitivity of pICA to the correlations between the diverse grey matter volume (GM) and white matter structure (FA) changes and the differences between healthy controls (HC) and subjects (SZ). Method: The images were acquired from 74 subjects with schizophrenia (SZ) and 82 healthy controls (HC). There were 5 iterations of pICA run using varied numbers of components and constraints, the most restrictive being 11×11 (components of each feature – GM and FA), constrained to 3, and the most permissive being 77×77, constrained to 39. Loading coefficients from each feature were t-tested for significant differences between HC and SZ. Results: Each iteration resulted in near maximum numbers of correlations possible for the number of components given. Only two iterations returned correlations that had significant differences in the loading coefficients between HC and SZ – iteration 2 (11×11, constrained to 5, 1 pair) and iteration 4 (39×39, constrained to 19, 3 pairs). HC showed significantly greater loading coefficients in each case. Conclusions: With fine tuning, pICA did return some significant results regarding the differences in correlated changes in GM and FA between HC and SZ. Further investigation into these areas of correlation is needed to determine their impact on our understanding of schizophrenia. .
A Quantum Mechanical Investigation of Interstellar Complex Organic Molecules
Rebecca Johnson and Samer Gozem
Abstract: Interstellar organic molecules of varying complexity have been detected in molecular clouds. For instance, Sagittarius B2, which is 390 light-years away from the center of our milky way, hosts molecules such as aldehydes, ketones, alcohols, and esters. These molecules may be key to understanding the formation of the building blocks of life (such as peptides and nucleotides), but their structural mechanisms are still being investigated. For example, Kaiser and co-workers have explored the formation mechanisms of cyclopropenone and propynal within ice struck by cosmic rays through experiment.1 Following a methodology supplied by a recent joint investigational and computational study by Abplanalp and colleagues,2 we will present preliminary computational data to aid in the interpretation of experimental spectroscopic data. Attention will specifically be given to the formation of cyclopropenone from carbon monoxide and acetylene. Our study also embraces ways of computing photoionization energies and cross sections of small organic molecules from ab initio quantum mechanical methods. Such information can provide fundamental details about the development of molecular foundations essential to universal existence.
SparkFSM: A Highly Scalable Frequent Subgraph Mining Approach using Apache Spark
Cynthia Khan, Bismita Jena and Rajshekhar Sunderraman
Abstract: Knowledge mining from graph data has attracted many researchers over the past several years. With the evolution of internet, computer technology, social networking sites, and web logs, graphs have become a very crucial dataset for mining and finding appropriate knowledge. Based on the application, graphs take different forms, such as airline flight information graph is mostly directed and smaller graphs, chemical compound structures are small and undirected, and social network graphs are very large single graphs based on the different types of associations between people. Earlier, our first attempt to use Hadoop (MapReduce model) to mine the directed frequent subgraphs from a large group of smaller graphs, which is famously known as transaction graphs, proved to be very scalable over the memory-based or database-oriented approaches. In this paper, we introduce SparkFSM, which not only handles undirected and directed graphs, but is also very scalable and efficient in handling isomorphism with the relatively new technology in industry (Spark/Scala). The combination of Spark with the functional style language Scala has established to be a de Facto standard while dealing with in-memory large data processing.
Solvation Effects on the Vibrational Frequencies of p-Benzoquinone
Abstract: This presentation will cover the effects of various solvent molecules on the vibrational frequencies of p-benzoquinone. Specifically, there will be a focus on the carbonyl stretching mode, because hydrogen bonding can cause the mode to split into two uncoupled vibrational modes. Quantum mechanical (QM) and Quantum Mechanical/Molecular Mechanical (QM/MM) simulations were performed using the Gaussian software package to simulate the vibrational frequencies of p-benzoquinone. Two important results are a trend of increased splitting between the two asymmetric carbonyl stretches with increasing hydrogen bond strength, and the necessity of a persistent solvent environment for the most accurate vibrational frequencies.
Simulating the Impacts of Ignoring Multiple Membership in a Piecewise Growth Model
Audrey Leroux, Chris Cappelli and David Fikis
Abstract: A study was conducted to investigate the way a three-level piecewise growth model (3L-PGM) can be extended into a multiple-membership piecewise growth model (MM-PGM) using both sample data from an existing source and simulation study. Estimates demonstrated less bias overall in the MM-PGM, but opportunities for improvements can still be observed in both models. The purpose of this poster presentation is to describe the application of scientific computing principles and resources to facilitate research that may not have otherwise been possible with the capabilities of the typical tools and environments used prominently in the field today. The major findings of this experience revolve around the observation that technical research support and interdisciplinary outreach proved just as essential as processing power to successfully complete the study.
Photoelectron Spectroscopy of Small Organic Molecules
Md Mahbub and Samer Gozem
Abstract: Photoelectron spectroscopy utilizes photoionization or photodetachment to analyze the energetics and movement of electrons and nuclei in molecules. However, interpretation of experimental results with quantum mechanical approaches is not simple, and often requires theory and computer modeling. Calculation of photoionization and photodetachment cross sections requires accurate wave functions of the initial and final states of the systems. Essential information of the initial and final states of the system can be found from Dyson orbital, which in part can be computed accurately from correlated electronic structure methods. However, the final state of the system also requires the description of the free electron wave function. At the moment, no black box method exists that can be applied in a fast and systematic way to obtain accurate photoelectron spectra of polyatomic molecular systems. We will present a series of benchmarks using approximate treatments of the photoelectron wave functions, where we will compare computed and experimental photoionization and photodetachment spectra.
Coding Strategies Driving Calcium-Mediated Cold-Evoked Behavior in Drosophila
Natalia Maksymchuk, Daniel Cox and Gennady Cymbalyuk
Abstract: Temporal coding strategies that sensory neurons use to distinguish between different stimuli, e.g. noxious and innocuous, remains an open question. Studying modality-specific neural activity patterns of Drosophila class III (CIII) multimodal sensory neurons has the potential to provide insights into this fundamental question. CIII neurons are activated by noxious cold temperatures which elicit full body contraction (CT) of an animal. We hypothesize that moderate cold stimulus is associated with spiking activity pattern and noxious cold is encoded with high-frequency bursting. Cold sensitivity of CIII neurons is mediated by the TRP channels Pkd2, Trpm, and NompC. We developed a computational model of CIII neurons that currently exhibits a wide spectrum of qualitatively different activity regimes. Depending on parameter set, our model exhibits two different types of resting states: hyperpolarized or depolarized; two types of spiking activity: small or large amplitude; and high-frequency bursting at noxious cold temperatures. In addition, it was shown that temperature factor Q10 for calcium dynamics determines bistability of the model. At certain values of Q10, high-frequency bursting coexists with small amplitude spiking in some parameter ranges. These regimes are associated with different levels of [Ca2+]i: both resting states and small amplitude spiking produce low levels of [Ca2+]i, while large amplitude spiking and bursting produce high levels of [Ca2+]i. The inclusion of the ITrpm and IPkd2 allows us to consider the problem of coding modality-specific activity patterns by coordinated modality-specific activation of these two TRP currents. The basic model with IPkd2 could represent an alternative scheme of the temperature coding following the sequence of transitions between regimes as a function of declining temperature: silent at room temperatures, spiking at moderately cold temperatures, period doubling cascade, high-frequency bursting at noxious cold temperatures, co-existence of high-frequency bursting and small amplitude spiking. This repertoire of regimes is common for biophysical models of neuronal activity equipped with slow and fast variables governing ionic currents. Finally, TRP channels are not uniquely required for CIII-mediated behavior responses to noxious cold, but together with calcium-activated K+ channels and calcium-induced calcium release mechanism they finely tune optimal [Ca2+]i levels and regulate neural activity patterns for coding modality specific CT behavior.
Computational insight into the rules governing substrate specificity of the glycosylase TDG
Kurt Martin, Thomas Dodd and Ivaylo Ivanov
Abstract: Thymine DNA glycosylase (TDG) is a pivotal enzyme with dual roles in both genome maintenance and epigenetic regulation. TDG is involved in cytosine demethylation at CpG sites in DNA. Here we have used molecular modeling to delineate the lesion search and DNA base interrogation mechanisms of TDG. First, we examined the capacity of TDG to interrogate not only DNA substrates with 5-carboxyl cytosine modifications but also G:T mismatches and nonmismatched (A:T) base pairs using classical and accelerated molecular dynamics. To determine the kinetics, we constructed Markov state models. Base interrogation was found to be highly stochastic and proceeded through insertion of an arginine-containing loop into the DNA minor groove to transiently disrupt Watson–Crick pairing. Next, we employed chain-of-replicas path-sampling methodologies to compute minimum free energy paths for TDG base extrusion. We identified the key intermediates imparting selectivity and determined effective free energy profiles for the lesion search and base extrusion into the TDG active site. Our results show that DNA sculpting, dynamic glycosylase interactions, and stabilizing contacts collectively provide a powerful mechanism for the detection and discrimination of modified bases and epigenetic marks in DNA.
Full Stack Serverless Processing Pipeline For The MUG-C Calcium Prediction Algorithm
Melchizedek Mashiku, Kun Zhao, Michael Kirberger, Jenny J Yang, Suranga Edirisinghe
Abstract:This web development project design and implement a science gateway for the MUG-C java application. MUGC application finds calcium binding sites provided the Protein Data Bank (PDB) file. Our project creates a full stack solution to hosting the front end on Amazon Web Services. We use the S3 storage, AWS Lambda functions, and API gateway to relay the PDB files to the back-end computing in GSU’s HPC ACoRE. We architected a full stack serverless processing pipeline that allows users to access the application. Our design optimizes for the scalability, reliability, security, performance, and cost.
Ultrafast valley polarization in two-dimensional materials
Seyyedeh Azar Oliaei Motlagh, Jhih-Sheng Wu, Vadym Apalkov and Mark Stockman
Abstract: The intense laser pulse can significantly populate the conduction band of two-dimensional (2D) materials. In 2D materials with honeycomb lattice structure, the strong circular pulse populates one valley, K or K’, significantly depending on the pulse polarization right or left respectively. Under this condition, the significant difference of the population of the valleys leads to the ultrafast valley polarization. In this work, we study theoretically and numerically the ultrafast valley polarization in the monolayer of transition metal dichalcogenide and the gapped graphene. The predicted ultrafast valley polarization in 2D materials has the potential to be used in ultrafast storage devices.
Femtosecond strong-field interferometry in Weyl semimetals: New degree of freedom, Chirality and Berry phase
Fatemeh Nematollahi, Vadym Apalkov, Jhih-Sheng Wu and Mark Stockman
Abstract: We theoretically introduce a new degree of freedom in three-dimensional topological Weyl semimetals by using a single oscillation circularly polarized optical pulse with a duration of a few femtoseconds. The population of electrons in the conduction band in reciprocal space is highly structured and defined by a topological phase. Also, we present interferometry in Weyl semimetals without a magnetic field and by employing strong ultrafast circularly polarized optical pulse. The pulse consists of two oscillations with different circularity. We illustrate the distribution of electrons in momentum space at two Weyl points is highly chiral which present the intrinsic chirality of Weyl nodes.
Modeling the spectral properties of flavin cofactor in different environments by using hybrid quantum/classical approaches
Mohammad Pabel Kabir, Yoelvis Orozco-Gonzalez and Samer Gozem
Abstract: Fluorescent proteins (FPs) are widely used genetically encoded reporters that have revolutionized the field of bioimaging. While the green fluorescent protein (GFP) is the most popular example, other examples include flavin-binding photoreceptors (FbFPs), which have been recently developed from Light Oxygen Voltage (LOV) blue-light sensing domains. FbFPs have an advantage over GFPs for some applications because they are smaller (less genetic content to encode) and can be used in anaerobic environments, while GFP derivatives and homologues require molecular oxygen for the maturation of their chromophore. Conversely, FbFPs bind FMN as chromophore, which is synthesized independently of molecular oxygen. However, FbFPs are at a nascent stage of development and have been utilized in only a handful of biological studies, unlike GFPs which have been studied extensively both experimentally and computationally. Therefore, we are still lacking a good understanding of the spectroscopic properties and color tuning mechanism of FbFPs. Toward such an understanding, we modeled the spectral properties of FMN using ab initio quantum mechanical methods. Our goal is to use hybrid quantum mechanical/molecular mechanical (QM/MM) methods to study the same chromophore in different protein environment. We will present preliminary data towards this goal.
Properties of a Fast Transient Rhythm Elicited in a Multifunctional Central Pattern Generator
Jessica Parker, Boris Prilutsky and Gennady Cymbalyuk
Abstract: Could two drastically different rhythms such as in cat locomotion and paw-shaking be controlled by the same network of neurons? To answer this question, we built a model of a multifunctional central pattern generator (CPG). Our model, constructed as a half-center oscillator (HCO), is able to produce multistability of a locomotion-like rhythm and a paw-shake-like rhythm. It uses a novel mechanism involving two slow currents, i.e. slowly inactivating calcium current and slowly inactivating sodium current . Transient paw-shake-like activity can be elicited in our model, and this transient activity exhibits asymmetric trends throughout consecutive bursts in accordance with experimental data. We investigated the model’s responses to various types of afferent stimulation during locomotion-like activity and transient paw-shake-like activity. We predict that applying a 1-second pulse of current to groups I and II afferents from cat hip flexors and extensors during locomotion, which have access to the flexor- and extensor half-centers of CPG rhythm generator , will evoke a paw-shake response in that hindlimb. According to our model, the duration of this transience depends on the phase of stimulation in the locomotion rhythm. Also, the duration of transience increases with the duration of the pulse. The duration of transient paw-shake-like activity could be extended when a short pulse of current is applied during transient paw-shake-like bursting. We predict that applying a short 20-millisecond pulse of excitatory current to groups I and II afferents from either hip flexors or extensors during a paw-shake response will extend the duration of the paw-shake response. Furthermore, the duration of the paw-shake response would increase as the duration of this stimulus increases until some threshold duration is reached at which the duration of the paw-shake response will remain roughly constant as the stimulus duration increases. In addition, the extension of the response would depend on the phase of pulse application in the paw-shaking cycle. The extended paw-shake response could last longer if the pulse is applied during the extensor phase as opposed to the flexor phase if the pulse is applied near the beginning of the paw-shake response. The asymmetry weakens if the pulse onset is delayed during the paw-shake response. These predictions are robust and can be tested experimentally to investigate whether the obtained responses during locomotion and paw-shaking are consistent with the idea that the two rhythmic behaviors are generated by the same multifunctional CPG. Confirming these predictions experimentally would provide strong evidence for the hypothesis that the paw-shake response in cats is generated as a transient response of the locomotion CPG.
Studies of four investigational protease inhibitors with HIV-1 protease bearing drug resistant substitutions of V32I, I47V and V82I
Abstract: HIV-1 protease (PR) has developed resistance mutations to all 9 FDA-approved clinical protease inhibitors (PIs). This study examines the effects of four investigational PR inhibitors against HIV-1 PR with drug resistant mutations of V32I, I47V and V82I (PRTri). From the total number of V32I, I47V and V82I single mutation patients, approximately, 60-90, 70-90 and 30-50% patients are found to be resistant to inhibitors FPV, ATV, IDV, LPV, NFV, SQV, TPV and DRV. The antiviral inhibitors, GRL-0249, GRL-0519, GRL-0739 and GRL-1111, introduce diverse chemical modifications on the darunavir (DRV) scaffold, and form new interactions with wild type PR. These inhibitors were expected to show increased interactions and improved inhibition towards the PRTri mutant relative to current PIs, however, the measured inhibition constants range from 5-10-fold higher than that for amprenavir. The structure of PRTri mutant in complex with GRL01111 was determined and compared with the corresponding wild type PR complex to analyze its interactions with inhibitors.
ONIOM Vibrational Frequency Calculations for the Prediction of the Orientation of Pigments in Protein Binding Sites
Leyla Rohani and Gary Hastings
Abstract: In all photosynthetic reaction centers, light induces the transfer of electrons via a series of acceptors across a biological membrane. In photosystem I photosynthetic reaction centers one of the central intermediates in electron transfer is a phylloquinone (PhQ) molecule, which occupies the so-called A1 protein binding site. Time resolved FTIR difference spectra have been obtained for PSI with PhQ, and several other non-native quinones incorporated into the A1 binding site. The goal of this work is to develop ONIOM-type QM/MM computational methods in order to simulate all of these experimental spectra. We show that the calculated spectra are sufficiently different for quinones in different orientations in the A1 binding site, and through comparing calculated and experimental spectra we can predict the precise orientation of the different quinones in the A1 binding site. The specific molecular structures, including water molecules, and the detailed computational methods employed, will be discussed.
Assessing parameter identifiability in compartmental dynamic models using a computational approach: Application to infectious disease transmission models
Kimberlyn Roosa and Gerardo Chowell
Abstract: Background: Compartmental dynamic models are commonly used in public health research to understand potential underlying mechanisms of infectious disease dynamics, assess patterns in epidemiological data, and forecast the trajectory of epidemics. However, the successful application of mathematical models to guide public health interventions and decisions lies in the ability to reliably estimate model parameters and their corresponding uncertainty. Here, we present and illustrate a simple, computational method for assessing parameter identifiability in compartmental epidemic models. Methods: We describe a parametric bootstrap approach to generate simulated data of the incidence curve in order to derive the empirical distributions of the estimated parameters. These distributions are then used to quantify confidence intervals and relative biases of estimated parameters to assess structural parameter identifiability. To demonstrate this approach, we begin with a low-complexity SEIR model and work through examples of increasingly more complex compartmental models that correspond with real-world applications to pandemic influenza, Ebola, and Zika. Results: Overall, parameter identifiability issues are more likely to arise with more complex models (based on number of equations/states and parameters). Also, as the number of parameters being jointly estimated increases, the uncertainty surrounding estimated parameters tended to increase, on average, as well. We found that, in most cases, R0 is often robust to parameter identifiability issues affecting individual parameters in the model. Despite large confidence intervals and high relative bias of other individual model parameters, R0 could still be estimated precisely and with little bias. Conclusions: Because public health decisions can be influenced by results from modeling studies, it is critical that modelers conduct parameter identifiability analyses prior to fitting the models to available data. The method described is helpful in this regard and adds to the essential toolkit for conducting model-based inferences using compartmental dynamic models
Robust Human Activity Recognition from RGB Video Stream with Limited Training Data
Abstract: Human activity recognition based on video streams has received numerous attentions in recent years. Due to lack of depth information, RGB video based activity recognition performs poorly compared to RGB-D video based solutions. On the other hand, acquiring depth information, inertia etc. is costly and requires special equipment, whereas RGB video streams are available in ordinary cameras. Hence, our goal is to investigate whether similar or even higher accuracy can be achieved with RGB-only modality. In this regard, we propose a novel framework that couples skeleton data extracted from RGB video and deep Bidirectional Long Short Term Memory (BLSTM) model for activity recognition. A big challenge of training such a deep network is the limited training data, and exploring RGB-only stream significantly exaggerates the difficulty. We therefore propose a set of algorithmic techniques to train this model effectively, e.g., data augmentation, dynamic frame dropout and gradient injection. The experiments demonstrate that our RGB-only solution surpasses the state-of-the-art approaches that all exploit RGB-D video streams by a notable margin. This makes our solution widely deployable with ordinary cameras.
Rainfall Variability in Western Uganda
Hae Seung Sung and Jeremy Diem
Abstract: It is widely believed that equatorial Africa has experienced a drying trend over the past several decades. And the decreasing rainfall should have had large negative impacts on the livelihoods of smallholder farmers. But those previous studies have not always relied on the best available rainfall data, and there now exist high-resolution, multi-decadal data to assess rainfall variability and trends in the region. Using multiple datasets for the 1983-2017 period, this study examines rainfall variability and the atmospheric controls of rainfall days in western Uganda, which is a climatologically transitional zone between central equatorial Africa and eastern equatorial Africa. Daily rainfall estimates were extracted from two satellite-based rainfall databases: Climate Hazards Group InfraRed Precipitation with Stations (CHIRPS) and TAMSAT African Rainfall Climatology And Time-series (TARCAT). Daily atmospheric data were extracted from the Reanalysis 2 data and ERA-Interim databases. Using hierarchical climate regionalization, we confirmed that western Uganda can be divided into five homogeneous rainfall regions. There are two rainy seasons in western Uganda, from March to May and from August to November, except the northernmost region which shows a unimodal rainy season from late March to mid-November. The typical atmospheric conditions during the rainy seasons are characterized by the decreased downward motion in the middle troposphere, westerly-wind anomalies in the lower troposphere, and high specific humidity in both the lower and middle troposphere. Rainy days with the rainy seasons (i.e., not all days in a rainy season have rainfall) had similar characteristics: a decline in the mid-tropospheric vertical velocity and more westerly wind compared to the typical rainy-season day. The temporal analysis of each rainy season revealed that the rainy seasons have started earlier– and thus increased in duration – over the past 35 years. The earlier season onset is caused by significantly increased rainfall in the weeks just prior to the typical start of the rainy seasons. Decreasing downward motion and increasing specific humidity in the middle troposphere are the principal causes of the increased rainfall. These results run counter to the prevailing view of western Uganda – based on analyses of inappropriate rainfall data – that the rainy seasons have been getting shorter and annual rainfall has decreased over the past several decades.
Literature keyword recommendations without negative examples: Semi-supervised learning with multiple views of negative feature space
Matthew Turner, Keven Haynes, Amber Tannahill and Jessica Turner
Abstract: Machine learning methods for assigning keywords to scientific literature suffer from a shortage of human labelled training data. Keyword assignments are incomplete; keywords are correctly assigned to true positive instances, but many positives are not labelled. We used full-text articles from Frontiers in Psychology (7481 articles) and their author-assigned keywords as data. The collection was divided into 3 sets: POSITIVES, papers with the keyword assigned; LIKELY NEGATIVES, papers with no keyword assigned and no occurrence of the keyword in the text; and UNKNOWN, papers with no keyword assigned, but which have the keyword appear in the text. We trained 100 individual naïve Bayes classifiers using the POSITIVES (same for each) and randomly chosen sets of LIKELY NEGATIVES as data. These individual classifiers were combined with a rule requiring 90% agreement to indicate the keyword should be applied. As an example, we manually evaluated the keyword “music.” The number of POSITIVES was 104; the UNKNOWNS were 682; and the rest were LIKELY NEGATIVES. The classifier assigned the keyword to 326 papers. The probability of assignment related to the number of occurences of the keyword in the text, from 0.19 when “music” occurs once to 0.96 when it occurs ≥ 10 times. An expert reviewed a random set of 40 articles and agreed with the classifier on 90% of the papers, indicating a surprising level of agreement on unknowns, given the absence of formally classified negative example
Intermittent Bursting regimes of a Model Neuron from the Pre-Bötzinger Complex
Alex Vargas and Gennady Cymbalyuk
Abstract: The Pre-Bötzinger Complex (PBC), located in the medulla of the brainstem, produces patterns responsible for the inspiratory phase of breathing (1,2). We focus our study on interactions between key currents supporting endogenous bursting in our model of a Pre-Bötzinger Complex neuron. We hypothesize that A-type current contributes to the maintenance of functional bursting regime along with the pump, persistent sodium, and h-currents. These currents have been implicated in the dynamics of intermittent bursting regime, recorded in the episodic swimming behavior of tadpoles (3) and mammalian locomotory CPG (4). We developed a single compartment model to simulate a PBC neuron which is intrinsically bursting through a persistent sodium mechanism. The model describes dynamic intracellular Na+ concentration which determines the reversal potential for all sodium currents. We demonstrate that our canonical model produces functional bursting under normoxic parameters and that the decrease of the pump strength corresponding to hypoxia generates intermittent bursting activity. The stronger A-type K+ current becomes, the longer the interbout interval becomes. We investigated dynamical mechanisms underlying the role of the Na+/K+ pump and its interactions with specific ionic currents found in the PBC and other mammalian CPGs. We find that this carries significance towards further understanding pathological vulnerabilities in the respiratory centers of the brain.
Bayesian Deep Learning for Active Data Acquisition
Xiulong Yang and Shihao Ji
Abstract: Deep neural networks have achieved the state-of-the-art performance on many recognition and learning tasks, such as image classification, speech recognition and natural language processing. However, training a deep model requires a large amount of labeled data, which is rather restrictive in practice since collecting a large set of labeled data is very expensive and time-consuming. One way to ease this problem is active learning, which intelligently chooses data samples for labeling to achieve the highest possible recognition accuracy with a limited labeling budget. The core idea of active learning is uncertainty estimation that evaluates the informativeness of data samples for label acquisition, while Bayesian learning is the most principled way for uncertainty estimation. In this work, we investigated three different Bayesian deep learning algorithms for active learning, including Bayesian dropout, deep ensembles and Stein Variational Gradient Descent (SVGD). Experiments on convolutional neural networks show that we can achieve similar accuracy with an order of magnitude less labeled data.