An e-Infrastructure for Collaborative Research in Human Embryo Development

0.83 MB·English

by  Adam Barker

An e-Infrastructure for Collaborative Research in
Human Embryo Development
Adam Barker∗, Jano I. van Hemert†, Richard A. Baldock‡ and Malcolm P. Atkinson†
.
∗Department of Engineering Science, University of Oxford, UK.
†National e-Science Centre, School of Informatics, University of Edinburgh, UK.
‡Human Genetics Unit, Medical Research Council. UK
9 Abstract—Within the context of the EU Design Study Devel- of work packages that span two major disciplines: biology
0 opmental Gene Expression Map, we identify a set of challenges and informatics. Gene expression studies on a small-scale
0 when facilitating collaborative research on early human embryo
are common place, the first work package investigates the
2 development. These challenges bring forth requirements, for
difficulties of automating systematic gene expression studies
whichwehaveidentifiedsolutionsandtechnology.Wesummarise
n
our solutions and demonstrate how they integrate to form an e- on a large-scale. An ethics study evaluates the inconsistent
a
J infrastructure to support collaborative research in this area of world-wide regulations and legislation on the use of human
developmental biology. embryos. A feasibility study focuses on a user requirements
5
1 Index Terms—Human embryo development, e-Science, e- of any resulting framework. The final work package and the
Infrastructure provision, In-situ gene expression studies
focal point of this paper focuses on designing a suitable

] e-infrastructure to support distributed laboratories. This
C I. INTRODUCTION
paper provides an overview of all involved solutions and
D
Characterising gene expression patterns is a crucial part technologies, it does not aim to give complete details of
s. of understanding the molecular determinants of embryonic each solution;instead referencesto complimentarypapers are
c developmentand the role of genes in disease. However, gene provided throughoutthe text.
[
expression studies in human embryos need to overcome a
1 number of difficulties. These include the sourcing and main- II. DGEMAP CHALLENGES
v tenance of collections of human material suitable for gene Establishing and maintaining suitable collections of human
0
expression studies, bridging the expertise in both biological developmentalmaterialrequirestime,significantresourcesand
1
and informatics areas, and amassing and making accessible knowledgethatasinglelaboratorywillfinddifficulttosustain.
3
2 data from multiple laboratories and studies. These difficulties may be overcome by linking multiple labo-
. Embryonic gene expression studies are common place in ratories and individual experts in the field, providing a single
1
0 model organisms such as drosophila (fly), mus musculus pointofaccess.Aframeworkthatfacilitatessuchcollaborative
9 (mouse)anddaniorerio(zebrafish),whereproceduressuchas research brings a number of key challenges:
0 in-situ hybridisationallow scientists to detect and localise the
: • Distributed laboratories Researchers distributed over
v presence of RNA transcriptomes that are likely to correspond
various physical locations around the world should be able
i togeneactivity.Thistypeofstudyislesscommoninhumans,
X to formulate and collaborate on research projects that involve
and currently only two laboratories are licensed in the UK
r wet-lab experiments on human embryo tissue. This requires
a to collect and analyse human embryonic tissue: the Human dealingwithprovidingsecureaccessofresearcherstoselected
DevelopmentalBiology Resource (HDBR) at Newcastle Uni-
subsets of several tissue banks, the experimental data and
versity and the MRC Fetal Tissue Bank (FTB) at University
results. In the DGEMap project this is dealt with by a Web
College London (UCL) [7]. Human embryonic material is
Portal,whichisdiscussedindetailinSectionIII-A.Moreover,
extremely scarce world-wide, dissemination of the results to
ethical legislation differs over countries, which can in many
the wider community is therefore vital for the progression of
locations make embryonic tissue unavailable, or even make
research.
experiments on human embryonic tissue illegal. In close col-
Developmental Gene Expression Map (DGEMap)
laboration with Policy, Ethics and Life Sciences at Newcastle
(http://www.dgemap.org/) is a EU-funded Design Study,
University, the project has evaluated the status of regulations
which will accelerate an integrated European approach to
and legislation on the use of human developmental material
gene expression in early human development. It runs from
and surveys professional standards [11].
April 2005 until March 2009. Due to the scarce nature
of human tissue, it is the first project of its kind in the • Distributed data handling Wet-lab experiments will be
world. DGEMap was granted e2.2 million and involves performedatseverallocations,theresultsofwhicharemostly
collaborations between the Institute of Human Genetics at stored at the site where the experiments were performed.
Newcastle University, and the National e-Science Centre at When researchers are querying, processing or analysing data
the University of Edinburgh. DGEMap comprises a number these data need to be transferred to other locations, either
for processing or for visualisation. To cope with the large The digital images are then aligned or mapped on to a 3D
quantity and highly complex research data any resulting e- reference embryo model of the same developmental stage.
infrastructure needs to provide transparentmechanisms to the Thereafter, we can query expression patterns from multiple
underlying workflows as well as allow for large amounts of genes, which allows us to do large scale analyses and data
data to be moved efficiently between sites. Issues regarding miningto extractknowledgeandconsequentlysynthetise new
the transportation and processing of large amounts of mostly hypotheses.
image data are discussed in Section III-B and Section III-C.
• Coping with a scarce resource Human embryo tissue WEB-BASED PORTAL
is a scarce resource, which makes finding techniques that IN-SILICO WET LAB DATA CURATION
maximise the results from each section of tissue a priority.
One of the objectives of the DGEMap project is to optimise
the physical use of tissue through new wet-lab techniques
Embryo Section ISH Curate
and more reliable instruments such as automatic systems for
immunohistochemistry and in situ hybridisation. Also, after DATA MANIPULATION
a small set of results from these instruments are captured
as images, we can apply image processing techniques to
accuratelypredictresults.Adetaileddescriptionofthisprocess
is provided in Section III-C. Data Mining Visualisation Workflows Mapping
• Few existing results on human development Where
Fig.1. Flowoftasksinageneexpressionstudyfromanembryothroughto
humanembryotissue is a scarce resource,with asa resultnot
in-situexperimentstomappingresultstoembryomodelstodatamanipulation
much data and information about the developmental process- throughworkflow,visualisationanddataminingtechniques.Thearchedarrow
ingintermsofgenetics;muchisknownaboutthedevelopment illustrates how the results of data manipulation tasks lead to hypothesis
generation which influences future wet-lab experimentation. All tasks are
in, for example, mouse models. By linking the results from
accessible via a Web browser through the DGEMap Web portal. Note that
DGEMap to results in external data sources, we can aid theimages shownhereareforillustrative purposes.
researchers in finding directions for future studies. Details
aboutthepossibilitiesforlinkingonvariouslevelsisprovided Due to the scarce nature of human embryonic material
in Section III-D. When these links are provided as services, it is essential to allow uniform access to these data to
they can be used by researchers to compose workflows that remote researchers. With this requirement in mind, a Web-
extractand processdata and informationto go back and forth based portal delivers a single entry point for EU research
between the mouse and human embryo databases. enabling collaborative in-situ experiments on human embryo
•Largeamountsofmouse-modelresultsWhendatafrom tissue. The federated approach allows researchers to query
tissue banks in multiple distributed sites, at the same time.
the human embryo is linked with data and information from
Workflows within the portal are data-centric, optimisation is
the mouse, we end up with a very rich resource. The disad-
therefore vital when moving large data for analysis. Image
vantageisthatitwillbecomemoredifficulttoextractrelevant
capturing techniques provide mechanisms to automatically
and significant ideas for future studies. It is vital for an e-
digitise embryonic samples and capture relevant meta-data,
infrastructurewithaccesstolargequantitiesofdatatoprovide
allowingexternalscientists tosearch,analyseandverifythese
methodsto enablehypothesisgeneration.In SectionIII-E, we
experimental data. After mapping 2D histological sections to
discuss in detail two methodologies from the field of data
standard3Dspatio-temporalmodels,dataminingisperformed
mining that directly operate on the spatio-temporal nature of
tofind“interesting”interactionpatternsbetweensetsofgenes.
gene expression patterns.
The resulting DGEMap e-infrastructureprovidesa unifiedso-
III. E-INFRASTRUCTURE lution forhumanembryoresearch, all aspectsof functionality
Thee-infrastructuremustsupportthetasksinvolvedinhigh are provided through a Web-based portal, making it possible
resolution in situ gene expression studies. In Figure 1 we to access data and results through a Web browser. In the
show the basic flow of tasks in such a typical study. First, following sections, we will discuss each componentin detail.
an embryo is frozen or embedded in a block of wax after
A. DGEMap Web Portal
which it can be cut into sections, which are all placed on
slides. These sectionswillundergoeither in-situhybridisation DGEMap provides a single point of access to multiple
or immunohistochemistry, which are histological techniques distributedgeneexpressiondatabases,thisisachievedthrough
to reveal by colourmetric, radioactive or fluorescent stains the use of portal technology, in particular a portal implemen-
the distribution of gene expression products. These products tation provided through the open-source Gridsphere project
are either the transcription product mRNA or the resultant (www.gridsphere.org/). Portals provide access to groups of
protein and are the evidence for gene-expressionin any given diverse Web sources and allow users to customise the way in
cell. The sections are viewed using microscopy and the tissue whichthesedataareviewed.Portalimplementationsprovidea
structureand expressionpatternsdigitally capturedto images. componentmodelthatallowsautonomouscomponentsknown
Logout avoidingtheneed to passlargequantitiesof intermediatedata
Welcome,
admin through a centralised server.
The solution is best illustrated through an example, with
Welcome Administration Project Registration Embryo management Experiments Management Report Project Management webforum
Embryo Management referencetoFigure3,anumberofdistributedlaboratoriescon-
Embryo Management
ducting human embryonic gene expression studies offer data
embryo collection
via a Web service interface. A scientist needs to retrieve and
ENmo b ryo EDmisbrurypotion ETemxbtr y o Disruption TBFiiexmfaoetrieon Abnormal SacLength CStaargneeg i e CTeaxrnt e g ie Stage collatedatafromthesethreelaboratoriesandruntheresulting
1 Mdiisldrulypted 0 0 0.0 1 datasetthroughadataminingalgorithm.Thefirststepsinthe
fingers long, workflow(phases1–2inFigure3)involveretrievingdatafrom
2 Mdiisldrulypted buRrseaesiftnu /Ol;hK ehatd a nbosetnt ?! 50 0 31.0 21 tpttohoauuadnccshh ;n t&ooet cwsh itmehso &re the laboratories via Web service interfaces. However, instead
3 Intact ibnu ts anco (sUigSnS o"ft woitnhse"r, 45 0 27.0 16 froooutn dpeladt,e retinal ofcontactingtheservicedirectly,acallismadetoa proxy(in
sac etc) pigment
toes more than thefirstinstanceP-1)whichisinstalledas“near”aspossible
notches, scalp
4 Intact snleigchkt rfergaicotnure of 5 0 37.0 21 vfNianusgcceh ra<sl 2 bc/rl3eo bsos?n;,l y?, to enrolled Web services; by near we mean preferably on the
full abdo; cord
cyst same Web server or network domain, so that communication
5 Mdiisldrulypted lgionuwstee rlrot isloitmn wbisth g oconred and 15 0 34.0 16 hreatninda pl lpaitgem,ent between a proxy and a Web service takes place over a local
some bowel loops hands just about
network. The proxy invokes the service and the output, in
Add a new embryo
this case R-WS1 is passed back to the proxy, tagged with
a unique identifier and stored within the proxy. Instead of
Fig. 2. A screenshot of the DGEMap Web Portal, showing the embryo
the proxy returning these data to the orchestration engine a
collection available attheHDBRinNewcastle
referencetothedataisreturnedinstead,inthiscase,$R-WS1.
This process is repeated for WS-2 and WS-3, retrieving the
as portlets to be plugged into a portal infrastructure. Portlets
necessary data from each laboratory. The output from the
representmodular,reusable software componentsthat may be
Web service invocations, $R-WS1, $R-WS2, $R-WS3 are
developedindependentlyofthegeneralportalarchitectureand
needed as input to the Data-mining Web service (phases
offer a specific set of operations.
3–4). The orchestration engine instructs P-1 to move these
TheDGEMapportalservesasakeyresourceforresearchers
data to a proxy sitting near the Data-mining Web service,
involved at each stage in the gene expression study cycle
P-4.ThesedataareusedasinputtotheData-miningWeb
detailed in Figure 1, in particular human embryonic gene
serviceviatheproxy(phases5–6),theoutputofwhich,R-DM
expression studies. The portal captures the major tasks of
isstoredattheproxy,whichinturnreturnsareferencetothese
HDBR managers, staff, and resource users, and provides
data to the orchestration engine, $R-DM.
facilities for project management, embryo search, resource
The Circulate architecture allows control flow messages to
distributionandtrackingandefficientinformationsharingand
pass through the orchestration engine, while larger data flow
manipulation,throughworkflowtechnology(discussedinSec-
messages (containing embryo images etc.) to be exchanged
tion III-B), image manipulation and visualisation (discussed
between proxies in a peer-to-peer fashion. Details of the
in Section III-C) and data mining techniques (discussed in
API and implementation of the proxy can be found in a
Section III-E).
complementarypaper[4]. Performanceanalysis[3] concludes
thatasubstantialreductionincommunicationoverheadresults
B. Workflow Optimisation - The Circulate Architecture
in a 2–4 fold performance benefit across common workflow
Efficiently executing large-scale, data-intensive work- patterns, essential when dealing with the data sizes involved
flows [2] common to scientific applications (i.e. scenarios in DGEMap scenarios.
detailed by DGEMap) must take into accountthe volume and
C. Image Processing and Visualisation
pattern of communication. Standard workflow tools based on
a centralised orchestrationenginecan easily becomea perfor- In situ gene expression studies intrinsically make use of
mancebottleneckforsuchapplications,extracopiesofthedata image output and image processing forms an important role
(intermediate data) are sent that consume network bandwidth in the e-infrastructure. To analyse the gene-expression data it
and overwhelm the central engine. In order to address this is important to be able to compare expression domains from
problem the DGEMap infrastructure proposes the Circulate multiplegenesthatcouldbepartofthesamegeneticnetwork.
architecture, a novel solution based on centralised control Toenablethiscomparisonofseveralgeneexpressionpatterns,
flow, distributed data flow [8]. The Circulate architecture databases such as the Embryonic Atlas of the Developing
sits between pure orchestration (completely centralised) and Human Brain (EADHB) and the Edinburgh Mouse Atlas
pure choreography (completely decentralised). A centralised Project (EMAGE) [6] have developed techniques to spatially
orchestration engine issues control flow messages to services mapdatafromsectionor3Ddatadirectlyontostandardrefer-
takingpartintheworkflow,howeverenrolledservicescanpass encemodels. Themappingprocessconsistsof image warping
dataflowmessagesamongstthemselves,likeinapeer-to-peer fromtheexperimentalembryoontothereferencemodel.With
system. This model maintains the robustness and simplicity embryonicmaterial the requiredspatial transformationcan be
of centralised orchestration but facilities choreography by very complex requiring significant expert input and image
1 2 3
LAB-1 LAB-2 LAB-3 LAB-1 LAB-2 LAB-3 LAB-1 LAB-2 LAB-3
WS-1 WS-2 WS-3 WS-1 WS-2 WS-3 WS-1 WS-2 WS-3
invoke R-WS1 Repeat until stage
WS-3
P-1 Data-mining P-1 Data-mining P-1 dePli-v4e, r, P-4 Data-mining
invoke, WS1, $R-WS1 [$R-WS1,
[params]; $R-WS2,
$R-WS3];
Orch. Orch. Orch.
Engine Engine Engine
4 5 6
LAB-1 LAB-2 LAB-3 LAB-1 LAB-2 LAB-3 LAB-1 LAB-2 LAB-3
WS-1 WS-2 WS-3 WS-1 WS-2 WS-3 WS-1 WS-2 WS-3
invoke
P-1 ack:true P-4 Data-mining invoke[$, RD-aWtaS-m1ining, P-4 Data-mining P-4 Data-mining
ack:true $$RR--WWSS32]; R-DM
$R-DM
Orch. Orch. Orch.
Engine Engine Engine
Fig.3. Centralised controlflow,distributed dataflowworkflowarchitecture
processing. These techniques are under active research, the
challengefor the e-infrastructureis to providethe tools to the
user,capturetheresultanttransformationsandtotransformthe
image data.
One of the major challenges is to deal with the scarce
nature of the material. In the context of the e-infrastructure,
this has lead to new methods for extrapolating results from
sparsely sampled gene expression patterns. An example is
shown in Figure 4 where a small set of gene expression
patterns have been mapped on to a 3D embryo model of the
same developmentalstage as the embryofrom which sections
werederived.Afterthis,itispossibletopredicttheexpression
patternbetweenthetwooutersectionsbyextrapolationallthe
sections in between the area of the original sections.
After images resulting from in-situ studies are mapped to
embryo models, with optionally, missing sections predicted
using extrapolation, it is vital good visualisation techniques (a) Asetoforiginal sections projected ina3Dreference embryo
are available. These techniques should support the general
questionsresearchersaskaboutgeneexpressionpatterns.Such
questions include comparisons between the relative spatial
positionsofpatternsofvariousgenesandbetweentheshapeof
expression patterns in different embryos. We have developed
tools that allow researchers to make use of techniques that
support answering these type of questions.
D. Human-Mouse Comparison
Tounderstandhowgenesdeterminethehumandevelopmen-
talprocessitisnecessarytocapturedatafromhumanembryos.
There are many examples, especially in the brain where the
pattern of expression in say mouse is significantly different
to that in human for the homologous gene. Nevertheless the
number of genes that have mouse homologues between say
(b) Thepredicted expressionpattern basedontheoriginal sections
mouse andhumanis about98%andabout91%of the human
genome lies in conserved syntenic segments (with respect to Fig. 4. Extrapolating gene expression patterns to form a prediction of the
mouse). The implication is that the mouse can serve as a full3-dimensional expressionpattern
good model for understanding mammalian development and
directcomparisonbetweenmouseandhumanwillbeawayto We have developed a methodology to define a given tar-
establish probable gene function and gene-networkactivity in get pattern by the combination of multiple gene expression
thematchingspatio-temporalcontext.Hypothesesgeneratedin patterns via several gene interactions operations. The target
thiswayofcoursethenneedtobecheckedinthehuman.Itis pattern can be the expression pattern of a particular gene, a
very important therefore that there is efficient interoperability pattern defined by a human, a pattern defined by anatomical
between the human and mouse data-resources and that data components or by any other means of defining spatial area
from one can be visualised and analysed in the contextof the within the context of the model mouse embryo. The method
other. searches for a set of genes and combines their expression
The key property of the gene-expression data associated patternsusingpredefinedoperationstocloselymatchthetarget
with DGEMap is the spatial pattern at a given stage of devel- pattern,therebyattemptingtodefinethispatternspatially.The
opment.Forinteroperabilitywethereforeneedbothspatialand objective of this study is to measure the robustness of the
temporalmappingsbetween the human and mouse. This is in methodologyandvalidatethesignificanceoftheresultinggene
additionofcoursetothe“standard”bioinformaticsmappingof interactions. This is important as much noise exists in the
genes and sequences via sequence comparison and homology acquisition of the patterns as well as much inaccuracy may
identification. For this purpose we are therefore developing existinthetargetpattern.Anexampleofasignificantresultis
directspatialmappings(3Dwarptransformation)betweenthe providedin Figure5(b)and a more detailedstudy is available
standardreferencemodelsofthemouseandhumandatabases. in [13].
These have been tested in prototypeand have been integrated
in thecontextof thequeryandanalysistoolsavailablevia the IV. RELATEDWORK
DGEMap portal. There are several other projects which propose simi-
lar e-Infrastructure to support the work of research scien-
E. Data Mining on Gene Expression Patterns
tists in the laboratory, a summary of such systems is pre-
The increase in efficiency of in-situ hybridisation experi- sented in [1]. The most widely known are myExperiment
ments brings with it a significant increase in the amount of (http://www.myexperiment.org/) a social networking Website
data that needs to be analysed. For instance, the Emage Gene for scientists to upload and share scientific workflows and
Expression Database Repository [14] contains 29560 assays, their associated results and ViroLab (http://www.virolab.org/)
where each assay can hold up to 24 images. With such a a Web-based virtual laboratory framework for infectious dis-
rich resource on data, it is impossible for a researcher to find eases.
relationshipsbetweendataina structuredmanner.Instead,we Our e-Infrastructure provides a unified solution for hu-
needtosupportresearcherswithmethodsthatallowstructured man embryoresearch, accessible througha Web-based portal,
searches and which can extract significant relationships from supporting the research cycle of a scientist by providing
massive amounts of data. Data mining [10] provides several functionality to: share physical lab resources, execute, share
methodologiesforextractingknowledgefromdatabyordering and monitor workflows and results, image processing, visu-
data,buildingrulestodescribedataandbyorderingtheserules alisation and data mining services. In comparison existing
according to measured properties such as significance and solutions focus on a particular aspect of this research cycle,
interestingness. As most of the methodologies are designed e.g. myExperiment focuses primarily on sharing workflow
for traditional nominal data, these need to be adapted to suit definitions in order to encourage reuse.
the domain of gene expression patterns. Here we discuss two
of those adapted methods. V. DISCUSSION
We introduce a novel application of mining for association We have identified some of the key challenges in enabling
rules in results of in situ gene expression studies. In ear- in-situ studies of human embryonic tissue. These challenges
lier studies in which association rules were applied to gene include overcomingthe problemsassociated with laboratories
expression results, these results originated from micro-array that are geographically distributed over many countries; han-
experiments, where the aim is to find associations between dling of large quantities of experiment and result data from
genes[5]inthecontextofbroadtissuetypes.Hereincontrast, these laboratories; working with human embryonic tissue,
we will operate on accurate spatial regions with patterns which is very a scarce resource; dealing with a relatively
derived from in situ experiments. This type of accurate data small amount of results on human development; and coping
enablesustoextracttwotypesofinterestingassociationrules, with large quantities of results from in-situ experimentswhen
first we can extract the same type of relationships between comparing with mouse-models.
genes;anexampleofwhichisshowninFigure5(a).However, The Developmental Gene Expression Map project has per-
we can also extract rules expressed in the form of spatial formed a three year design study that focused on tackling
regions, thereby providing knowledge on how areas in an thesechallengesinthecontextoffeasibility,ethics,biological
embryo are linked spatially. The only other study in the aspects and electronic infrastructure. In this paper, we have
direction of spatial association rules the authors are aware of, focussedonthelatter,whichhasresultedinane-infrastructure
is solely based on synthetic data [9]. A more detailed study consisting of several components. These components, when
on our application is available in [12]. integrated, address the following important aspects of the
!"
#$%&'#('#$%)'"$*34'5+6%,-./0-12%2& 7!"
#$%&'#('#$%)'$*'+%,-./0-1268+ #$%&'#('#$%)'$*'+%,-./0-186
/39:;%% <9=&
(a) Spatial context of a strong association rule (b) Left,thetarget pattern, which consists ofstrongandmoderate expression ofthegeneMid1.Right,
Brap,Zfp354b⇒9830124H08Rik,whichhasasup- the best match with a similarity index of 0.753, which uses aspecific combination of the spatial gene
port of 6.0%, a confidence of 97.9% and a lift of expressionpatterns ofthegenesRnf34,Pax5andAnapc11
10.2
Fig.5. Examplesofresults fromapplyingnoveldataminingtechniques inthedomainofcurated spatio-temporal geneexpressionpatterns
challenges. A portal was designed to enable a single point [3] A.Barker, J.B.Weissman,andJ.vanHemert. EliminatingtheMiddle
of entry to data and methodologies, which allows researchers Man: Peer-to-Peer Dataflow. In HPDC ’08: Proceedings of the 17th
International SymposiumonHighPerformanceDistributedComputing,
from any geographical location to collaborate on human em-
pages55–64.ACM,June2008.
bryonic tissue by facilitating access to the several human em- [4] A. Barker, J. B. Weissman, and J. van Hemert. Orchestrating Data-
bryo resources, to results from experiments, and to advanced CentricWorkflows.InThe8thIEEEInternationalSymposiumonCluster
Computing and the Grid (CCGrid), pages 210–217. IEEE Computer
methodsforprocessingandvisualisingtheseresults.Newdata
Society, May2008.
transportationprotocolswere investigatedto allow processing [5] C. Becquet, S. Blachon, B. Jeudy, J. Boulicaut, and O. Gandrillon.
of large volumes of data over distributed sites using standard Strong-association-ruleminingforlarge-scalegene-expressiondataanal-
ysis:acasestudyonhumansagedata. GenomeBiology, 3(12),2002.
workflow tools. To make optimal use of scarce human em-
[6] J.Christiansen,Y.Yang,S.Venkataraman,L.Richardson,P.Stevenson,
bryonic tissue, new image and visualisation techniques were N.Burton,R.Baldock, andD.Davidson. Emage:aspatialdatabase of
developedthatallow missingdata to bepredicted.Spatialand gene expression patterns during mouse embryo development. Nucleic
AcidsResearch,34:637–641, 2006.
temporal mappings between human and mouse models were
[7] S.LindsayandA.Copp.Mrc-wellcometrusthumandevelopmentalbiol-
established to allow much for a much richer environment for ogyresource:enablingstudiesofhumandevelopmentalgeneexpression.
building new hypotheses.Last, to cope with the large volume TrendsGenet,21(11):568–590, Nov2005.
[8] D.Liu,K.H.Law,andG.Wiederhold. AnalysisofIntegrationModels
of data made available through these mappings, novel data
ofServiceComposition.InProceedingsofThirdInternationalWorkshop
miningtechniqueswereappliedtosynthesisenewhypotheses. onSoftwareandPerformance, pages158–165. ACMPress,2002.
Membersofthe DGEMapprojectareinvestigatingpossible [9] C.OrdonezandE.Omiecinski. Discoveringassociation rulesbasedon
imagecontent.InProceedingsoftheIEEEAdvancesinDigitalLibraries
ways of implementing the design on a larger scale over the
Conference (ADL’99),Baltimore, Maryland, 1999.
laboratoriesin Europethat collect humanembryonictissue as [10] P.-N.Tan, M. Steinbach, and V. Kumar. Introduction to Data Mining.
well as those that perform research on human development. Addison-Wesley, 2005.
[11] K.TaylorandS.Woods. Reportonthecomparative systematic review
oftheliteraturefromkeyscientific,medical,socialscienceandbioethics
VI. ACKNOWLEDGEMENTS journals. Technical report, Policy, Ethics andLifeSciences, Newcastle
University,2008.DevelopmentalGeneExpressionMap(EU/FP6Design
The e-infrastructure work package is part of an EU-funded
Study011993),Deliverable D3.1.
Design Study Developmental Gene Expression Map (contract [12] J.vanHemertandR.Baldock. Miningspatialgeneexpressiondatafor
number 011993; http://www.dgemap.org/), which is led by associationrules. InS.HochreiterandR.Wagner,editors,Proceedings
of the 1st International Conference on BioInformatics Research and
professor Susan Lindsay at Newcastle University, UK.
Development, Lecture Notes in Bioinformatics, pages 66–76. Springer
Verlag,2007.
REFERENCES [13] J. van Hemert and R. Baldock. Matching spatial regions with com-
binations of interacting gene expression patterns. In M. Elloumi,
[1] G.Aloisio,V.Breton,M.Mirto,A.Murli,andT.Solomonides. Special J. Ku¨ng, M. Linial, R. Murphy, K. Schneider, and C. Toma, editors,
section: Lifescience grids for biomedicine and bioinformatics. Future Proceedings of the 2nd International Conference on BioInformatics
Generation Computer Systems,23(3),2007. ResearchandDevelopment,volume13ofCommunicationsinComputer
[2] A. Barker and J. van Hemert. Scientific Workflow: A Survey and andInformationScience, pages 347–361.SpringerVerlag,2008.
Research Directions. In R. Wyrzykowski and et al., editors, Seventh [14] S. Venkataraman, P. Stevenson, Y. Yang, L. Richardson, N. Burton,
International Conference on Parallel Processing and Applied Mathe- T. P. Perry, P. Smith, R. A. Baldock, D. R. Davidson, and J. H.
matics,RevisedSelectedPapers,volume4967ofLNCS,pages746–753. Christiansen. Emage—edinburghmouseatlasofgeneexpression:2008
Springer, 2008. update. Nucleic AcidsResearch,36(D):860–865, Jan2008.