TACCSTER 2022 Proceedings
Permanent URI for this collectionhttps://hdl.handle.net/2152/115318
Browse
Browsing TACCSTER 2022 Proceedings by Issue Date
Now showing 1 - 20 of 31
- Results Per Page
- Sort Options
Item Automated Identification of Cotton Diseases and Pests(2022) Noore, Adly; Beksi, William J.There is a constant need for optimizing food production and farming across the world, espe- cially in impoverished countries where the economy heavily depends on agriculture. The major contributing factors to crop yield loss are pest infestations and plant diseases. The U.N. Food and Agriculture Organization reports that 40% of crops are lost due to pests and diseases. This research project aims to contribute to the field of AI-based assistive technologies in precision agriculture by helping develop models to aid farmers in accurately identifying cotton pests and diseases from uploaded cotton leaf pictures. Farmers will be able to mitigate crop loss by (i) taking remedial ac- tion before additional damage occurs, (ii) minimizing pesticide waste by only spraying unhealthy crops, (iii) mapping areas of the field impacted by infestations, and (iv) reducing the health risks associated with the presence of excessive pesticides in the consumer’s food. However, training a high-quality model is challenging since real-world data contains more healthy leaves than diseased leaves and some plant diseases are more prevalent than others. This results in undesirable class im- balances and biases which lower the model’s accuracy. To address these imbalances, advanced data augmentation with generative adversarial networks (GANs) were used to create realistic-looking diseased plant images. These images were added to the underrepresented classes to balance the dataset and improve the model’s accuracy. Training high-quality GANs is computationally expen- sive. Each GAN took approximately 48 hours to train on a node with two V100 GPUs using TACC resources. Traditional affine data augmentation was also performed separately to compare with the GAN results.Item Applications of Machine Learning Algorithms for Coral Disease Fate in Caribbean Corals(2022-09-29) Van Buren, Emily W.; Beavers, Kelsey; MacKnight, Nicholas; Wang, Li; Mydlarz, Laura D.The Caribbean is known as a coral disease “hot spot” due to the high prevalence of acute and chronic diseases that have plagued corals in the area. Two diseases, Stony Coral Tissue Loss Disease (SCTLD) and White Plague (WP), are common and infect many coral species. These two diseases have been studied in a genotype- matching study that looked at transcriptomics of baseline, and post-exposure to disease in four species of corals. While transcriptomic studies have improved our knowledge of host response, a knowledge gap regarding the disease risk corals have prior to disease exposure still exists. Understanding disease risk before an outbreak is an important step in modeling disease dynamics of corals as it will help conservation efforts and disease response protocols. One way to identify disease risk is the application of machine learning to identify patterns of expression based on disease outcome. By applying novel but proven layers of machine learning programs from medical research and using healthy corals whose disease fate are known, we can identify which biological processes are relevant to disease susceptibility. We examined six different types of machine learning algorithms for detection of presence/absence of genes and expression patterns correlated o whether the coral got disease when exposed or not. We will report what types of data these algorithms provide and how it can be applied for disease motoring and modeling.Item Automated Detection and Tracking of Infield Cotton Bolls(2022-09-29) Muzaddid, Md Ahmed Al; Beksi, William J.Cotton fiber accounts for nearly 25% of world-wide textile fiber use. Texas produces more cotton than any other state. It contributes approximately 45% of the U.S. cotton production, with about 25% of the entire U.S. crop, and plants more than 6 million acres. Cotton is the state’s leading cash crop and it ranks third behind the beef and nursery industries. The number of cot- ton bolls on a given farm is arguably the most important phenotyping trait. It provides a better understanding of the physiological and genetic mechanisms of crop growth and development that supports breeding research. Currently, the standard approach to obtain cotton boll counts is by manual sampling via human visual inspection, which is tedious, labor intensive, and error prone. To address the inefficiencies in this approach, we developed an automated vision-based system for cotton boll counting from infield videos where each track uniquely identifies a cotton boll, and the total number of tracks equals the estimated cotton boll count as follows. First, we identify the relationship among the locations of neighboring cotton bolls and model them in a probabilistic framework to handle occlusions. In addition, we exploit dense optical flow and utilize particle filtering to guide each tracker. Then, correspondences between detections and tracks are found through data association via direct observations and indirect cues, which are then combined to obtain an updated observation. We highlight the efficacy our approach in detecting and tracking cotton bolls against other cotton boll counting methods, along with three state-of-the-art tracking methods. TACC’s computational and storage resources were essential for obtaining the results reported in this project.Item Real-space Methods for Electronic Structure Calculations of over 100,000 Atoms(2022-09-29) Dogan, Mehmet; Liou, Kai-Hsin; Chelikowsky, James R.Two factors limit our ability to accurately describe the properties of materials: (1) the ability characterize multiple electron interactions, and (2) the computational tools to solve the resulting equations. With density functional theory (DFT) and the use of pseudopotentials, the electronic structure problem can be effectively solved for many weakly coupled systems. Computational cost of the Kohn–Sham equations is still a problem, frequently restricting the systems of interest to just a few thousand or fewer atoms. Here, we discuss novel methods that let us solve systems that contain more than 100,000 atoms. We concentrate on new computational algorithms based on real-space DFT and pseudopotentials. Our strategy has several benefits. The global communication required for fast Fourier transforms is avoided by real-space formalisms, such as finite differences and finite elements, which also provide superior scalability for big calculations across hundreds or thousands of computer nodes. Furthermore, finite-difference techniques with a uniform real-space grid offer simple implementation; for instance, the grid spacing alone determines how quickly a Kohn-Sham solution converges. Based on a Chebyshev-filtered subspace iteration method (CheFSI), we developed a promising approach for solving the Kohn–Sham equations in real space. We will illustrate two improvements on CheFSI to enhance scalability and accelerate the calculations: (1) a hybrid method that combines a spectrum slicing method and CheFSI, which divides a Kohn–Sham eigenvalue problem into subproblems wherein each subproblem can be solved in parallel using CheFSI; (2) a grid partitioning method based on space-filling curves which improves the efficiency of the sparse matrix–vector multiplication—the key component in CheFSI. We show with computations of confined systems with over 100,000 atoms or 400,000 electrons, that this method effectively reduces the communication overhead and improves the utilization of the vector processing capabilities provided by most modern parallel computers.Item Phosphorylation of Tyrosine 841 Strongly Affects JAK3 Kinase’s Activation(2022-09-29) Sun, Shengjie; Li, LinJanus Kinase 3 (JAK3) plays a key role in the proliferation, development, and differentiation of various cells. It regulates gene expression by phosphorylation of Signal Transducer and Activators of Transcriptions (STATs). A new JAK3 kinase domain phosphorylation site was found, tyrosine-841 (Y841). The effects of phosphorylated tyrosine-841 (pY841) on ATP/ADP binding affinities of the JAK3 kinase domain were systematically studied and reported here. With the support of TACC, we applied long all-atom molecular dynamic simulations to study the effects of phosphorylation on Y841. The results show that pY841 reduces the size of the cleft between the N-lobe and C-lobe of the JAK3 kinase domain. However, when an ATP/ADP is bound to the kinase pY841 was found to enlarge the cleft. Additionally, for unphosphorylated JAK3 (JAK3-Y841), the binding forces between the kinase domain and ATP or ADP are similar. After phosphorylation of Y841, JAK3-pY841 exhibits more salt bridges and hydrogen bonds between ATP and kinase than ADP and kinase. Consequently, the electrostatic binding force between ATP and kinase is higher than that between ADP and kinase. The result is that compared to ADP, ATP is more attractive to JAK3 when Y841 is phosphorylated. Therefore, JAK3-pY841 tends to bind ATP rather than ADP.Item Investigating the Permissive Environment of Perisynaptic Astroglia for Information Storage in the Dentate Gyrus(2022-09-29) Nam, A.J.; Kuwajima, M.; Mendenhall, J.M.; Hubbard, D.D.; Hanka, D.C.; Parker, P.H.; Wetzel, A.; Bartol, T.M.; Sejnowski, T.J.; Abraham, W.C.; Harris, K.M.Perisynaptic astroglial processes (PAPs), are active modulators of neuronal activity and directly contribute to information processing in the brain. Both in vivo and in vitro experiments have demonstrated that PAPs undergo activity-dependent structural changes. Thus, here we employ cutting-edge resources at the Texas Advanced Computing Center (TACC) to explore PAP structural remodeling associated with long-term potentiation (LTP) and long-term depression (LTD) that may help support local changes in information processing. Long-term potentiation (LTP) and long-term depression (LTD), widely accepted cellular mechanisms of learning and memory, were induced in vivo in the awake adult rat hippocampal dentate gyrus. LTP induction in the middle molecular layer (MML) was achieved by delta-burst stimulation in the medial perforant pathway, a procedure that produced concurrent long-term depression (cLTD) in the outer molecular layer (OML). The contralateral control hemisphere received only baseline stimulation to the medial perforant path. Three-dimensional electron microscopy (3DEM) offers significant advantages over two-dimensional approaches including a more complete view of ultrastructure in all X-Y-Z planes. AlignEM Swift, the state-of-the-art interactive application available at TACC, is integral for achieving the standard of perfect serial section image alignment needed for 3DEM analysis. Furthermore, Blender at TACC, equipped with the computing power of TACC’s supercomputers, similarly facilitates large-scale and realistic PAP reconstructions for visualization and quantitative mesh analysis. Changes to PAP ultrastructure have important implications on the spatiotemporal dynamics of astrocyte calcium signaling. Thus, TACC resources will further enable computational modeling to investigate the functional consequences of PAP morphological changes. Preliminary analysis suggests that more than 80% of all dentate gyrus synapses exhibit some degree of PAP apposition at the axon-spine interface (ASI). Results from this study made possible using TACC systems will contribute to our overall understanding of the cellular mechanism of information processing and the role of specifically astrocytes in this process.Item Flow and Scalar Transfer Characteristics for a Circular Colony of Vegetation(2022-09-29) Kingora, Kamau; Raza, Mishal; Sadat, HamidLocal and global flow structures, as well as transfer and transport of a passive scalar from a circular colony of uniformly distributed vegetation stems, are investigated at Re = 2100, Re = 4200, and Re = 8400. The number of stems in the colony is varied from 1 to 284 yielding a solid fraction of 0.0<𝜙𝜙<0.65. The following three flow regimes are identified: a co-shedding flow regime prevails at low solid fraction where wakes of individual cylinders have minimal interaction; a bleeding-wake flow regime is identified at intermediate solid fraction in which stream-wise bleeding flow delays the formation of colony-scale vortices yielding a steady wake between two separated shear layers; and a single-body flow regime is observed at high solid fraction and is accompanied by the commencement of colony-scale vortex shedding. As Reynolds number increases, the separated shear layers observed at intermediate solid fraction break up to form stem- scale vortices that organize themselves in colony scale coherent structures. As the solid fraction increases, drag and Sherwood number experienced by colonies increases linearly and at a reducing rate at low and intermediate solid fractions, respectively, while the net lift remains negligible. At high solid fraction, the commencement of colony-scale vortex shedding is accompanied by a jump in lift and base suction. Pressure and friction lift/drag increase and decrease with an increase in solid fraction, respectively, toward the value experienced by a solid cylinder. Sherwood number, on the other hand, decays exponentially toward the value experienced by a solid cylinder at high solid fraction. Colonies at intermediate solid fraction exhibit the highest scalar transfer but weakest transport in their near field wake. Scalar transfer in colonies with high solid fraction deteriorates with an increase in solid fraction yielding less scalar concentration in their downstream wake. Each case consist of about 14M computational points and computations were performed on TACC LS6 clusters. A typical case converges in 128,000 processor hours.Item GPU-Acceleration of PDE Solvers for Large-Scale Wave Simulation(2022-09-29) Hanindhito, Bagus; Gourounas, Dimitrios; Fathi, Arash; Trenev, Dimitar; Gerstlauer, Andreas; John, Lizy K.Large-scale simulations of wave-type equations have many industrial applications, such as in oil and gas exploration. Realistic simulations, which involve a vast amount of data, are often performed on multiple nodes of an HPC cluster. Using GPUs for these simulations is attractive due to considerable parallelizability of the algorithms. Many industry-relevant simulations have characteristics in their physics or geometry that can be exploited to improve computational efficiency. Furthermore, the choice of simulation algorithm impacts computational efficiency significantly. In this work, we exploit these features to significantly improve performance for a class of problems. Specifically, we use the discontinuous Galerkin (DG) finite element method, along with the Gauss- Lobatto-Legendre (GLL) integration scheme on hexahedral elements with straight faces, which then greatly reduces the number of BLAS operations, and simplify the computations to Level-1 BLAS operations, reducing the turnaround time for wave simulation. However, attaining peak performance of GPUs is often not possible in these codes that exacerbate bottlenecks caused by data movement, even when modern GPUs enjoying the latest high-bandwidth memory are being used. We have developed an efficient and scalable GPU-accelerated PDE solver for Wave Simulation, by using hardware- and data-movement-aware algorithms. While significant speed-up over CPUs can be achieved, data movement still limits GPU performance. We present several optimization strategies, including kernel fusion, Look-Up-Table-based neighbor search, improved shared memory utilization, and SM-occupancy-aware register allocation. They improve performance up to 84.15x over CPU implementations and 1.84x over base GPU implementations on average. We then extend the functionality to support multi-GPUs on multi-node HPC clusters for large-scale wave simulations, and perform additional optimizations to reduce communication overhead. We also investigate the performance of several MPI libraries in order to fully overlap communication and computation. We are able to reduce the communication overhead by 70%, and achieve weak-scaling over 128 GPUs in TACC Longhorn GPU cluster.Item MuBuCo: Mutation Burden Composition(2022-09-29) Pugalenthi, Lokesh; Richardson, Jensen; Jiang, Wenxuan; Srinivasan, Harish; Reddy, Himanshu; Pritha, Jafrin; Nanduri, Rahul; Hong, Raymond; Kuhlman, Christopher; Prasad, Rohit K.; Arasappan, Dhivya; Kowalski-Muegge, JeanneGene mutations can vary by type, in terms of affecting a single site, spanning hundreds of base pairs, or their over/under-representation in cancer cells. Collectively, these mutation types include: single nucleotide variants (SNVs), structural variants (SVs), and copy number variants (CNVs). Tumor mutation burden is one measure widely used throughout cancer research but it is often limited to a single dimension, using SNVs only. We derive a sample mutation burden for each mutation type and combine them to define their relative contribution, forming a mutation burden composition (MuBuCo). We applied MuBuCo to multiple myeloma, a well-recognized, genomically heterogeneous blood cancer. Using 70 multiple myeloma cell lines, we first computationally assessed more than 15 bioinformatics tools to detect each type of variant and selected the best performing ones (using known features) to calibrate and characterize these cell lines. This required more than 10,500 node hours to run on Texas Advanced Computing Center clusters. We also developed a snakemake pipeline incorporating preprocessing and SNV calling. Each cell line’s variant calls were used to calculate each mutation type burden. We further defined expressed mutations by variants found in expressed genes to predict neoantigens. We implemented the results in a query-able application that enables cancer researchers to select MM cell lines of interest and visualize its MuBuCo relative to other cell lines. With this information, we hope to improve our understanding of the molecular background against which these cell lines are used for testing new treatments. We further provide an in silico look at changes in cell lines’ MuBuCo from user specified removal of a single or multiple genes, mimicking a ‘knock out’ experiment. Our application offers a novel mutation analyses whose results are not readily attainable, until now.Item Improving Structural Brain Connectomes through Statistical Evaluation via Model Optimization(2022-09-29) Heinsfeld, Anibal Sólon; McDonald, Daniel J.; Pestilli, FrancoAccurate mapping of the structural brain connectomes is fundamental to understanding the role of white matter in health and disease. Diffusion-weighted magnetic resonance imaging (dMRI) and fiber tractography provide the only way to map brain connectomes in living human brains. Several studies have shown technical gaps in robustly mapping brain connectomes. The lack of connectome evaluation methods is evident from the recent findings. The present work focuses on developing methods for the statistical evaluation of brain connectomes. We present a new method that builds on LiFE and COMMIT2 methods to reduce a candidate tractography to an optimized one by identifying the brain connections that best model the dMRI signal. We used sparse group regularization, which requires finding a parameter (λ) for the trade-off between better fitting the signal with individual streamlines while maintaining the bundle's cohesion. Previous methods using regularizations to evaluate connectomes set fixed λs, refitting the model for several values of λ. We propose an efficient approach to selecting the optimal λ value. We performed experiments to test the complexity and efficacy of the approach using two datasets: simulated and real datasets. The simulated data were generated using Phantomas, with simple bundles and tissue factors. In addition, we used diffusion data from the Human Connectome Project (HCP). Results show that our approach can identify the optimal λ in a reliable amount of time. The full λ optimization process for 100 different λ took 17 min on a standard desktop computer, while it takes 4x more time than COMMIT 2 to select the optimal λ. In addition, the model's mean squared error is 0.0036 for the HCP dataset and 3.89e-5 for the simulated dataset. This is 14.78x less than COMMIT 2 (0.0544). The reduction in error is due precisely to the optimized selection of λ.Item Molding Aortic Valve Hemodynamics Using a Novel Immersed Boundary Method(2022-09-29) Raza, Mishal; Kingora, Kamau; Sadat, HamidThis research entails the study of the transfer and transport of a passive scalar around the aortic valve to aid in understanding Calcific Aortic Valve Disease (CAVD). Simulations were conducted using a novel interpolation-free sharp-interface immersed boundary method. The method is generic in nature, enabling imposing boundary conditions for scalar concentration to investigate CAVD. In this study, the 3D geometry of the native tricuspid AV including the cusps, commissures, and sinuses will be reconstructed based on the parametric model developed by (Rami Haj-Ali 2012) based on the AV anatomy and measurements reported in the literature. We will solve advection-diffusion transport equations to find the scalar transport, albeit in a Fluid- Structure Interaction (FSI) setting. The FSI framework will be based on the developed immersed boundary coupled with a solid solver (Calculix (Guido Dhondt 2020)) using preCICE (preCICE 2021). The results will be employed to evaluate the distribution of scalar concentration on leaflets as well as to understand the correlation between the level of concentration and valve movements. The correlation between the predicted scalar concentration and several WSS-based parameters (WSS, WSSG, OSI, GON, RRT) will be also investigated. Parallel simulations are to be conducted on a High-Power Computer from Texas Advanced Computing Center (TACC). Approximately 15 million computational points will be decomposed into 560 processors.Item How To Incorporate CI/CD When Your Science Requires HPC(2022-09-29) Stuart, GeorgiaItem AlphaFold 2 Monomer: Deployment in an HPC Environment(2022-09-29) Yang, Yuntao; Li, Zhao; Shih, David J. H.; Zheng, W. JimAlphaFold2, developed by Google DeepMind, is a breakthrough in the grand challenge of protein structure prediction. While the breakthrough will have profound impact on biomedical research, its application faces significant hurdles due to the computing intensive nature. We overcome this challenge by deploying the AlphaFold 2 pipeline in an HPC environment that fully utilized the computing resources and accelerated the workflow. Specifically, the CPU component of the AlphaFold 2 that includes multiple sequence alignment and template search was deployed on a computer cluster at the Texas Advanced Computing Center (TACC). The high performance of CPU cores and I/O requests on the cluster allowed us to complete over 200 jobs within 10 hours. The GPU component that includes model prediction and refinement was deployed on the latest Nvidia GPU server, and 200 jobs could be completed within 24 hours when 2 jobs run in parallel. The deployed workflow can efficiently use different computing environments to process many protein structure predictions to advance biomedical research.Item Discovering Pathophysiologic Networks of Temporal Lobe Epilepsy Using the BrainMap Community Portal through the Texas Advanced Computing Center(2022-09-29) Towne, Jonathan M.; Eslami, Vahid; Fox, P. Mickle; Cavazos, José E.; Fox, Peter T.Temporal lobe epilepsy (TLE) seizures cause regional damage, detectable on imaging of brain structure (VBM) and function (VBM). Damage is mediated by aberrant neuronal activity propagating along existing network architecture. TLE networks remain ill-defined and are of great interest to diagnostic and therapeutic development. Independent component analysis (ICA) can detect neural-networks by computing multi-variate co-occurrence patterns across a volume and is validated for coordinate-based meta-analysis (CBMA/Meta-ICA). Meta-ICA is methodologically distinct from mass-univariate meta-analytics (ALE) that simply detect robust regions/hubs of pathology. Although meta-ICA is typically used to extract canonical/healthy networks, we applied meta-ICA to VBM/VBP reports of TLE-pathology, to infer TLE-specific network anomalies. To identify TLE-networks, BrainMap Community Portal applications were used to access coordinate-results (Sleuth) of 74 experiments (n=1599), model coordinates as spatial probability distributions (weighted by sample-size) and apply ICA (Meta-ICA) at dimensions:d=1&2 (per sample-size restrictions), computing coordinate co-occurrence within and across experiments. TLE pathology-hubs were computed (GingerALE) separately as spatial convergence across studies (ALE), agnostic to within-study co-occurrence. Two anatomically distinct TLE-networks were identified (i.e. no overlap, excepting ALE hubs). Network-IC1 included brain-regions involved in language processing (speech-execution:Z=6.95; speech-cognition:Z=4.10) at d=2, with similar results (spatial-correlation:R=0.89) at d=1. Network-IC2 included brain-regions involved in emotion (reward:Z=4.76) and cognition (attention:Z=4.02; memory:Z=3.80). Meta-ICA implicated a superset of hub regions from GingerALE (IC1-tonsil; IC2-pulvinar, caudate, superior-temporal; Both-hippocampus, MDN-thalamus), uniquely identifying anterior nucleus, insula, supramarginal, and pre/paracentral gyri. Neither network matched canonical networks. VBM/VBP distribution to ICs was homogenous (χ2:p=0.07). Meta-ICA networks align with TLE symptom profiles. IC1 (verbal/visual) disruption can impair communication and cause visual hallucinosis. IC2 (limbic) aligns with social-emotional deficits; dyscognitive seizures disrupt cognition (attention/memory), impair awareness, and induce postictal amnesia. These findings reveal two novel TLE-networks, debut the first low-d meta-ICA detection of disease-networks, and highlight a BrainMap Community Portal use-case with implications in biomarker development within/beyond epilepsy pathologies.Item Job Losses, Marriage Troubles and Rich Uncles: Foreclosure Prevention Policy when Borrowers Hold Private Information about their Financial Health(2022-09-29) Kytömaa, LauriMy dissertation studies foreclosure prevention in environments where borrowers have an incentive to appear distressed in order to receive mortgage reductions. Such behavior is possible when borrowers have knowledge about their abilities to repay debt that cannot be observed by their lenders. Using a sample of Fannie Mae loans originated in California between 2004 and 2007, I show that mortgage providers only offer debt relief when they are highly informed about borrower default probabilities. I then use the estimated model to explore the effects of the Federal Home Affordability Modification Program, which was launched in 2009 in response to the Great Recession. I find that subsidies offered to banks under the program were more effective at preventing foreclosures in loans originated earlier in the 2000’s, even though banks tended to be equally well informed about borrower financial health in all sample cohorts. The results suggest that government subsides decreased foreclosures by 7.2% for loans originated in 2004, but that this rate steadily declines to 1.0% for loans originated in 2007. I also find that the average subsidy expenditure per prevented foreclosure increased from $17,000 to $150,000 between my sample origination cohorts. Jointly, the results offer a comprehensive look at how borrower financial well- being and the behavior of financial institutions influence debt relief policy. This project benefits greatly from access to TACC resources. Estimation uses a maximum likelihood routine in which I solve for a high-dimensional grid that rationalizes bank behavior, and then match model-predicted loan outcome probabilities to data. Solving for bank behavior is conducive to parallel computing since the optimum can be computed independently for every set of inputs. Leveraging many nodes allows me to solve for optimal policy at 60.2 million input combinations in under an hour. All numerical maximization takes place using Python on the Stampede2 cluster.Item Modeling Developmental Signaling Proteins Using Alphafold2(2022-09-29) Dingal, P. C. Dave P.Vertebrate animal development is governed by the signaling proteins that pattern fields of embryonic cells and direct their intended tissue fate. Vg1 and Nodal are two such signaling proteins that induce the formation of mesoderm and endoderm tissues (e.g., muscle, bone, blood tissues). Vg1 and Nodal are members of the transforming growth factor-beta superfamily of proteins that have no existing crystal structures yet. We utilized Alphafold2, a deep learning algorithm, to computationally predict (1) the monomeric structures of Vg1, Nodal, their receptors, agonists and antagonists; and (2) the heteromeric structures that will indicate amino acid residues potentially involved in their interactions and downstream signaling. We will show several Alphafold2 models of the multi-protein complex of Vg1-Nodal bound to their receptors, which inspired multiple exciting hypotheses that we now plan to experimentally validate.Item Pore-scale Simulations of Multiphase Flow for CO2 Migration Through Saline Aquifers in the Capillary- Dominated Regime(2022-09-29) Larson, Richard; Bakhshian, Sahar; Hosseini, Seyyed A.Carbon capture and storage intends to inject anthropogenic carbon dioxide from large point sources into the geologic formations for emissions mitigation. In geological carbon sequestration, it is critical to understand the behavior of carbon dioxide as it displaces subsurface, resident fluids during storage to assure its safety and permanence. The multiphase nature of carbon dioxide displacing saline water over long-term periods of post-injection relies heavily on the buoyancy forces arising from the density contrast between CO2 and saline water and the capillary forces controlled by the pore geometry at the pore scale. The competition of those governing forces controls the field-scale migration and confinement of CO2 in reservoir formations. In this study, we use high fidelity pore-scale simulations of multiphase flow to investigate this phenomenon in porous geometries representative of sedimentary rock formations. A computational fluid dynamic technique known as the volume of fluid method is taken to model the buoyancy-driven flow of CO2 in subsurface reservoirs. There are many computational burdens in these simulations. High-resolution meshes with a high number of grid cells are required to capture the complexity of the pore morphology leading to a high computational burden when considering the number of necessary simulated timesteps. Even reducing the domain size of the models to an effective 2-dimensional (2-d) structure with an area less than 1 cm2, nearly prohibitive computational time is needed. Furthermore, to replicate the capillary dominated flow, low velocity values are needed. However, the slower the flowrate, the more an interfacial issue known as spurious currents occurs, leading to numerical instabilities. To handle this numerical issue, simulations with smaller timesteps are required, but in tandem with the slow velocities, computational expense is further compounded. The usage of TACC’s parallel processing resources along with optimization techniques facilitates solutions and results that inform us about the fundamental mechanics of this flow.Item Discovering Spatially Coherent Gene Modules from Spatial Transcriptomics Data(2022-09-29) Larina, Maria; Singh, Salvi; Samee, Md. Abul HassanSpatial transcriptomics (ST) is an emerging technology that quantifies gene expression at spatial resolution from intact tissue sections. Although ST is enabling unprecedented studies on spatial gene expression, it has posed new challenges to biological data science. A typical ST dataset contains information of ~20K genes from 50K-100K cells. It is challenging to design efficient and scalable algorithms that generate new biological insights from these datasets. Here we feature an efficient and scalable non-negative matrix factorization (NMF) algorithm for identifying “spatial gene modules” (spatial-gems), i.e., groups of genes that express at spatially adjacent locations, in ST data. Spatial-gems are fundamental aspects of multi-cellular organisms. NMF is suitable for this problem since, in theory, NMF can identify the “informative parts” constituting a dataset, e.g., lips and eyes in human facial images and spatial-gems in ST data. The basic NMF formulation, however, can give sub-optimal results for spatial datasets – it ignores spatial locations of data points and thus does not guarantee informative parts that are spatially coherent. Graph-regularized NMF (GNMF) overcomes this issue by constraining the informative parts to comprise spatially adjacent data points. We introduce three changes to tailor the state-of-the-art GNMF algorithm for ST data. First, we statistically determine the optimal number of spatial-gems in an ST dataset. Secondly, we introduce regularizations that minimize the number of genes common between spatial-gems. Finally, we leverage numerical libraries and efficient data structures to obtain a scalable implementation. We benchmarked our GNMF against alternative algorithms on a brain ST dataset. Our algorithm comprehensively charted the spatial-gems in this dataset with a 20x speedup in execution time, making this an attractive tool for large-scale ST consortia like HuBMAP (Human BioMolecular Atlas Program). This tool and our multifaceted approach to enhance efficiency and scalability will be of major interest to the broad userbase of TACC.Item An Automated MRI Analysis Tool to Measure the Tumor Volume and Assess the Treatment Response for Glioblastoma(2022-09-29) Kabir, Tanjida; Hsieh, Kang-Lin; Nunez-Rubiano, Luis; Cai, Yu; Hsu, Yu-Chun; Quintero, Juan C. Rodriguez; Arevalo, Octavio; Zhao, Kangyi; Zhang, Jackie Jiaqi; Zhu, Jay-Jiguang; Riascos, Roy F.; Jiang, Xiaoqian; Shams, ShayanGlioblastoma is the most common and aggressive grade IV glioma tumor. During surgical resection, complete tumor removal is impossible due to irregular structure and can be infiltrative into the adjacent brain tissue. Therefore, measuring residual tumor volumes and their detailed location becomes a vital predictor of patients’ survival. Furthermore, radiologists manually segment tumor regions from normal brain tissue and compare the pre-surgery and follow-up MRIs to conduct post-surgery evaluations. This process is time-consuming and operator-dependent due to postoperative changes and tumors' irregularity. Therefore, the overarching aim of this work is to design an artificial intelligence (AI) framework to estimate the residual tumor volume considering the brain structural variations and assess the efficacy of the therapy utilizing imaging and clinical features. We used 419 pre-surgical and 310 follow-up segmented MRIs. All cases include four MRI modalities- T1, T1+Gd, T2, T2-FLAIR. Our study demonstrates that because of the morphological changes in the post-surgical brain, the state-of-the-art AI segmentation models trained on pre-surgery MRIs suffer from a 3-20% performance drop on follow-up MRIs. Besides, these models show a significant drop in generalizability and consistency on independent MRIs (p-value<0.05), indicating the necessity of training follow-up MRI-based segmentation models to assess the treatment response. We propose an encoder-decoder-based segmentation model where encoders utilize contrastive learning schemes to identify tumor location and shape by consolidating domain-specific and problem-specific features to reduce variation of the tumor’s irregularity. We replace the decoder’s deterministic layers with Bayesian layers to quantify the uncertainty in the model's prediction. The proposed model achieved >0.85 dice score on average for segmenting and measuring the volume of different tumor sub-regions- edema, enhancing, and non-enhancing regions. These areas’ prompt and accurate identification is crucial to measure residual disease, detect tumor recurrence, and identify treatment-associated side effects.Item Robust Software Vulnerability Detection Using Bayesian Gated Recurrent Unit(2022-09-29) Aminul, Orune; Dera, DimahSoftware systems are prone to source code vulnerabilities, resulting in deadlock, hacking, information leakage, and system failure. It is crucial to spot the security holes and identify the vulnerable software components prior to such encounters to prevent any security breach or system crash. While conventional machine learning and neural network-based vulnerability detection techniques have been reported in the literature, it is still challenging to determine the trustworthiness of these models. In this study, we develop a robust approach for detecting software vulnerabilities using a Bayesian gated recurrent unit. The proposed model identifies source code vulnerability while simultaneously learning the output prediction uncertainty. In the Bayesian framework, the Gaussian distribution is introduced as a prior distribution over the network parameters, which are then propagated through the network layers and activation functions. At the network output, the mean of the predictive distribution indicates the predicted vulnerability in source codes, while the covariance matrix reflects the uncertainty in predicted vulnerability. To evaluate the robustness of the model, we conducted rigorous experiments with 1.27 million source codes in C/C++. We reported the performance for five types of Common vulnerabilities encountered under different levels of Gaussian noise and adversarial attacks and compared the results with the state-of-the-art methods. The training and testing were done on the HPC resources in The Texas Advanced Computing Center (TACC), particularly on the lonestar6 supercomputer. This work is under the TACC project "TRUST- TRustworthy Uncertainty Propagation for Sequential Time-Series Analysis". By submitting multiple jobs to the queue, we could complete significant simulations within a reasonable period. The investigations have confirmed that the proposed model produces a significantly higher uncertainty when encountered with high levels of natural noise or adversarial attacks and could be used as a measure to alert human users in many high-stake applications.