A WORLDWIDE E-INFRASTRUCTURE FOR COMPUTATIONAL NEUROSCIENTISTS Innovative ICT Solutions for Multi-centric European Research Networks 2011-10-27 ITU Telecom World – WHO eHealth Pavilion Palexpo – Geneva, Switzerland David MANSET outGRID Technical Coordinator From the Internet to the Grid Internet Grid The worldwide network of networks... The environment for sharing computing resources... Web The protocol for sharing Information over the Internet... *Courtesy of EGI Project neuGRID e-Infrastructure 3 participating clinical centers: FBF (Brescia, Italy), VUmc (Amsterdam, The Netherlands), KI (Stockholm, Sweden), 600 CPU cores (+ EGI elastic resources), Unique bridge to transatlantic initiatives CBRAIN (Canada) and LONI (USA), Expands to support clinical decision making in neuro-degenerative diseases High-capacity Network Sim-e-Child e-Infrastructure Leverages on FP6-funded Health-e-Child project and excellence network, 5 participating clinical centers: OPBG(Rome, Italy), IGG (Genova, Italy), NECKER (Paris, France), GOSH (London, UK), JHU (Baltimore, USA) 300 CPU cores (+ EGI elastic resources), Unique bridge to transatlantic initiatives GenTAC and COAST clinical trials and databases (USA), High-capacity Network Disease Markers Development, Test and Validation neuGRID A Grid-brained e-Infrastructure for Neurosciences Imaging Markers for Alzheimer’s Gray Matter Loss Isolated Memory Problems Early Disability Consolidated Disability Recipe to Develop Markers for Alzheimer’s: 1. Large Databases (‘0,000s) … T0 US Australian Japanese European T6 T12 ADNI ADNI (AIBL) ADNI ADNI (IMI Pharma-cog) T18 ≈10,000 images ≈ 4,000 images (≈ 3,000 images) (≈ 1,500 images) … Recipe to Develop Markers for Alzheimer’s: 2. Sophisticated Algorithms for Image Analysis Cortical thickness at month 0 1.50 mm Cortical thickness at month 30 4.50 mm Difference analysis +.50 -.50 Recipe to Develop Markers for Alzheimer’s: 3.Sophisticated Statistical Models Gray matter change in Healthy elders Very early Alzheimer’s -3 | -2 | -1 | 0 | 1 | Years from dementia 2 | 3 TOMORROW neuGRID neuGRID Infrastructure Scalable Robust Distributed Grid SOA Workflow Provenance Pipeline neuGRID DACS are connected to GEANT2 Network neuGRID Infrastructure integrates EGI Grid resources LEVEL 0 Thousands of CPUs Petabytes of storage Data LORIS Grid Coordination Center 20 Mb/s Coordination Strong liaison with BiomedVO/LSVRC Center LEVEL 1 Slave LORIS DACS1 100 Mb/s USERS Slave LORIS DACS2 100 Mb/s Slave LORIS DACS3 1 Gb/s neuGRID Platform CoreLab •Data Acq. (Imaging + Clinical Variables) •Data Quality Control •Data Anonymization Pipelining •Pipeline Authoring •Pipeline Optimization •Pipeline Enactment •Multiple Toolkits (CIVET, BrainVISA, FreeSurfer, ITK/VTK, R etc) Collaboration •Document Sharing •Pipeline Sharing •Provenance Sharing Pipeline Output Visualization Augmented Reality Service LEVEL 0 PHP JSON MyProxy CAS Grid Coordination Service Service GW Center Auth Service GSI LORIS Data Coordination Center LEVEL 1 Augmented Reality Spot AR Service Slave LORIS DesktopFusion DACS2 Service Slave LORIS DACS3 Pipeline authoring and debugging AC/DC2 “Highway to Hell” - Data Challenge LEVEL 0 Scalable Robust Distributed Grid SOA Workflow Provenance Pipeline Fast transfers through GEANT2 US-ADNI Database GCC US-ADNI Dataset • 6’235 MRData Scans LORIS • 70 GB Coordination Center LEVEL 1 Slave LORIS DACS1 CE WN1 WN2 WN3 WNn Slave LORIS Slave LORIS DACS3 DACS2 SE CIVET Pipelines Execution AC/DC2 “Highway to Hell” - Data Challenge EGEE’09 Best Live Demonstration Award! Expected Results Experiment duration on the Grid < 2 Weeks Experiment duration on single computer Analyzed data Patients MR Scans Images Voxels Total mining operations Max # of processing cores in parallel Number of countries involved Volume of output data produced > 5 Years 715 6’235 ~1’300’000 ~9’352’500’000 286’810 184 4 1 TB AC/DC3 “Thunderstruck” - Data Challenge Alzheimer’s Disease Neuroimaging Initiative - The ADNI is a multisite, multiyear program which began in October 2004, - More than 800 subjects recruited, more than 7’500 MR scans - CIVET, FreeSurfer and BrainVISA pipelines executed over the 7’500 MR scans, - Exercised 3 DACS sites and the EGEE Seed Resources Expected Results AC/DC3 “Thunderstruck” - Data Challenge (3/3) Experiment duration on the Grid < 3 Months Experiment duration on single computer > 100 Years Analyzed data Patients MR Scans Images Voxels Total mining operations Max # of processing cores in parallel Number of countries involved Volume of output data produced 800 7’500 ~1’563’750 ~10’791’500’000 ~700’000 1’300 6 2.2 TB Research Infrastructures Interoperability In A Nutshell To promote interoperability among three e-infrastructures for computational neuroscience to converge into one unique worldwide facility Home to US-ADNI/IDA LONI Pipeline Cluster Facility CIVET Pipeline HPC Facilities NIH-funded http://www.loni.ucla.edu/ outGRID Facts - FP7-funded - Started 11/09 - 24 Months http://cbrain.mcgill.ca/ Access to large image datasets (future home to EU-ADNI) Image Analysis Pipelines Grid Facilities http://www.neuGRID.eu/ Developing Interoperability Current HPC Integration Syntactic Interoperability Syntactic Interoperability Semantic Interoperability (~13,500 cores 1,000,000+ core hours) JUROPA – Julich (26304 cores) CLUMEQ – U. Laval (7600 cores) Westgrid – UBC (3090 cores) RQCHP - U.Sherbrooke (2464 cores) SHARCNET - UWO (3082 cores) • • Dedicated HPC cluster f acility at UCLA/LONI, 306-node, dual-processor SUN Microsy stems V20z cluster. Each V20z node has dual 64-bit 2.4 GHz AMD Opteron 250 CPUs with 8 GB of memory 64-node Dell development cluster , w ith each node using dual 64-bit 3.6 G Hz Intel EM 64T pr o cesso rs and 4 G B o f m em ory 64-pr ocessor SGI Or igin 3800 SMP super co mputer w ith 32 G B of m em ory • Dedicated Grid nodes combined with public computing resources from EGI • EGI LSGC VRC • 10’000 CPU Cores LSGC • • Semantic Interoperability **SciNET - UoT (30240 cores) CLUMEQ - McGill (350 - 20000? cores) Sunday, March 14, 2010 • Shared HPC resources from public facilities • 7 HPC Centers sharing computing resources, 6 across Canada and 1 in Europe • 13’500 CPU Cores Interoperability Thinking Framework • Focus on interoperability across DCIs and user communities • Formulation of general interoperability guidelines • Syntactic • • • • Communication protocols/standards Security Networks Semantic • • Workflow standards/technologies Provenance, Data and Meta-data standards/technologies Interoperability Impact Preliminary Impact: unified space for developing new biomarkers Current HPC Integration Algorithms/Pipelines (~13,500 cores 1,000,000+ core hours) JUROPA – Julich (26304 cores) CLUMEQ – U. Laval (7600 cores) Westgrid – UBC (3090 cores) Data RQCHP - U.Sherbrooke (2464 cores) SHARCNET - UWO (3082 cores) **SciNET - UoT (30240 cores) CLUMEQ - McGill (350 - 20000? cores) Sunday, March 14, 2010 • • Dedicated HPC cluster f acility at UCLA/LONI, 306-node, dual-processor SUN Microsy stems V20z cluster. Each V20z node has dual 64-bit 2.4 GHz AMD Opteron 250 CPUs with 8 GB of memory 64-node Dell development cluster , w ith each node using dual 64-bit 3.6 G Hz Intel EM 64T pr o cesso rs and 4 G B o f m em ory 64-pr ocessor SGI Or igin 3800 SMP super co mputer w ith 32 G B of m em ory • • Dedicated Grid nodes combined with public computing resources from EGI • EGI LSGC VRC • Shared HPC resources from public facilities • 7 HPC Centers sharing computing Frisoni, Manset, et 6al., resources, across Canada and 1 in Europe 10’000 CPU Cores Nature Reviews •Neurology 2011 13’500 CPU Cores LSGC • • in press Super Infrastructure Continental Access for Communities CBRAIN oGEP LONI oGEP McGILL neuGRID oGEP UCLA FBF outGRID LInked Neuroscientific Grand chAllenge (LINGA) Scientific Objective DATA: ≅ 2’000 subjects acquired in different time points at different ages neuGRID AD – MCI - CTR ICBM CTR LONI ANM CBRAIN T12 T24 T36 T48 missing Longitudinal CIVET/CLASP CORTICAL THICKNESS ADNI T6 AD – MCI - CTR Baseline LINGA – LInked Neuroscientific Grand chAllenge Technical Objective International Web Portal Shared Workflow Authoring Space - Objective: Validating interoperability demonstrator and architecture, - “1 variable” approach, i.e. “same application with different data”, - Developing a “super workflow” involving all 3 infrastructures, - CIVET Pipeline to be run in all 3 infrastructures, - Different datasets to be processed in each infrastructure - CBRAIN: ICBM - LONI: US-ADNI - neuGRID: EU-ADNI AddNeuroMed Grand challenge in numbers… CIVET @ CBRAIN CIVET @ LONI CIVET @ neuGRID 11’000 MR Scans 100’000 CPU Hours Post-processing Post-processing 1.5 TB of Scientific Data 2 weeks runtime Quality Control & Statistical Analysis neuGRID LONI CBRAIN Interoperability Demonstrator In Collaboration with LINGA – LInked Neuroscientific Grand chAllenge Political Objective International Web Portal Shared Workflow Authoring Space - Objective: Interconnecting major international/national initiatives and having them collaborating towards one common goal - Neuroscience Infrastructures - CBRAIN CA - LONI US - neuGRID EU - NeuroLOG FR - DCI Infrastructures - HPC - EGI - SHIWA CA EU EU CIVET @ CBRAIN CIVET @ LONI Post-processing CIVET @ neuGRID Post-processing - Research Infrastructures - GéANT EU - CAnet CA - Internet2.0 US Quality Control & Statistical Analysis neuGRID LONI CBRAIN Interoperability Demonstrator LINGA Challenge EGI’11 Best Live Demonstration Award! MetaWorkflow Management Sub-Workflows 17/09/11 SSC & Virtual Laboratory Users Pipelines & Algorithms Analyses New Datasets Markers FBF VUmc N4U Tier 1 GRID KI LSVRC N4U - IS N4U Science Gateway NSPIN UNIG E GéANT N4U Tier 2 Grid Coordination Center HPC GBRAIN LONI CAnet Data Coordination Center INTERNET2 N4U Tier 3 CLOUD From Markers to Diagnostic Tools Production quality e-Infrastructure to service Diagnostic Tools • • • • • • • • • • • • Interoperability • LifeRay-based Web portal • • Diagnostic services as Portlets User-friendly access • • • Simple login + password • Single Sign On + Shibboleth Multiple IdP security architecture Robot-certificate mapping DIAGNOSIS OF ALZHEIMER’S AT THE MCI STAGE Structure: Gray matter volumetry Metabolism: FDG PET Alzheimer CSF total tau Amyloid deposits: PET Biochemistry: CSF tau/Abeta Normal CSF Abeta42 Neural activity. EEG From Research Infrastructures to Translational Medicine Sim-e-Child Grid-Enabled Platform for Simulations in Paediatric Cardiology Toward the Personalized Virtual Child Heart – ITU Telecom World 2011 – 36 October 2011 ?? Sim-e-Child Sim-e-Child First Trans-Atlantic platform towards personalized & predictive modeling of congenital heart disease • • • 37 FP7 STREP – Virtual Physiological Human (VPH) Objective – Building upon FP6 IP Health-e-Child – January 2010 to June 2012 Technical focus – Computational modeling & simulation of cardiovascular anatomy, function and hemodynamics – Grid-enabled infrastructure for data management & distributed high-performance computation Clinical focus – Model validation – Multi-scale quantification of disease extend – Planning, simulation & assessment of efficacy and safety of therapies October 2011 ?? Simulation Platform Sim-e-Child Web-Portal Grid-Enabled, Distributed Data Management & High-Performance Computing Local Simulation Application in Hospital http://sec-portal.maatg.fr 38 October 2011 ?? Sim-e-Child Network Sim-e-Child Health-e-Child Sim-e-Child COAST and genTAC (>1000 patients) HeC and OPBG (>1200 patients) Siemens SCR TUM JHU MAAT ACC OPBG Lynkeus US Grid Resource Point 39 EU Grid Resource Point October 2011 ?? Simulation Platform: Web-Portal Sim-e-Child Information System PHP 1. Create and export proxy to MyProxy (done only once) JSON MyProxy CAS Grid Coordination Service Center Service GW Auth Service 2. Standard CAS Authentication through Web portal 40 GSI 3. Access all facilities connected to the SSO October 2011 ?? General Modelling Approach Sim-e-Child Image Data (CT, MRI, US) Anatomical Model Fitting and Clinical Parameter Estimation Haemodynamics Simulation *Courtesy of SIEMENS 41 October 2011 ?? High-level seminar at the International Telecommunication Union (ITU) next Feb 2012 • To start forming the global structure • To discuss coordinated calls between funding agencies • To define seed initiatives in low-income countries GLOBIOS project preparation to structure the global initiative Conclusion