Texas Advanced Computing Center Research Articles

Data Integration and Data Engineering Techniques

The Keeneland Science Application Partnership (Keeneland SAp) is a National Science Foundation (NSF) funded project in its third year to enable short-term, on-demand, and large-scale computational capability to 21 scientific computing teams involved in computation- and data-intensive projects, for a combined total of 700,000 Service Units (SUs). A unique aspect of the K-SAP is its strong alignment with NSF-supported research and the level of engagement from both the National Institute for Computational Science (NICS) and the Texas Advanced Computing Center (TACC) that are partners in this K-SAP. This report outlines the technical plan of the K-SAP and discusses the experiences of the first two years. Data Integration is the process that takes data of various shapes and structures, from a wide range of sources and makes them available to users and processes. It is a result of advancements in technology and international regulations that result in large amounts of data being generated every day in various sectors. To gain useful insights from this data, it needs to be accessible, of high quality, relevant to an organization’s objectives, and capable of meeting an ever-increasing regulation and compliance load. Data Engineering is the first step in making data available to stakeholders. Data Engineering involves data integration, transforming the data, and loading the data into a database so the data can be managed by data professionals, business analysts, and data scientists.

Shear speed in Arctic ice and underwater acoustic reflection

The underwater reflected sound wave under a sheet of Arctic ice contains information about the ice including thickness and mechanical properties. The ideal case of a perfectly flat ice floe floating above the water may be modeled using the OASES wavenumber integration code. A spectrogram of the acoustic response contains features related to the compressional and shear waves. The resonance associated with the shear wave speed in the ice is particularly distinctive. In the real world, the ice-water interface is not perfectly smooth. The finite element code SPECFEM2D is used to simulate the response of ice with a rough ice–water interface. It is particularly well suited to this problem since it accounts for all orders of multiple reflections. It shows the effects of fine-scale roughness on the acoustic response. Depending on the severity of the roughness, it may enhance or diminish the acoustic features related to the properties of the ice. [Work supported by the U. S. Office of Naval Research, Code 32 Grant N00014-20-1-2041 (N. Chotiros). G. Bayrakci and A. Best were funded by the UK Defence and Security Accelerator, Grant ACC2016927. The Texas Advanced Computing Center provided the high-performance computing resources.]

Real-space solution to the electronic structure problem for nearly a million electrons.

We report a Kohn-Sham density functional theory calculation of a system with more than 200 000 atoms and 800 000 electrons using a real-space high-order finite-difference method to investigate the electronic structure of large spherical silicon nanoclusters. Our system of choice was a 20nm large spherical nanocluster with 202 617 silicon atoms and 13 836 hydrogen atoms used to passivate the dangling surface bonds. To speed up the convergence of the eigenspace, we utilized Chebyshev-filtered subspace iteration, and for sparse matrix-vector multiplications, we used blockwise Hilbert space-filling curves, implemented in the PARSEC code. For this calculation, we also replaced our orthonormalization + Rayleigh-Ritz step with a generalized eigenvalue problem step. We utilized all of the 8192 nodes (458 752 processors) on the Frontera machine at the Texas Advanced Computing Center. We achieved two Chebyshev-filtered subspace iterations, yielding a good approximation of the electronic density of states. Our work pushes the limits on the capabilities of the current electronic structure solvers to nearly 106 electrons and demonstrates the potential of the real-space approach to efficiently parallelize large calculations on modern high-performance computing platforms.

Solving the electronic structure problem for over 100 000 atoms in real space

Using a real-space high-order finite-difference approach, we investigate the electronic structure of large spherical silicon nanoclusters. Within Kohn-Sham density functional theory and using pseudopotentials, we report the self-consistent field convergence of a system with over 100 000 atoms: a ${\mathrm{Si}}_{107,641}{\mathrm{H}}_{9,084}$ nanocluster with a diameter of 16 nm. Our approach uses Chebyshev-filtered subspace iteration to speed up the convergence of the eigenspace, and blockwise Hilbert space-filling curves to speed up sparse matrix-vector multiplications, all of which are implemented in the parsec code. For the largest system, we utilized 2048 nodes (114 688 cores) on the Frontera machine in the Texas Advanced Computing Center. Our quantitative analysis of the electronic structure shows how it gradually approaches its bulk counterpart as a function of nanocluster size. The band gap is enlarged due to quantum confinement in nanoclusters, but decreases as the system size increases, as expected. Our work serves as a proof of concept for the capacity of the real-space approach in efficiently parallelizing very large calculations using high-performance computer platforms, which can straightforwardly be replicated in other systems with more than ${10}^{5}$ atoms.

EESD special issue: AI and data‐driven methods in earthquake engineering – (Part 1)

EESD special issue: AI and data‐driven methods in earthquake engineering – (Part 1)

Context-aware learning of hierarchies of low-fidelity models for multi-fidelity uncertainty quantification

Multi-fidelity Monte Carlo methods leverage low-fidelity and surrogate models for variance reduction to make tractable uncertainty quantification even when numerically simulating the physical systems of interest with high-fidelity models is computationally expensive. This work proposes a context-aware multi-fidelity Monte Carlo method that optimally balances the costs of training low-fidelity models with the costs of Monte Carlo sampling. It generalizes the previously developed context-aware bi-fidelity Monte Carlo method to hierarchies of multiple models and to more general types of low-fidelity models. When training low-fidelity models, the proposed approach takes into account the context in which the learned low-fidelity models will be used, namely for variance reduction in Monte Carlo estimation, which allows it to find optimal trade-offs between training and sampling to minimize upper bounds of the mean-squared errors of the estimators for given computational budgets. This is in stark contrast to traditional surrogate modeling and model reduction techniques that construct low-fidelity models with the primary goal of approximating well the high-fidelity model outputs and typically ignore the context in which the learned models will be used in upstream tasks. The proposed context-aware multi-fidelity Monte Carlo method applies to hierarchies of a wide range of types of low-fidelity models such as sparse-grid and deep-network models. Numerical experiments with the gyrokinetic simulation code Gene show speedups of up to two orders of magnitude compared to standard estimators when quantifying uncertainties in small-scale fluctuations in confined plasma in fusion reactors. This corresponds to a runtime reduction from 72 days to four hours on one node of the Lonestar6 supercomputer at the Texas Advanced Computing Center.

A suite of broadband physics‐based ground motion simulations for the Istanbul region

AbstractPhysics‐based earthquake ground motion simulations (GMS) have acquired significant growth over the last two decades, mainly due to the explosive developments of high‐performance computing techniques and resources. These techniques benefit high/medium seismicity regions such as the city of Istanbul, which presents insufficient historical ground motion data to properly estimate seismic hazard and risk. We circumvent this reality with the aid of the Texas Advanced Computing Center (TACC) facilities to perform a suite of 57 high‐fidelity broadband (8–12 Hz) large‐scale physics‐based GMS for a region in Istanbul, Turkey. This paper focuses on the details of simulated GMS: (i) validation of the GMS approach against recorded ground motions produced by the 2019 Silivri earthquake; (ii) characteristics of 57 different source models, which aim to consider the uncertainties of many fault rupture features, including the length and width, dip, strike, and rake angles of considered fault planes, as well as hypocenter locations and earthquake magnitudes ranging between 6.5 and 7.2; (iii) high‐resolution topography and bathymetry and seismic data that are incorporated into all GMS; (iv) simulation results, such as PGAs and PGVs versus and distances to fault ruptures (), of 2912 surface stations for all 57 GMS. More importantly, this research provides a massive database of displacement, velocity and acceleration time histories in all three directions over more than 20,000 stations at both surface and bedrock levels. Such site‐specific high‐density and ‐frequency simulated ground motions can notably contribute to the seismic risk assessment of this region and many other applications.

(Invited) Small Polarons in Metal-Oxide Based Solar Absorber Materials

The key materials for photo-electrochemical (PEC) processes are the photoelectrocatalysts that are required to absorb sunlight, form electron-hole pairs, and transport the charge carriers efficiently towards the electrolyte-semiconductor interface. Metal-oxides are thought to be the most stable materials under intense interfacial reactive conditions. However, the overall efficiencies of these metal-oxide photoelectrocatalysts are not as high as expected, even for those with near-suitable bandgap energies required for efficient solar energy absorptions. One of the main reasons is attributed to the poor transport properties in metal-oxides with strongly correlated electrons. Typically, strongly correlated systems exhibit polaron conduction via electron-phonon interactions. Polarons act like a bottleneck for photoconductions in such metal-oxides. Recently, there have been significant efforts to characterize polaronic states and their effects on transport properties in photoactive materials. First-principles theoretical and computational studies can reveal the electronic origin of the polaron formation and, therefore, can suggest how to improve overall transport properties in such photoactive materials. In this presentation, representative metal-oxides photo-catalysts will be considered to demonstrate the small polaron formation mechanism. The polaronic states in metal-oxides will be depicted from detailed electronic structure calculations by density functional theory (DFT) based methods. We will show how these polaronic states affect the PEC cell's transport properties and photo-voltages. We will also discuss how these computational results compare with the recent experimental outcome. Some part of the work was funded by National Science Foundation, USA. Computations were performed at the Texas Advanced Computing Center (TACC).

A DFT+U Study of the Electrochemical Oxidation of H2 and CO on SrLaFeO4-Δ

INTRODUCTION Solid oxide fuel cell (SOFC) anodes are commonly composed of cermet materials, typically Ni-YSZ. Despite success in electrochemical performance, cermet materials suffer from sulfur poisoning, coke formation under low operating temperatures, and metal coarsening. Due to success of material design focused on the cathode such as LSCF, single-phase mixed ionic-electronic conducting (MIEC) materials have displayed promising oxidation activity [1]. The Sr-based Ruddlesden-Popper (RP) family has found recent use as a base stoichiometry in anodes where experimentalists have used both A and B-site doping and nanoparticle exsolvation strategies in order to tune activity [2,3]. In order to understand the fundamental electrochemical activity, we want to elaborate on the mechanism of H2 and CO oxidation on the promising material SrLaFeO4- \U0001d6ff (SLF). In the present study, we used ab initio methods to model H2 and CO oxidation on two FeO2-based models of SLF with and without B-site doping with Co and Ni. COMPUTATIONAL DETAILS All calculations presented in this work were performed using the spin-polarized DFT+U method with periodic boundary conditions as implemented in the VASP 5.4.4 code. We choose the PBE functional to describe exchange and correlation effects. Dudarev’s approach for DFT+U calculations is used with a U-J value of 4.0, 3.32, and 6.2 eV, respectively. We used the projector-augmented wave (PAW) method to represent the inner core potentials. The kinetic energy cutoff was set for all calculations to 700 eV, and integration over the Brillouin zone was performed with 3 × 3 × 1 Monkhorst-Point k-point mesh and the Gaussian smearing method with a sigma value of 0.05 eV. Slab models utilized a 15 Å vacuum layer and the bottom two layers were fixed to bulk position values. The DFT-derived parameters were employed in a microkinetic model to calculate the surface coverage of all adsorbed species for a given reaction condition. RESULTS AND DISCUSSION To begin our study, we tested all non-symmetrical conformers of SLF for the lowest relative energy on a model 2 × 2 × 1 supercell, as displayed in Figure 1a. A total of 3 structures were tested – P4/mmm, P4/nmm, and I4mm. We confirm that the lowest energy structure (I4mm) is equivalent to the experimentally reported structure. Next, we develop two 1.5 × 1.5 × 1 non-identical (001) FeO2-based slab models – so-called FeO2-LaO and FeO2-SrO (Figure 2b). The oxidation mechanism of H2 suggests oxidation begins with a dissociative adsorption step. Subsequent steps include formation of surface H2O, a surface vacancy, and a subsurface vacancy. To complete the catalytic cycle, we computed the bulk vacancy migration of SLF and assumed fast cathode kinetics for the overall cell model. At an operating voltage of 0.7 V, we predict that the highest rate occurs on the FeO2-SrO surface. We predict that H2 dissociative adsorption step is the rate-determining process based on Campbell’s degree of rate control. With the application of a single B-site dopant (Co and Ni), Co and Ni both decrease the barrier of the H2 dissociative adsorption step. Ni causes a slight decrease in energy for the formation energy of H2O. The oxidation mechanism of CO suggests oxidation begins with adsorption of the CO molecule on a reduced Fe site. The CO migrates to a neighboring Fe-O complex in order to form surface CO2. Based on degree of rate control, CO2 formation is likely the rate-determining step for both surface models.In conclusion, this is one of the first mechanism investigations on the anodic oxidation reactions of H2 and CO on RP-based oxides. We find that H2 has promising catalytic performance similar to the experimental evidence. The inclusion of B-site dopants displays that catalytic performance can be tuned with the introduction of different transition metals. Future work will examine the role of exsolved particles for RP-based oxides. ACKNOWLEDGMENTS This work has been funded by the Division of Materials Research, a National Science Foundation organization under Award Number 1832809. Computing resources provided by the San Diego Supercomputer Center (SDSC) and Texas Advanced Computing Center (TACC) under XSEDE grant number TG-CTS090100 and USC High-Performance Computing clusters are gratefully acknowledged. REFERENCES L. Xu, Y. Yin, N. Wang, Z. Ma, Electrochem. Commun., 76, 51–54 (2017)N. Wu, W. Wang, Y. Zhong, G. Yang, J. Qu, Z. Shao, ChemElectroChem, 4, 2378–2384 (2017)H. Chong, H. Chen, Z. Shao, J. Shi, J. Bai, S. Li, J. Mater. Chem. A, 4, 13997–14007 (2016) FIGURE 1a: 2×2×1 supercell model of I4mm SrLaFeO4 FIGURE 1b: Slab models of (001) FeO2-LaO and FeO2-SrO Figure 1

Hyperparameter Tuning with High Performance Computing Machine Learning for Imbalanced Alzheimer's Disease Data.

Accurate detection is still a challenge in machine learning (ML) for Alzheimer's disease (AD). Class imbalance in imbalanced AD data is another big challenge for machine-learning algorithms working under the assumption that the data are evenly distributed within classes. Here, we present a hyperparameter tuning workflow with high-performance computing (HPC) for imbalanced data related to prevalent mild cognitive impairment (MCI) and AD in the Health and Aging Brain Study-Health Disparities (HABS-HD) project. We applied a single-node multicore parallel mode to hyperparameter tuning of gamma, cost, and class weight using a support vector machine (SVM) model with 10 times repeated fivefold cross-validation. We executed the hyperparameter tuning workflow with R's bigmemory, foreach, and doParallel packages on Texas Advanced Computing Center (TACC)'s Lonestar6 system. The computational time was dramatically reduced by up to 98.2% for the high-performance SVM hyperparameter tuning model, and the performance of cross-validation was also improved (the positive predictive value and the negative predictive value at base rate 12% were, respectively, 16.42% and 92.72%). Our results show that a single-node multicore parallel structure and high-performance SVM hyperparameter tuning model can deliver efficient and fast computation and achieve outstanding agility, simplicity, and productivity for imbalanced data in AD applications.

SciApps: An Automated Platform for Processing and Distribution of Plant Genomics Data.

SciApps is an open-source, web-based platform for processing, storing, visualizing, and distributing genomic data and analysis results. Built upon the Tapis (formerly Agave) platform, SciApps brings users TB-scale of data storage via CyVerse Data Store and over one million CPUs via the Extreme Science and Engineering Discovery Environment (XSEDE) resources at Texas Advanced Computing Center (TACC). SciApps provides users ways to chain individual jobs into automated and reproducible workflows in a distributed cloud and provides a management system for data, associated metadata, individual analysis jobs, and multi-step workflows. This chapter provides examples of how to (1) submitting, managing, constructing workflows, (2) using public workflows for Bulked Segregant Analysis (BSA), (3) constructing a Data Analysis Center (DAC), and Data Coordination Center (DCC) for the plant ENCODE project.

Immune Repertoire Analysis on High-Performance Computing Using VDJServer V1: A Method by the AIRR Community.

AIRR-seq data sets are usually large and require specialized analysis methods and software tools. A typical Illumina MiSeq sequencing run generates 20-30 million 2×300bp paired-end sequence reads, which roughly corresponds to 15GB of sequence data to be processed. Other platforms like NextSeq, which is useful in projects where the full V gene is not needed, create about 400million 2×150bp paired-end reads. Because of the size of the data sets, the analysis can be computationally expensive, particularly the early analysis steps like preprocessing and gene annotation that process the majority of the sequence data. A standard desktop PC may take 3-5days of constant processing for a single MiSeq run, so dedicated high-performance computational resources may be required.VDJServer provides free access to high-performance computing (HPC) at the Texas Advanced Computing Center (TACC) through a graphical user interface (Christley et al. Front Immunol 9:976, 2018). VDJServer is a cloud-based analysis portal for immune repertoire sequence data that provides access to a suite of tools for a complete analysis workflow, including modules for preprocessing and quality control of sequence reads, V(D)J gene assignment, repertoire characterization, and repertoire comparison. Furthermore, VDJServer has parallelized execution for tools such as IgBLAST, so more compute resources are utilized as the size of the input data grows. Analysis that takes days on a desktop PC might take only a few hours on VDJServer. VDJServer is a free, publicly available, and open-source licensed resource. Here, we describe the workflow for performing immune repertoire analysis on VDJServer's high-performance computing.

A DFT+U Study of the Oxygen Defect Formation in (Sr1-xPrx)2FeO4

INTRODUCTION Solid oxide fuel cell (SOFC) electrodes are commonly composed of cermet materials, a composite of ceramic and metal materials. Despite success in electrochemical performance, cermet materials suffer from sulfur poisoning at the triple-phase boundary (TPB), high operating temperatures, and metal coarsening. To mitigate these problems, single-phase mixed ionic-electronic conducting (MIEC) materials have been investigated as alternative electrode materials. One such class of materials is the Ruddlesden-Popper (RP) phase (An+1BnO3n+1), a perovskite composed of n layers of octahedra in a perovskite (ABO3) stack. The RP family of Srn+1FenO3n+1 has found recent use as a base stoichiometry in both cathodes1 and anodes2 with and without A-site trivalent dopants. In order to rationalize the design of RP electrodes, we elaborate on the relationship of MIEC properties as a function of A-site dopant concentration at the atomistic level. In the present study, we used ab initio methods to investigate the relationship between Pr3+ concentration, iron oxidation state, and charge compensation upon defect formation to better understand the role of the A-site on the ionic and electronic properties in (Sr1-xPrx)2FeO4± \U0001d6ff (x = 0, 0.125, 0.25, 0.375, 0.5). COMPUTATIONAL DETAILS All calculations presented in this work were performed using the spin-polarized DFT+U method with periodic boundary conditions as implemented in the VASP 5.4.4 code. We choose the PBE functional to describe exchange and correlation effects. Dudarev’s approach for DFT+U calculations is used to correct the inadequate description of localized 3d electrons on Fe and 4f electrons on Pr with a U-J value of 4.0 and 6.0 eV, respectively. We used the projector-augmented wave (PAW) method to represent the inner core potentials. The kinetic energy cutoff was set for all calculations to 750 eV, and integration over the Brillouin zone was performed with 5×5×3 Monkhorst-Point k-point mesh and the tetrahedron method with Blöchl corrections. RESULTS AND DISCUSSION To sample all possible structural conformations at a given dopant concentration, we tested all non-symmetrical conformers for the lowest relative energy on a model 2×2×1 supercell, as displayed in Figure 1a. A total of 7, 42, 123, and 181 structures were tested for each dopant configuration, respectively. We calculate metallic behavior for x = 0.125 and 0.25 and semi-conducting behavior for x = 0, 0.375 and 0.5, indicating electronic conductivity decreases upon increasing A-site dopant concentration. Figure 1b displays the relationship between the final optimized dopant configurations and defect stability. The interstitial formation energy decreases linearly with Pr concentration with a deviation at x = 0.125. This relationship is primarily a function of the neighboring cation elemental identity and the decreasing interatomic distance between the interstitial defect and the neighboring cations in the rocksalt regime. The vacancy formation energy appears constant with Pr except for a large energy penalty at x = 0.5. The origin of the harsh energy penalty is rooted in the over-reduction of Fe. Upon vacancy formation, the released electrons directly reduce the next-nearest Fe atoms from +3 to +2 as opposed to delocalizing upon the oxide lattice. Figure 1c displays the relationship between the degree of delocalization and the energy of vacancy formation. The degree of delocalization3 is a dimensionless quantity that describes the extent of charge delocalized upon the oxygen sublattice with a vacancy. For x = 0 to 0.375, vacancy formation delocalizes the released electrons onto the lattice as opposed to x = 0.5, where charge from the lattice directly reduces Fe. Covalence between Fe-O bonds is decreased upon the formation of Fe2+, causing a harsh penalty in vacancy formation.In conclusion, tuning MIEC behavior is a balance between reducing Fe and charge delocalization upon the oxide lattice. In terms of electronic conductivity, a small introduction of Pr yielded metallic behavior. In terms of ionic conductivity, a threshold value of Pr yields an unfavorable vacancy formation energy. Future studies will focus on the optimization of the catalytic activity by B-site doping. ACKNOWLEDGMENTS This work has been funded by the Division of Materials Research, a National Science Foundation organization under Award Number 1832809. Computing resources provided by the San Diego Supercomputer Center (SDSC) and Texas Advanced Computing Center (TACC) under XSEDE grant number TG-CTS090100 and USC High-Performance Computing clusters are gratefully acknowledged. REFERENCES T. Hong, M. Zhao, K. Brinkman, F. Chen, C. Xia, ACS Appl. Mater. Interfaces, 9, 8659–8668 (2017)C. Yang, Z. Yang, C. Jin, G. Xiao, F. Chen, Adv. Mater., 11, 1439-1443 (2012)A. M. Ritzmann, J. M. Dieterich, E. A. Carter, Phys. Chem. Chem. Phys., 17, 12260-12269 (2016) Figure 1

Large Scale Science on NSF's Frontera System

In 2019 the National Science Foundation in the United States began operation of its Frontera supercomputer at the Texas Advanced Computing Center, debuting at number 5 on the TOP500 list. Unlike more general purpose supercomputers, Frontera was designed for the focused community of users doing computational science at the largest scale. This special issue features articles from Frontera's science users spanning the sciences from climate change to tornadoes and organic solar cells. We also include two computational science articles highlighting the tools and techniques that are needed to enable users running single jobs beyond 8,000 nodes.

Preparing Frontera for Texascale Days

Recently, the Texas Advanced Computing Center at the University of Texas at Austin held its third Texascale Days event on Frontera. Frontera is the most powerful academic supercomputer in the U.S. This tightly connected system allows experienced users with sophisticated parallel applications to run across thousands of cores. Frontera offers a unique opportunity for National Science Foundation researchers. Texascale Days provides researchers the opportunity to execute half- to full-system runs. Events such as Texascale Days allow our colleagues to experiment outside their regular workflow and push their applications and our systems to the limit. The issues we encounter and overcome provide insight on how to better deploy our systems and support their research. These events are stepping stones to the science problems of tomorrow and the HPC resources required to investigate them. Here, we describe the effort required to offer the Texascale Days opportunity to our researchers.

Tapis v3 Streams API: Time‐series and data‐driven event support in science gateway infrastructure

SummaryThe explosion of IoT devices and sensors in recent years has led to a demand for efficiently storing, processing and analyzing time‐series data. Geoscience researchers use time‐series data stores such as Hydroserver, Virtual Observatory and Ecological Informatics System (VOEIS), and Cloud‐Hosted Real‐time Data Service (CHORDS). Many of these tools require a great deal of infrastructure to deploy and expertise to manage and scale. The Tapis framework, an NSF funded project, provides science as a service APIs to allow researchers to achieve faster scientific results, by eliminating the need to set up a complex infrastructure stack. The University of Hawai'i (UH) and Texas Advanced Computing Center (TACC) have collaborated to develop an open source Tapis Streams API that builds on the concepts of the CHORDS time series data service to support research. This new hosted service allows storing, processing, annotating, archiving, and querying time‐series data in the Tapis multi‐user and multi‐tenant collaborative platform. The Streams API provides a hosted production level middleware service that enables new data‐driven event workflows capabilities that may be leveraged by researchers and Tapis powered science gateways for handling spatially indexed time‐series datasets.

Revealing the mechanism of SARS-CoV-2 spike protein binding with ACE2.

A large population in the world has been infected by COVID-19. Understanding the mechanisms of Severe Acute Respiratory Syndrome CoronaVirus 2 (SARS-CoV-2) is important for management and treatment of the COVID-19. When it comes to the infection process, one of the most important proteins in SARS-CoV-2 is the spike (S) protein, which is able to bind to human Angiotensin-Converting Enzyme 2 (ACE2) and initializes the entry of the host cell. In this study, we implemented multi-scale computational approaches to study the electrostatic features of the interfaces of the SARS-CoV-2 S protein Receptor Binding Domain (RBD) and ACE2. The simulations and analyses were performed on high-performance computing resources in Texas Advanced Computing Center (TACC). Our study identified key residues on the SARS-CoV-2, which can be used as targets for future drug design. The results shed lights on future drug design and therapeutic targets for COVID-19.

Deep Learning‐Based Point‐Scanning Super‐Resolution Imaging

Point scanning imaging systems (e.g. scanning electron or laser scanning confocal microscopes) are perhaps the most widely used tools for high resolution cellular and tissue imaging. Like all other imaging modalities, the resolution, speed, sample preservation, and signal‐to‐noise ratio (SNR) of point scanning systems are difficult to optimize simultaneously. In particular, point scanning systems are uniquely constrained by an inverse relationship between imaging speed and pixel resolution. Here we show these limitations can be mitigated via the use of deep learning‐based super‐sampling of undersampled images acquired on a point‐scanning system, which we termed point‐scanning super‐resolution (PSSR) imaging. Oversampled, high SNR ground truth images acquired on scanning electron or Airyscan laser scanning confocal microscopes were ‘crappified’ to generate semi‐synthetic training data for PSSR models that were then used to restore real‐world undersampled images. Remarkably, our EM PSSR model could restore undersampled images acquired with different optics, detectors, samples, or sample preparation methods in other labs. PSSR enabled previously unattainable 2 nm resolution images with our serial block face scanning electron microscope system. For fluorescence, we show that undersampled confocal images combined with a multiframe PSSR model trained on Airyscan timelapses facilitates Airyscan‐equivalent spatial resolution and SNR with ~100x lower laser dose and 16x higher frame rates than corresponding high‐resolution acquisitions. In conclusion, PSSR facilitates point‐scanning image acquisition with otherwise unattainable resolution, speed, and sensitivity.Linjing Fang1Fred Monroe2Sammy Weiser Novak1Lyndsey Kirk3Cara R. Schiavon1Seungyoon B. Yu4Tong Zhang1Melissa Wu1Kyle Kastner5Yoshiyuki Kubota6Zhao Zhang7Gulcin Pekkurnaz4John Mendenhall3Kristen Harris3Jeremy Howard8Uri Manor11. Waitt Advanced Biophotonics Center, Salk Institute for Biological Studies, La Jolla, CA, USA2. Wicklow AI Medical Research Initiative, San Francisco, CA, USA3. Department of Neuroscience, Center for Learning and Memory, Institute for Neuroscience, University of Texas at Austin, Austin, TX, USA4. Neurobiology Section, Division of Biological Sciences, University of California San Diego, La Jolla, CA, USA5. Montreal Institute for Learning Algorithms, Université de Montréal, Canada6. Division of Cerebral Circuitry, National Institute for Physiological Sciences, Okazaki, 444‐8787 Japan7. Texas Advanced Computing Center, University of Texas at Austin, Austin, TX, USA8. Fast.AI, University of San Francisco Data Institute, San Francisco, CA, USAFigure 1

Machine Learning Prediction for Complete Response to Hypomethylating Agents with or without Additional Agents in Patients with Newly Diagnosed Myelodysplastic Syndrome

Machine Learning Prediction for Complete Response to Hypomethylating Agents with or without Additional Agents in Patients with Newly Diagnosed Myelodysplastic Syndrome

The Impact of Treatment Recommendation By Leukemia Artificial Intelligence Program (LEAP) on Survival in Patients with Chronic Myeloid Leukemia in Chronic Phase (CML-CP)

The Impact of Treatment Recommendation By Leukemia Artificial Intelligence Program (LEAP) on Survival in Patients with Chronic Myeloid Leukemia in Chronic Phase (CML-CP)