Serono Science

Scientific computingand high performance

applications

Research Computing in Serono

Hardware environment

High performance computing applications

Drug development pipeline

TargetTargetdiscoverydiscovery

TargetTargetvalidationvalidation

ScreeningScreening+ H2L+ H2L PCDPCD Phase I/IIPhase I/II Phase III/IVPhase III/IV

MarketingMarketing

Drug Development pipelineDrug Development pipeline

Proteomics

Genomics Chemistry

Human GeneticsBiostatistics

Transcriptomics

Mouse genetics

Cell biology Pharmacology

SciencesSciences

WGSuArrays

Protein arrays

siRNAcaliper

combichem

MS

Taqman

imaging

cellomics

x-ray RMN

Y2H HTS

TechnologiesTechnologies

genomes

Transcript mapSNPs

Protein structure Patient data

protein map

interactions

screening data

phenotypeimageshaplotype

Data typesData types

pathwaysStructure/activity

Research Computing Vision & Missions

Use in-silico technologies to help identify and progress therapeutic proteins and small molecules that will

successfully feed the development pipelineBy:

Research Knowledge Management Delivering across Research an integrated information environment that puts scientific data, information and knowledge at the scientist’s fingertips

Computational Life Science Developing cutting-edge scientific applications enabling in-silico drug discovery and driving Serono’s competitive advantage

Advanced Data Analysis Providing advanced and pervasive data analysis competencies to make sense of high-throughput and complex data.

Research IT environmentProviding the computing and communication infrastructure to deliver the vision

Research computing activities

1. Data processing Technology driven

2. Predictions and simulation

3. Data analysis Interpretation1

-1

4. Data management

Advanced Data Analysis - Issues

Amount of data

Data complexity

2000 2002 2004

High content cell assays, genomic sequence, QSAR

High density microarrays as a discovery platform

Genome scan data – 100’000 SNP’s, hundreds of patients, several diseases

Biomarkers identification through proteomics and trasncriptomics

Compendium, Virtual Combinatorial Library

Multidimensional decision making

2001-3

2004-7

Analysis cannot be performed in silos – we need information systems able to correlate data available from all sorts of experimental information (Genome scans, DGE, RNAi, Cell assays, proteomics, interactions, phenotypes)

Grid for the life sciences – differences

Physics Biology

Theory « complete »Inexistent or

imprecise

Level of abstraction (model)

Single Multiple

Volume Very high Low-medium

Data complexity

Low Very high

Sample: GP07127N01

Sample Set: PAT

Acquisition_18Project: 20944, 20945 Result Id: Date Acquired: 3/12/03 9:14:21 AM

Acq. Method Set: Med Throughput 8 min TFA

3/12/03 9:24:25 AM, 3/12/03 Date Processed:

processing Processing method:

MED 2 HPLC Sy stem Name:

Run Time: 8.0 MinutesVial: 1

Injection #: 1

Injection Volume: 10.00 ul

Injection Id: 20930

Channel Id 20931 Mobile_Phase H2O TFA 0.1%- ACN TFA 0.05% HPLC_column XTerra MSC8, 4.6x50mm,3.5 uM

Processed Channel Descr. PDA MaxPlot (230.0 nm to 400.0 nm)

AU

0.00

0.10

0.20

0.30

0.40

0.00

20.00

40.00

60.00

80.00

100.00

Minutes

1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00

3.40

4

3.96

0

5.33

8

1

2

3

RetentionTime (min)

Area % Area

3.40

3.96

5.34

11433

13936

2494928

0.45

0.55

98.99

Processed Channel Descr. PDA 254.0 nm

AU

0.00

0.05

0.10

0.15

0.20

0.00

20.00

40.00

60.00

80.00

100.00

Minutes

1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00

2.72

7

3.40

4

3.96

0

5.33

8

1

2

3

4

RetentionTime (min)

Area % Area

2.73

3.40

3.96

5.34

8860

7383

7794

1266728

0.69

0.57

0.60

98.14

1

2

SampleName Processed Channel Rt (min) Area %

GP07127N01

GP07127N01

PDA MaxPlot (230.0 nm to 400.0 nm)

PDA 254.0 nm

5.34

5.34

98.99

98.14

SampleName: GP07127N01

CorporateDatabase

Core

Generic End-user Access

LIMSQC

LIMSQC

LIMSQC

PublishDrill-down

IntegratedOracle-Based

Systems

User-friendly interfaces (Web based)

Specialized, complex power-user interfaces

E-notebook

in silico generation tools e.g. Text mining, Data Analysis

HPC Hardware environment

• SGI Origin 3900 64 proc (cc-numa), IRIX, 128 GB

• SGI Altix 3700 BX2 16 proc

• Timelogic Decypher FPGA bioinformatics accelerator x 4

• SGI Origin 3900 32 proc• Linux Xeon cluster, 50 proc

• 10 TB CXFS SAN (Geneva only)

Computational chemistry (docking, combichem, compendium, pharmacophore, structure resolution)

Bioinformatics (public domain tools, sequence databases, peptide identification, in-

silico modeling)

Blast, SW, profiles

Same as above, Boston, Paris

Distributed data storage

High performance computing applicationsin Serono today

• Large scale sequence to sequence comparisons• Genome wide analysis (microRNA, focused gene

prediction, gw profiles, etc.)• Sequence data base monitoring• Gene index and data mapping• Large scale proteomics (peptide identification)• Virtual screening• In-silico biology

Smart is better than More

In combinatorial chemistry design, one scaffold and 4 groups of 800 reagents each generate a library of 320 billions virtual compounds

Enumerated substracture search would take years of CPU time and 1 petabyte of storage

A proprietary non-enumerated search retrieves hits in just a few seconds

VirtualVirtualCombinatorialCombinatorialDatabaseDatabase

VirtualVirtualScreeningScreening

Fast pre-filtering of compounds reduces amount of compounds for time-consuming docking studies

Useful for new compound acquisition, known protein target structure, not for primary screen (replating)

Usual size of virtual screens in Serono: ~1000 compounds

Future grid applications

• Large scale in-silico modeling• Protein-protein interaction• QM-based, dynamic virtual screening • Data grids• Imaging

Past grid evaluations (corporate PC idle cycles)

• High deployment costs – IT resources• Concern about availability of PC resources – habits and

procedures• Foreseen replacement of desktop by even less available

laptops• Modification of software to run effectively on the grid• Previous studies show that a large corporate grid of

1000 desktops is not more efficient than a 64 proc dedicated cluster (Novartis)

The in-house idle-cycle grid model is not efficient

Issues in the pharma industry

• IP considerations• Competitive intelligence• Security policies• Obsession with proprietary data and know-how

Is the current model of « all in-house » sustainable?Distributed (grid-enabled) public domain bioinformatics services will anyway become pervasive and will superceed capabilities available in-house