Presentation by Sven Warris at SURFacademy Autumn School, Tools for Researchers, 2-4 november 2009, Leiden.

1. Dare to compute
Mastering bioGrid
Sven Warris
Instituut
Leiden,
November 3, 2009
voor Life Science & Technology

2. Overview
 Bioinformatics
 What is a grid?
 Use of grid technology in research @ Hanze
• miRNA prediction
• Interproscan
 Mastering bioGrid project
 GPGPU: emerging technology
 Current use of grid/GPGPU @ Hanze
• Dingo @ UNSW
• Mixed Cultures
 Future & PhD project
Instituut
voor Life Science & Technology

3. Bioinformatics
 Biology and computer science
 Basic: using software and computing power to
perform database searches
 Advanced:
•Development of software
•Creating large databases
 Why computers?
•Complex analysis
•Large scale: gigabytes per run
•Visualization
Instituut
voor Life Science & Technology

4. Grid infrastructure
 Many computers in a network
 Software for access and control
 Central manager to distribute work
 Easy access to idle computers
 No need to start several
computers by hand
 Fail safe
Instituut
voor Life Science & Technology

5. Grid infrastructure
 Condor software: easy to use
 Bioinformatics network:
• Linux
• NFS
• central user authentication
 Use of almost all standard software
• BLAST, meme, ClustalW
• Own applications / scripts
• Controlled from web server
Instituut
voor Life Science & Technology

6. Grid infrastructure
 Current setup:
 ~70 nodes
• many dual cores
• at least 1gb ram per core, usually 2gb
 Calculix: 8gb -> 16gb, 2x dual core
 GPU: 4gb, quadCore, GTX295 & 8800 GTX
 Already used for many years of CPU-time!
Instituut
voor Life Science & Technology

7. Grid technology & research
 Start in 2005
 Installation of Condor grid software
 Needed for high throughput computing
 Basic configuration: resource hogging!
• full use of claimed computer
• no user access possible anymore
 Used for BLAST/SW-type applications
 Only possible during weekends/holidays
 No students allowed (unstable configuration)
 However: new possibilities!
Instituut
voor Life Science & Technology

8. Grid technology & research
 Example: miRNAs
 small, interfering RNAs 50-300b
 distinct hairpin structure
 Bonnet et.a.: minimum free energy significantly lower than
random sequences (population)
 For prediction: 1 + 1000 MFE values
 Slow software and too many MFEs
 Precalculate on grid: many different 'populations'
• ~1,400,000 populations (= ~1,400,000,000 MFEs)
 Estimate population for candidate: single MFE needed
 Use grid for prediction as well!
Instituut
voor Life Science & Technology

9. Grid technology & research
 Example: interproscan
 Used for protein prediction/classification
 Large databases
 Slow software
 Students developed Java application Condor-
Interproscan
 Introduction to GPGPU: clawhmmer
Instituut
voor Life Science & Technology

10. From research to education
 Grid successful in research
 However: unstable, resource hogging configuration
 No educational materials present
 Only two lecturers familiar with technology
 Needed: time and resources to integrate research and
education
Mastering bioGrid Project
Instituut
voor Life Science & Technology

11. Mastering bioGrid
 Cooperation between Hanzehogeschool & Hogeschool
Arnhem Nijmegen
 Development of education materials
 Reconfiguration of grid setup
 Testing of configurations
 Setup of grid infrastructure @ HAN
 Research into possibilities GPGPU
 Education of lecturers
 Important input from students
 Budget: 160,000€, subsidized 100,000€:
SURF,Utrecht
Instituut
voor Life Science & Technology

12. Mastering bioGrid, results
 Small grid @ HAN
 Documentation for use Condor software @ Hanze
 Education materials
 Further development of Interproscan-Condor
 Lecturers, researchers and students consider it already
as a standard tool
 Students start using it in second year
 More research possible during student projects
 Very flexible grid infrastructure
 Students: don't use openGL, use CUDA
Instituut
voor Life Science & Technology

13. GPGPU
 General Purpose Graphics Processing Unit
 Highly parallel, data driven
 Graphics processing: many independent, simple but
mathematically intensive calculations
 Bioinformatics: HMM, Smith-Waterman, SNP-detection!
 OpenGL: very complex, not very suitable for GPGPU
 CUDA:
•NVIDIA GPU
•C-based, 'simple' API
 Compatible with openCL
 Cheap hardware!
 Redesign of algorithms needed
Instituut
voor Life Science & Technology

14. GPGPU
2003 2003 2004 2005 2006 2007 2008
Instituut
voor Life Science & Technology

15. GPGPU
Instituut
voor Life Science & Technology

16. Current research use
 Dingo SNP detection (together with UNSW, Sydney)
•~500,000 '454' reads
•Reference genome: Dog
•Find reads with a single SNP: SOAP software
•Find unique reads not in Dog SNP database
•Results: 8242 new SNPs
•Now: tests in lab for pure Dingo selection
 Mixed cultures
•Sample water/soil/etc
•Sequence all DNA
•Find organisms/species/etc
 Ongoing miRNA-research
Instituut
voor Life Science & Technology

17. Current educational use
 Several projects now use grid
 More and faster BLAST results: no more days waiting
 High throughput / high performance computing
•Theme 12 specialisation
•Focus on GPGPU, CUDA
•Very enthusiastic students
•Student from HAN for minor
•Huge PR appeal
(graphics cards are sexy)
Instituut
voor Life Science & Technology

18. Future & PhD
 RAAK Pro Project
•Started in September 2009
•1M Euro, 4 years, 0.6M Euro subsidized
•WUR, UMCG, UNSW, HAN, PRI, KeyGene
 More use of current grid infrastructure by others
 Life Science Grid (SARA)
 Larger grid @ HAN
 More grids @ Hanze
•School of ICT, Media
 Connect those grids
 Development of GPGPU-based software
 Use these and third party for research
Instituut
voor Life Science & Technology

19. Future?
2
Next and next generation sequencing
demand high throughput / high performance!
More and more other chip-based lab
technologies come available!
Research asks, we develop and teach
Instituut
voor Life Science & Technology

20. Questions and remarks
Instituut
voor Life Science & Technology