Nationale HPC pilotprojekter via DeiC

Få inspiration fra mere end 50 pilotprojekter der har kørt i perioden 2015-2019

DeiC kompetencecenter har bl.a. til opgave at hjælpe nye brugere godt i gang inden for High Performance Computing (HPC). Forskere har kunnet indstille et relevant projekt til at blive et nationalt eScience pilotprojekt ved et af DeiCs tre nationale HPC-centre, hhv. ABACUS2.0, Computerome eller Kulturarvsclusteret (KAC).

Som et nationalt pilotprojekt i DeiC Kompetencecenter er man udpeget som "HPC frontrunner" inden for sit forskningsfelt. Der sigtes efter at nå ud til nye grene i det kontinuerligt voksende HPC-miljø, og et særligt fokus har bl.a. været på hhv. humaniora og samfundsvidenskabelige projekter (Fig. 1). Pilotprojekterne favner bredt og inkluderer ligeledes fagområderne biologi, miljø, kemi og biokemi, medicin, materialer og ingeniørvidenskab, computer science og kunstig intelligens.

Indhold på siden

 

Hvordan har regnekraften været brugt?

Ialt 55 projekter har modtaget midler til at få regnekraft på en national supercomputer. Et budget på 3.000.000 DKr har været tildelt denne ordning i perioden 2015 til primo 2019. Samfundsvidenskab og humaniora er støttet med flest projekter (40%) efterfulgt af biologi (24%), medicin (18%) og Computer Science og AI (11%, Fig. 1). Projekterne har involveret 7 danske universiteter og fordelt sig med 21 projekter via ABACUS2.0 (Bred mht. fagområder), 24 projekter via Computerome (Life Science) og 10 projekter via Kulturarvsclusteret (Samfundsstudier og Humaniora, Fig. 2). Du kan finde en samlet oversigt over pilotprojekterne her.

Figur 1: Fordeling (%) af 55 HPC Pilotprojekter på fagområder.

Figur 2: Antal HPC Pilotprojekter fordelt på HPC-centre og universiteter i Danmark. "Andre" omfatter industri, sektorforsknings-institutioner samt hospitaler. Forkortelser: CBS, Copenhagen Business School; DTU, Danmarks Tekniske Universitet; ITU, IT-Universitetet i København; KU, Københavns Universitet; SDU, Syddansk Universitet; AAU, Aalborg Universitet; AU, Aarhus Universitet.

 

Bevillingspakker

Som nationalt pilotprojekt får du stillet gratis teknisk support og regnetid til rådighed på supercomputerne. Tilbudet afhænger af hvilket HPC-center der benyttes:

  • ABACUS2.0: Pilotprojektet får op til 10.000 timers gratis regnetid og tilhørende gratis teknisk support.
  • Computerome: Bevilling af op til 40 timers teknisk support og 10.000 "compute node hours".
  • Kulturarvsclusteret (KAC): Pakkeløsning med fuldt finansieret projekt op til 6 mdr. og 120 timers teknisk support.

 

Sådan søger du et pilotprojekt

Find mere information om hvordan du søger et HPC pilotprojekt:

 

Forskningsresultater

Den tildelte regnekraft fra et DeiC HPC Pilotprojekt har i adskillige tilfælde indgået i en videnskabelig publikation (Fig.3).

I listen over pilotprojekter finder du aktive links til mere information via projektets titel.

Mere inspiration til fagområder der har brugt DeiC national HPC findes også via den totale liste over publikationer.

Figur 3: DeiC HPC pilotprojekt status mht. udarbejdelse af peer-reviewed publikationer. Hver projekt ansvarlig er blevet spurgt om der er lavet videnskabelig publikation(er), hvori regnekraft fra pilotprojektet var inkluderet.

 

Oversigt over pilotprojekter

Her får du en oversigt over pilotprojekter inden for de forskellige forskningsområder:

 

 

 

 

 

 

 

Pilotprojekter inden for biologi

Projekt titel HPC-center Beskrivelse
Disease Detection in Mink Industry with Most Economic Impact Computerome The aim of the project is to create a better understanding of the disease with most economic impact in the mink industry. Methods for whole genome sequencing of the virus and phylogenetic analyses of these sequences are developed and optimised. The resulting knowledge will facilitate viral detection, typing, and outbreak investigations.
HLA Typing in the Genome Denmark Cohort Computerome The project is part of the Genome Denmark project for the creation of a Danish reference genome. The aim is to benchmark the capabilities of existing Human Leukocyte Antigen (HLA) typing methods with 150 individuals divided in 50 parents-offspring trios and to expand the information in HLA databases with HLA haplotypes representative of the Danish population.
Rnas Cclust Being Docker'Ized and Hosted on Computerome Cloud Computerome The project considers the secondary structure of RNA molecules, which is often crucial to their function. A programme, RNAscClust, is developed to identify similar RNA structures in large-scale data sets thus shedding light on potential functional relationships between RNAs.
Using Evolution to Predict Protein Structures Computerome The project takes a look at the evolution of protein structure using the backbone's dihedral angles. A large database of aligned protein structures (HOMSTRAD) is used to study the overall features of protein evolution in terms of the backbone dihedral angles and based on these features, suggest and evaluate probabilistic models that can capture protein evolution in a stochastic way.
Protein Prediction with LSTMs Computerome The project is about the Type 3 Secretion System (T3SS), which is used by many bacteria to inject proteins directly into other cells. The aim is to identify the signal sequences using state of the art machine learning algorithms. Specifically, the LSTM improvement on Recurrent Neural Nets will be applied along with Convolutional Neural Nets in architectures selected using cross-validation.
Conformational Changes Induced upon Post-Translational S-Nitrosylation of Cysteine in an Oncogenic Mitochondrial Chaperone Computerome The project uses state-of-the-art atomistic and explicit-solvent enhanced sampling techniques based on the metadynamics and parallel tempering algorithms to provide the first rationale of the effects induced by S-nitrosylation (SNO) of TRAP1. The aim is to investigate if and how cancer-mutations reshape the conformational landscape of TRAP1 in absence or presence of the SNO modification.
Human Past Populations Computerome The project revolves around DNA sequencing used for estimating biodiversity through metagenomic analyses of the biological material. Through de-novo assembly a consensus sequence for previously unsequenced species can be obtained. The core samples from different regions and timescales provide unprecedented insight and direct evidence for past human population migrations and admixture events.
Supercomputing enabling ancient population genomic history studies Computerome In the project, billions of DNA sequence reads are generated and transformed into biological knowledge through bioinformatic and statistical methods. An analysis of high coverage genomes extracted from Late Pleistocene and Early Holocene samples is carried out, that is instrumental in understanding the early peopling of the Americas and Siberia.
Sequencing Technology Enables New Applications in Non-Model Species Computerome The project focuses on full genome resequencing and reduced genome representation sequencing, that enable new applications in non-model species. The data analyses provide novel insights on evolution and demography in the studied species and contribute to securing sustainable exploitation of natural fisheries resources.
Exploring the Power and Limitations of NGS Data for Comparative Genomics Analysis Computerome The project intend to explore capture data from 46 species of hummingbirds, and assess the impact of different bioinformatic choices on both the extent of information that can be recovered from the sequence data, and the potential biases in the biological interpretation of the results produced.
Holofish: Use of Microbiome-Genome Co-Optimisation to Improve Gut-Health and Growth in Farmed Salmon Computerome The objective of the project is to combine genome, transcriptome, epigenome, metabolome, metagenome, and metatranscriptome data in a hologenomic framework to decipher interactions between the host genome and the microbiome in affecting health and growth in farmed salmon. The results will be used to optimize future fish feed strategies, by matching feed additives to the concrete genetic background of a salmon broodstock.
Genomic Analysis of DNA from Archived Shark Jaws Computerome The aim of the project is to apply cutting edge genomic tools to abundant samples of tiger sharks in order to describe historical changes in population distribution, abundance and evolutionary response to global change and exploitation. The project is expected to initiate and facilitate a wealth of spatiotemporal genomic research on archived and uniquely abundant fish samples.
Beluga Whale Genome Project Computerome The project revolves around sequencing the genomes of more than 230 individual belugas through next generation sequencing and novel genotype likelihood approaches. The aim is to study population structure and adaptation across the circumpolar species range. The insights will aid in the informed conservation and management of this iconic Arctic species.

Pilotprojekter inden for miljø

Projekt titel HPC-center Beskrivelse
Impact of Climate and CO2 on the Terrestrial Carbon Sink in the North ABACUS2.0 The aim of the project is to determine the uncertainties in sources and sinks of CO2 for the future. Simulations are applied to differentiate the effects of CO2 rise, temperature changes and precipitation changes. Results can be applied in an atmospheric transport model and in an Earth system model to provide a coherent representation of the carbon cycle, as well as its link to climate processes.

Pilotprojekter inden for kemi og biokemi

Projekt titel HPC-center Beskrivelse
Graph Canonicalization in Cheminformatcis ABACUS2.0 The project intends to improve the state-of-the-art approaches in graph canonicalization of chemical compounds. A generic framework is developed for implementing various versions graph canonicalization algorithms that allow for the inclusion of edge types or the order of incident edges. This order is essential in order to encode the spatial layout of atoms leading to stereoisomers, which are important in a biological setting. 
Virtual Screening to Identify Novel Receptor Ligands Computerome The computational drug design group is conducting virtual screening for novel ligands that can be used to characterise unknown receptor functions or serve as starting points for drug design. The orphan receptors are characterised in pharmacological experiments revealing new physiological signalling networks.

Pilotprojekter inden for medicin

Projekt titel HPC-center Beskrivelse
Automatic Detection and Classification of Cancer Tissue in Hematoxylin and Eosin Stained Slides Using Computer-Aided Diagnosis ABACUS2.0 The purpose of the project is to develop an automatic computer-aided diagnosis to detect and classify cancer tissue within tissue samples stained with hematoxylin and eosin (H&E) using medical image analysis. Based on the cellular structures within the H&E stains a pathologist has to manually make a cancer diagnosis or determine if a special stain has to be made to support the diagnosis.
Forøgelse af informationsoverførselshastigheden i ikke-invasive aynkrone brain-computer-interfaces, ved optimering af samspillet mellem antal og type kontrolsignaler og klassifikations hastighed ABACUS2.0 Projektets formål er at identificere udviklingspotentialer for brain computer interfaces (BCI) systemer og anvendelse af stimulation af patienten. Navnligt informations­overførselshastigheden og latenstiden, da disse to faktorer i vidt omfang afgør hvilke anvendelser systemer egner sig til.
STAGING; Sequence of Tumor And Germline DNA - Implications and National Guidelines Computerome STAGING aims to reduce childhood cancer mortality, morbidity, and total health costs by setting the stage for integrating extensive genomic profiling into future individualized treatment. A whole genome sequencing (WGS) is performed of host germline DNA, and targeted DNA sequencing of tumor and gut microbiomes. This is linked with deep phenotyping of cancer development, drug metabolism, treatment efficacy, and toxicity. 
Machine Learning Analysis of Single Cell Proteomics Data for Disease Characterization and Discovery Cancer Immunotherapy Targets Computerome The aim of this project is to perform comparative analysis of healthy and tumor cells, with the goal of discovering tumor defining features that could help distinguish disease cells and provide targets for chimeric antigen receptor T cell therapy. xisting machine learning methods for processing the raw mass cytometry data from the CyTOF HELIOS system are improved.
Supercomputere til Hvidovre Hospitals NIPT Center Computerome Hvidovre Hospitals NIPT Center arbejder med at optimere og udvikle Non-Invasive Prenatal Testing (NIPT) af gravide. Centeret anvender Massive Parallel Sequencing (MPS), hvor hele genomet hos gravide kvinder sekventeres med Next Generation Sequencing (NGS) teknologi.
Personalised Health Service That Follows a Subscription Model Where the Customers Can Track Their Gut Health over Time Computerome The project carries out gut microbiome analyses through DNA extraction, 16S rRNA sequencing and bioinformatics tools. The analyses are used to give tailored dietary and lifestyle information that can be practically applied to optimize daily lifestyle habits.
MINPLAN – En app til selvmordstruede Computerome Forskningsprojektet har til formål at undersøge om en app-baseret kriseplan til selvmordstruede er bedre til at mindske selvmordstanker end en kriseplan i papirformat.
Identifying Genetic Causes of Retinal Dystrophies Computerome The goal of the project is to find genetic causes of retinal dystrophies. The strategy is to use whole genome sequencing (WGS) of approximately 300 persons with no molecular genetic diagnosis. The findings can establish a molecular genetic diagnosis and contribute to the knowledge of the retinal genetic network. 
Identification of Disease-Causing Genetic Variants in Patients with Cancer or Immune Deficiency Computerome Center for Genomic Medicine at Rigshospitalet uses genomic information in disease research. The aim is to identify disease-causing genetic variants in patients with cancer or immune deficiency. The identification of variants helps the discovery of new disease determinants and improves diagnostics.
The Hematological Relapse Project (ProGen) Computerome The Hematological Relapse Project (ProGen) at Aalborg University Hospital investigates the mechanisms of treatment of hematological malignancies and how to design specific therapeutic strategies aimed at overcoming them. High throughput sequencing (HTS) of tumor DNA and RNA is used to identify possibly therapeutic targets in patients.

Pilotprojekter inden for materialer og ingeniørvidenskab

Projekt titel HPC-center Beskrivelse
Large Scale Topology Optimization of Bridge Girders in Cable Supported Bridges ABACUS2.0 The project will carry out a large scale topology optimization utilizing the PETSc parallel computing framework to study the optimal structure of bridge girders. A new design concept will be based on the results and a more detailed optimization carried out of the individual parts of the structure.

Pilotprojekter inden for Computer Science og AI

Projekt titel HPC-center Beskrivelse
Quantifying the Strength of Hash Functions ABACUS2.0 The project focuses on risk assessment of SHA-1 and testing of the hypothesis by Bruce Schneier. The aim is to attempt both a second preimage-attack and a collision attack on the SHA-1 algorithm, performed with the parallelization capabilities of CUDA cores.
Evolutionary Robotics and Embodied Cognition (Podcast) ABACUS2.0 The aim of the project is to use simulations to combine insights from robotics, artificial intelligence, and evolutionary biology to abstract and synthesize robotic artifacts that can adapt behaviorally and evolutionary to different environments.
Popular Parallel Programming (P3) ABACUS2.0 The goal of the project is to achieve automatic parallelization of dataflow programs for execution on modern shared-memory multicore computers, in standard laptop, desktop and server hardware. The core ideas are to view spreadsheets as a dataflow language and further improve compilation of dataflow languages to shared-memory multicore machines.
ITU NLP – Pilot Projects ABACUS2.0 ITU has four pilot projects using Natural Language Processing (NLP): Multilingual Stance Prediction, Named Entity Recognition, Dependency Parsing and Language Modelling for Danish Languages. The research areas include dependency parsing, social media, multilingual NLP, stance detection, fake news analysis and entity detection.
Using Deep Learning to Represent and Predict Outcomes from Behavioral, Clinical, and Molecular Level Data Computerome A deep learning approach to learn a new representation of data in a fixed dimension is applied in the project. It explores how this optimal new representation improves prediction of e.g. gene expression levels and phenotype outcomes, and how data from individuals can be combined to make predictions on stratified groups of individuals.
Sample Size Optimization for Catch Comparison Trials Computerome The project concerns fishing gear testing and data collection. The aim is to optimize the industry data collection process by determining the minimum number of fish needing to be sampled during commercial fishing. Stochastic data simulations, state of the art catch comparison and catch ratio analyses are used.

Pilotprojekter inden for sociale studier og humaniora

Projekt titel HPC-center Beskrivelse
CBS Shared ABACUS2.0 CBS has three pilot projects on (1) movement and the spread of citations, (2) geography, trade and wage inequality and (3) life-cycle housing, fertility and female labor supply choices. 
Conditional Simulated Methods of Moments Estimation ABACUS2.0 The aim of the project is to develop and to apply a ‘conditional simulated methods of moments’ estimation technique. It is applied to answer a well-known question in financial economics: Do companies invest in overly risky projects just prior to bankruptcy and can features of debt contracts prevent this value destroying behavior?
General Equilibrium on the Danish Housing Market ABACUS2.0 The project focuses on solving general equilibrium models for the Danish housing market. It will answer relevant policy questions about tax gains in the housing market and design of housing market policies to obtain macroeconomic stability.
Trading on Deviations in the Put-Call Parity and Momentum ABACUS2.0 The project seeks to investigate the informative trading content in options prices. It will investigate these prices through the deviations of the Put-Call Parity as a measure of price pressure on the underlying stock. 
Computational Creativity and Optimizing for Aesthetics ABACUS2.0 The IT University of Copenhagen’s Intermedia Lab is doing an aesthetic exploration of Neural Networks and Deep Learning. Tasks involve subjective optimizations and explorative qualification of trained models.
3d og digital fotogrammetri workshop ABACUS2.0 HUMlab og DeIC afholder en workshop for at skabe læring om digital fotogrammetri, supercomputing inden for forskning og formidling af kulturarv.
Automatic Generation of Rap Lyrics ABACUS2.0 The aim of the project is to automatically generate rap lyrics, using a dataset with over 30,000 rap songs. In rap lyrics generation, both hybrid and pure unsupervised approaches can be used, in addition to Generative Adversarial Networks and similar methods.
Symbolic Music Generation with Deep Learning Systems ABACUS2.0 The project involves using LSTMs with attention mechanism to construct a system capable of generating music in MIDI form. The argument is that the attention mechanism will provide better results, as it can learn to focus on the most informative parts.
[HUMlab] Photogrammetry Working Group ABACUS2.0 [HUMlab], Department of Cross-Cultural and Regional Studies, and SAXO Institute archaeology photogrammetry working group consists of both students and researchers. The idea is to build up competences and facilitate the use HPC in humanities.
Durable Consumption Patterns in Equilibrium ABACUS2.0 The project study focus on households changing consumption decisions in response to changes in unemployment risk. Empirical facts are related with models of microeconomic behavior.
Multivocal – The creation of Collaborative Synthetic Voices with Non-Singular Identities ABACUS2.0 This project is a continuation of an art-/research-project called multivocal. It aims to explore the aesthetics and politics of synthetic voices by creating an alternative synthetic voice, in that the voice is based on a multitude of people having different ages, genders and geographic origin.
Moesgaard 3D Arkæologi ABACUS2.0 Undersøgelse af dynamikken i implementering af billedbaseret 3D-optagelse ved hjælp af empiriske erfaringer fra feltarkæologer. Implementering af modeller, der tegner sig for det arkæologiske feltarbejdes tidsmæssige, rumlige og organisatoriske kompleksitet. Fotogrammetri er en metode sammensætning betinget af en række kontekstuelle forhold.
Probing a Nation's Web Domain - The Historical Development of the Danish Web (Podcast) KAC The project contains an analysis of the Danish web development from 2005 to 2014. It answers the question: How many websites are there in Denmark and what knowledge is there in the Danish web if you analyse up to 1 million web pages?
Probing a Nation’s Web Domain - The Historical Development of the Danish Web — Part 2 KAC This project is guided by the following research question: What has characterised the Danish web and its development from 2005 onwards? The aim is to develop procedures for extracting, preprocessing/cleaning and analysing hyperlinks and multimedia content.
Digital Footprints KAC The project investigates what data patterns are seen in social media data internationally and how these patterns divide in terms of demographic profiles, across countries and over time. 
Udviklingen af det danske sprog på de sociale medier KAC Projektet analyserer sprogbrugen i de sociale medier og undersøger hvilke sproglige konventioner og normer der udvikler sig i forskellige genrer. Der arbejdes kvantitativt med analyse af ordformer, brug af emotikoner, stavefejl og ordfrekvens samt kvalitativt med analyse af dialoger og kommentarer.
Lingvistisk opmærkning af gamle danske aviser med fokus på navne og steder KAC Projektet har som mål at fastsætte hvor godt man kan opmærke navne og steder i aviserne i Mediestream for perioden ca. 1872 til 1916, samt at gøre den bedst opnåede opmærkning tilgængelig via Mediestream.
DeepAnon: Deep Data Anonymization for Large-Scale Analysis of Text-Heavy Data (Podcast) KAC The project develops a novel automated approach, DeepAnon, to anonymization of unstructured text-heavy data. A generic technique is established based on machine learning, which learns the anonymization function from a set of domain-relevant training data.
Temporalities of the Danish Newspaper Sphere, 1749-1877 KAC The project focuses on mapping and analyzing how journalistic texts and formats historically have constituted interrelated temporalities. It investigates how the constitutions of temporality within print newspapers have developed over a long time-span in Denmark and how such developments can be related to broader societal and technological developments.
The Historical Development of Tracking and e-Commerce on the Danish Web KAC The purpose of the project is to map and analyse the historical development of tracking technologies (e.g. http and Flash cookies, beacons, fingerprinting, html web storage etc.) and shopping baskets (e-commerce) on the Danish web.
Developing Tools for Large-Scale Analysis of Danish Radio and Other Audio Media Archives KAC The project is in continuation of research in music radio from the FKK collective project RAMUND. The distribution of music and talk developed through the timespan, in addition to the distribution of gender in the talk and musical style and genre, is examined.
Controversial Healing: Making Sense of Medicinal Cannabis Debates KAC The project aims to track the debate about medicinal cannabis on the Danish web leading up to its introduction in 2018. It seeks to investigate assumptions by mapping the evolution of the online debate about medicinal cannabis.