Research Article DNAseq Workflow in a Diagnostic Context and an Example of a User Friendly ...

12 Pages·2015·3.18 MB·English

Hindawi Publishing Corporation
BioMed Research International
Volume 2015, Article ID 403497, 11 pages
http://dx.doi.org/10.1155/2015/403497
Research Article
DNAseq Workflow in a Diagnostic Context and
an Example of a User Friendly Implementation
BeatWolf,1,2PierreKuonen,1ThomasDandekar,2andDavidAtlan3
1UniversityofAppliedSciencesandArtsofWesternSwitzerland,Perolles80,1700Fribourg,Switzerland
2UniversityofWu¨rzburg,AmHubland,97074Wu¨rzburg,Germany
3PhenosystemsSA,137RuedeTubize,1440BraineleChateau,Belgium
CorrespondenceshouldbeaddressedtoBeatWolf;[email protected]
Received13March2015;Revised10May2015;Accepted18May2015
AcademicEditor:HongLu
Copyright©2015BeatWolfetal.ThisisanopenaccessarticledistributedundertheCreativeCommonsAttributionLicense,which
permitsunrestricteduse,distribution,andreproductioninanymedium,providedtheoriginalworkisproperlycited.
Overrecentyearsnextgenerationsequencing(NGS)technologiesevolvedfromcostlytoolsusedbyveryfew,toamuchmore
accessible and economically viable technology. Through this recently gained popularity, its use-cases expanded from research
environmentsintoclinicalsettings.Butthetechnicalknow-howandinfrastructurerequiredtoanalyzethedataremainanobstacle
for a wider adoption of this technology, especially in smaller laboratories. We present GensearchNGS, a commercial DNAseq
softwaresuitedistributedbyPhenosystemsSA.ThefocusofGensearchNGSistheoptimalusageofalreadyexistinginfrastructure,
whilekeepingitsusesimple.Thisisachievedthroughtheintegrationofexistingtoolsinacomprehensivesoftwareenvironment,as
wellascustomalgorithmsdevelopedwiththerestrictionsoflimitedinfrastructuresinmind.Thisincludesthepossibilitytoconnect
multiplecomputerstospeedupcomputingintensivepartsoftheanalysissuchassequencealignments.WepresentatypicalDNAseq
workflowforNGSdataanalysisandtheapproachGensearchNGStakestoimplementit.Thepresentedworkflowgoesfromraw
dataqualitycontroltothefinalvariantreport.Thisincludesfeaturessuchasgenepanelsandtheintegrationofonlinedatabases,
likeEnsemblforannotationsorCafeVariomeforvariantsharing.
1.Introduction the increasing number of documented variations in well-
known disease related genes, this type of analysis became
Nextgenerationsequencing(NGS)technologiessawarapid verycommon.Increasingly,othertypesofanalysesarealso
progressionoverrecentyearsandimprovedinmanyaspects used in diagnostics, such as the detection of structural
since their introduction. The data quality, the length of the variations,whichbecamepossiblewiththeincreasedquality
sequences, and the speed at which the data is generated of sequencing data. The sequencing data, which can come
improved massively all while decreasing the costs associ- from various sequencing technologies, can be either tar-
ated with the technology [1]. Cost reductions and quality geted, such as targeted gene panel sequencing and whole
improvements now allow the technology to be used in exomesequencing(WES),ornontargetedlikewholegenome
a diagnostic setting, replacing older technologies such as sequencing (WGS). Those different approaches to sequence
Sangersequencing[2].Anincreasingnumberoflaboratories the patient’s genome produce a varying amount of data [3]
are either considering or already implementing NGS for andaresuitedfordifferenttypesofanalyses.Thecomplexity
routinediagnosticsprocedures.Nextgenerationsequencing oftheanalysisincreaseswiththeamountofdatatoprocess,
has many use-cases in diagnostics and, depending on the bothfromacomputationalpointofviewandfromahuman
analysis, different types of information are searched for in resources point of view. The general tendency today is to
the sequencing data. A very common type of analysis is moveaway fromtargetedsequencingtowardsWGS,asthis
thesearchforSNPs(single-nucleotidepolymorphisms)and removes the need for targeted amplicon libraries, as well as
smallindels(insertionsanddeletions)inpatientDNA.With makingiteasiertoreanalyseasampleatalaterdatewithout
2 BioMedResearchInternational
the need to resequence it. This tendency to increase the positiononthereferencegenome.Whatthemostfittingposi-
amount of sequenced data creates increasing challenges for tionwilldependonthealgorithmandparametersused.This
smallerdiagnosticslaboratories.Theyoftendonothavethe process is specially complicated when aligning sequences
requiredcomputinginfrastructureortechnicalknow-howto fromrepeatedregionsorcontaininginsertionsanddeletions.
handle this type of data. This is especially true in emerging The alignment process can be done by using one of many
marketswhichdonotalwayshavetherequiredfinancialand availabletools[6].Manyofthosetoolsareactivelydeveloped
humanresourcesaswellastheextensiveknow-howrequired andpossesstheirindividualqualities,dependingontheuse-
tosetuptheirownanalysispipeline. case and sequence data to be aligned. Some of the more
Today many types of NGS data analysis exist, all of commonlyused tools for sequence alignment are BWA [7],
themansweringdifferentsortsofquestionsaboutasample. Bowtie 2 [8], or BLAST [9]. This list is not exhaustive and
One example being copy number variation (CNV) analy- furtherinformationonthesubjectcanbefoundin[6].
sis suited to replace traditional MLPA (multiplex ligation-
dependent probe amplification) analysis. Another one is de (C)Realignment.Inadditiontothealignment,toolsexistto
novo sequence alignment to detect large changes in the improve the quality of an existing alignment, for example,
genome, for example, in cancer patients. But ultimately the with local indel realignment. Tools like GATK [10] and the
detection of SNPs and small indels in a sample based on CompleteGenomicsrealigner[11]providethisfunctionality.
a standard reference sequence remains the most prevalent This and the previous Alignment analysis step are some of
typeofanalysis.Thisiswhy,inthecontextofthispaper,we the heaviest in terms of computing power and require a
limitourstudytothistypeofanalysis:thedetectionofSNPs good computing infrastructure to be performed in a short
and small indels in targeted or nontargeted resequencing. timeframe.Thisproblemisaccentuatedwiththeincreasing
Althoughthistypeofanalysiscanbeapproachedinvarious amount of produced raw data and expected qualities of the
ways,itcanbedescribedasasetofgenericstepsthatremain alignment,notablyforindels.
common to most approaches. The next chapter describes
thecommonlyusedstepsintheanalysisprocess,providing (D) QC. Another round of quality control is usually per-
examplesofexistingtoolsthatcanbeusedtoperformthem. formedafterthefinalalignmenthasbeencreated.Itisused
to verify that the sample has been sequenced correctly, in
NGS Data Analysis. This chapter describes a common NGS particularintheregionsofinterest,likethegenesassociated
data analysis workflow used in diagnostics. Figure 1 shows with the patients phenotype. Several tools allow the extrac-
an overview of the described workflow, using a Unified tionoftheinformationneededforthisstep,suchassamtools
ModelingLanguage(UML)diagram. [12] or bedtools 2 [13]. They allow the user to get metrics
suchasthenumberofalignedreadsandthecoverageofthe
regionsofinterestoutofthealignmentfile.Ifthequalityof
(A)QC/Filtering.Theanalysisstartswiththequalitycontrol
thealignmentisnotsufficientatthisstage,itcanbedecidedto
of the raw sequencing data to determine if the sequencing
resequencethesample.Thisstepofqualitycontroliscritical
processadherestothedefinedqualityrequirements.Different
in a diagnostic setting where it is necessary to show how
metrics, such as the amount of reads sequenced, the length
well the regions of interest for a specific sample have been
distribution of the sequenced reads, and their quality, are
covered.Itisespeciallyimportantforcaseswherenovariants
verifiedandcomparedtotheexpectedvaluesofthespecific
of interest can be found in the specified zones and thus a
sequencer technology that has been used. If the quality of
negativefindingsreportneedstobejustified.
thesequenceddatadoesnotmatchtherequiredquality,the
sampleisusuallyresequenced.Toverifythequalityoftheraw
(E)VariantCalling.Aftertherawsequencingdatahasbeen
data,onecommonlyusedtoolisFastQC[4],whichgenerates alignedanditsqualityverified,SNPsandindelscanbecalled.
a comprehensive list of statistics of the raw data. During Theprocessofvariantcallingcomparesthealignedsequences
this initial quality control stage, the raw sequencing data is tothereferencesequenceinordertocreatealistofvariants
oftenfiltered,discardingreadsthatdonotmatchthedefined present in the sample. This list can be filtered on different
quality criteria. This raw data filtering is commonly done criteria,suchasminimumcoverageatthevariantpositionor
withtheFASTXtoolkit[5],whichallowsremovingreadsthat thefrequencyofthevariantalleleinthesample.Commonly
donotmatchtherequiredquality,byremovinglowquality usedtoolsareGATK[10],samtools[12],andVarscan2[14].
readsorreadsthataretooshort.Simultaneouslytofilterthe Differentvariantcallersfollowdifferentunderlyingmodels,
raw data based on its quality, it is increasingly common to resulting in differences in the called variants [15]. Variant
preprocessthedata,forexample,toremovebarcodesattached callingusuallyyieldsalargeamountofvariants,evenwhen
tothesequencedreads.Barcodesareaddedtothesequenced restrictingthecallingtoonlyoutputvariantswithaminimum
reads when multiple samples are sequenced together and qualityandprobabilitynottobeasequencingartifact.
needtobeseparatedagainbeforetheanalysis.Thissplitting
basedonbarcodescanalsobedoneusingatoollikeFASTX. (F) Annotation. This is why in the next step the list of the
variantsisannotatedwithmoreinformationthatispertinent
(B) Alignment. The verified raw sequencing data is then to the analysis being done on the sample. This includes
aligned against the human reference genome. During this the detection of variants present in public databases, such
process every sequenced read is placed at the most fitting as dbSNP [16], or the information about which genes are
BioMedResearchInternational 3
Data preparation Data analysis Validation and reporting
Samtools,
(D) QC bedtools 2
FastQC,
(A) QC/filtering
FASTX
GATK,
(E) Variant calling samtools, (I) Sanger validation
Varscan2
BWA,
Bowtie2, dbSNP,
BLAST, (B) Alignment Phenomizer
and so forth (F) Annotation VEP, (J) Report generation
and so forth
Samtools
GATK, (G) Variant filtering
complete genomics (C) Realignment
realigner
(H) Visualization
IGV,
Tablet,
genome view
UCSC
browser, and so forth
Figure1:UMLactivitydiagramofaNGSdiagnosticsworkflowseparatedintodifferentsteps.Acollectionoftheavailabletoolsforevery
stepismentioned.
affected and their predicted effects on those genes. Various (I) Sanger Validation. Before submitting the final findings
tools exist to annotate variants, either locally or through a to the clinician, it is still common practice to validate
webservice.SomeofthemorepopularonesaretheVariant the identified variants through Sanger sequencing. While
Effect Predictor (VEP) from Ensembl [17] and Phenomizer future advances in NGS technologies might make this step
[18] that add clinically relevant information to the detected redundant,itisstillconsideredthegoldstandardforvariant
variants.Oncetheannotatedlistofvariantshasbeencreated, validation.
thephenotyperelevantvarianthastobeidentified.
(J)ReportGeneration.Aftervisuallyinspectingandvalidating
(G) Variant Filtering. Analysis specific filters allow the user the variants, the identified variants or the lack thereof are
to filter the number of variants down to a smaller number. documentedinafinalreportthatissentofftotheclinician
One example is to only keep variants affecting genes with toallowhimtotakedecisionsonthecase.
phenotypesmatchingtheonesofthepatientandthathavea The described process can be done by taking several
predictedconsequenceonthegenefunction.Theexacttype approaches. A common approach is to use a selection of
of filters to reduce the amount of candidate variants varies the previously described tools to perform the individual
from case to case, depending on the available information. analysis steps. The different tools are usually connected
The filtering is often done using bcftools, which is part together through a series of automated scripts, performing
of the samtools project [12] or vcftools [19]. If during this the required analysis. The creation of the resulting pipeline
step no clinically relevant variant related to the current is a complex technical task, as it requires insight in the
analysis can be found, depending on the analysis protocol, working of the different tools and the ability to chain them
the filters may be relaxed. This allows the user to include, together.Thisisanontrivialtask,asnotalltoolsfollowthe
for example, variants on genes where the relation with the same standards in terms of data file formats and handling.
currentphenotypeislesscertain,orvariantswithuncertain Nonetheless,thisisaverycommonapproachwhichresulted
consequencesonaknowndiseasegene. in a variety of custom made NGS analysis pipelines, either
privateorpublic[25].
(H)Visualization.Oncealistofcandidatevariantshasbeen Complete software packages like CLCBio [26],
identified,itiscommontovisuallyverifytheminagenome NextGENe[27],ortheweb-basedGalaxy[28]alsoexist.They
browser. The genome browser displays the sequenced reads allow the user to combine many of the mentioned analysis
around the variant and allows the user to verify the quality tools,withouthavingtoworrytoomuchonhowtoconnect
ofthedataaroundthatposition,allowingtheusertoidentify the different tools to perform the analysis. Those tools
anddiscardvariantsoriginatingfromsequencingerrors.This are focused on giving the user the possibility to automate
is done usingone of the variousgenome browsers installed the analysis process as much as possible, abstracting the
locally such as IGV [20], Tablet [21], or Genome view [22]. complexity of the analysis. GensearchNGS belongs to this
There also exist web-based genome browsers, such as the family of tools, with its own approach and focus on the
UCSC[23]orEnsembl[24]genomebrowsers. issuesofsmallerlaboratories.Itfocuseslessonthepossibility
4 BioMedResearchInternational
to generate a custom automated pipeline but more on an and favoring an interactive analysis approach instead of
intuitive interface to handle the data-files to perform the a fully automated analysis pipeline. The user interface is
analysisinapredefinedpipelineoptimisedfordiagnostics. focused on providingthe functionalityrequiredto perform
The features of GensearchNGS are limited to those thepreviouslydescribedanalysis,withoutoverwhelmingthe
neededtoperformthepresentedNGSdiagnosticsworkflow. userwithrarelyusedfeatures.
Thismeanstheoverallfeaturesetissmallerthanthatofother
complete software packages, especially CLCBio or Galaxy, 2.MaterialsandMethods
which both offer a wide variety of functionalities. Having
a limited feature set allows for a cleaner and more focused GensearchNGShasbeendevelopedasamultiplatformappli-
user interface, one of the major goals of GensearchNGS, cation,runningonWindows,Linux,andOSX,usingJava6+.
something that is highly appreciated by current users of Theapplicationisalocallyinstalledsoftwareandisnotweb
GensearchNGS. based. In order to perform NGS analysis steps, Gensearch-
Aswillbediscussedinthefollowingchapters,Gensearch- NGSincludesseveralexistingexternaltoolsthroughplugins.
NGS uses either external applications through plugins or Thosetoolsaremainlyusedduringthesequencealignment
customimplementationsofstandardalgorithmsforthedif- phase,givingtheuserthechoicebetweendifferentalignment
ferentanalysissteps.Theresultinganalysisqualityistherefore algorithms which may have specific advantages related to
comparable to the one observed when using those tools the data to be analysed. The external alignment algorithms
separatelybyhandorthroughasoftwarepackagelikeGalaxy. supported through plugins are BWA [7] and Bowtie 1 + 2
[8] as well as Stampy [29]. GensearchNGS can either use
locallyinstalledversionsofthoseexternaltoolsordownload
Approach.Theissuesfacedbysmalllaboratorieswhenusing
and update them automatically. If the tools are installed
the previously mentioned tools, which are often limited
by the user he can update them when desired; otherwise
in terms of computing and human resources, are various.
thesameupdatemechanismusedtoupdateGensearchNGS
Based on our experience, the biggest issue faced by smaller
itself is used to update the external tools. For sequencing
laboratoriesisthelackoftechnicalknow-howanddedicated
data,GensearchNGS supportsvariousinputandoutputfile
bioinformaticians. The usage and configuration of many of
formats. For raw sequencing data, the standard FASTQ
thedescribeddataanalysistoolsrequireadedicatedemployee
formataswellasFASTA,SFF,andFNAissupported.Allof
whichnotavailableineverylaboratory.Thisisespeciallytrue
themareconvertedtothestandardFASTQfileformatduring
inemergingmarketswheremanymonetaryrestrictionsarein
thedataimportphase.AsFASTQfilesrequireaqualityscore
place.TheapproachofsolutionslikeGalaxyandCLCBioto
foreverynucleotideinthesequencedreads,adefaultquality
thisproblemistoabstracttheusageoftheunderlyinganalysis
scoreisassumedforallthenucleotidesifthesourceformat,
toolsandprovideawayfortheusertoautomatetheanalysis
like FASTA, is missing this information. For aligned data,
process.Whiletheabstractionoftheunderlyingtoolsreduces
bothSAMandBAMfilesaresupported,whereasSAMfiles
thetechnicalcomplexityoftheanalysisprocess,thisisoften
are converted to BAM during import. BAM and SAM file
notenoughtomaketheanalysisprocessmoreaccessible.The
are accessed through the HTSJDK library. HTSJDK is an
citedtoolscanbeusedformanyothertypesofanalysesthan
open source Java library, developed as part of the samtools
theoneslookedatinthispaper,whichleadstoamuchmore
project[12].ItprovidesastandardizedAPItoaccessvarious
complexuserinterfaceandoftenoverwhelmstheuserwith
NGS data file formats. The library can be downloaded at
manyoptions.Thecitedtoolsalsofocusmoreontheability
http://samtools.github.io/htsjdk/. Annotations can be pro-
to create automated analysis pipelines, leading to very little
videdthroughtheBED,WIG,andGFFfilesandtheVCFfile
interactivitybetweentheuserandthedata.
formatisusedtoimportorexportvariants.
Thesecondissueoftenencounteredisthelackofasuitable To benefit from the knowledge stored in various online
computing infrastructure to perform the computationally databases, GensearchNGS has been connected to several of
intensive parts of the analysis. Sequence alignment in par- them to annotate the NGS data on various levels, from the
ticularisverydemandinginthatregard.Eventhoughmany genesandtheirtranscriptsdowntotheindividualvariants.
laboratoriesoutsourcethesequencingandalignmentprocess, For most of the annotation data, GensearchNGS accesses
the possibility to realign the raw sequencing data is often the Ensembl [24] web service through its biomart API,
required.Toperformthedescribedanalysisstepsandmore, givingtheapplicationaccesstogeneandvariantinformation.
GensearchNGS providesa complete software suite focusing Ensemblprovidesaunifiedwaytoaccessseveralotheronline
onoptimizedalgorithmsandusageofdistributedcomputing databases,likedbSNP[16]orOMIM[30],throughasingle
for optimal usage of the existing infrastructure. All the onlineservice.InadditiontoEnsembl,GensearchNGSinte-
analysisstepsmakeusageofmultiprocessorsystemstospeed gratesotheronlineservicesforadditionaldataandfeatures:
uptheanalysis,whereasonlysequencealignmentusingthe Cafe Variome [31] provides a standardized way to share
customGensearchNGSalignmentalgorithmusesdistributed variants between various laboratories, UCSC [32] provides
computing. The entire functionality is provided through accesstothehumanreferencesequence,andHPO[18]pro-
an intuitive user interface to lower the technical know- videsinformationaboutphenotypesassociatedwithgenes.In
how needed to perform NGS data analysis. GensearchNGS situations with a direct conflict in the information between
provides its own way on how the analysis is approached thedifferentdatabases,wherepossible,theinformationwith
by providing the user with a patient centric user interface the most severe clinical impact is displayed. An example of
BioMedResearchInternational 5
thisiscontradictorySIFT[33]andPolyPhen2[34]scoresfor
variants.Bydefaultthemostseverewillbeshowntotheuser,
withthepossibilitytomanuallygetmoreinformationwhen
requested. GensearchNGS also integrates the possibility to
talk to other software installed by the user, such as Alamut
[35], allowing the user to easily open a variant found by
GensearchNGSinsidealocalAlamutinstallation.
3.Results
This chapter discusses how GensearchNGS implements the
differentanalysisstepsandhowittriestolowerasmuchas
possiblethetechnicalknow-howandcomputinginfrastruc-
Figure2:MaininterfaceofGensearchNGS,showingapatientwith
turerequiredforNGSdataanalysis.
itsvalidatedvariants.
3.1.SoftwareSuite. GensearchNGSprovidestheuserwiththe
possibilitytocreatedifferentprojectstogrouptheanalysisof
variouspatientsbasedontheirtypeortheirrelation.While ACMG variant classification guidelines [36], with the user
many other NGS analysis tools provide the possibility to requestedcategoriesArtefact andFalsereferenceadded.The
organizethedatainprojects,GensearchNGSputsthenotion savedvariantsareusedforanyfuturevariantdetectioninnew
ofpatientsandtheirassociatedmetainformationinthecenter patients,immediatelytellingtheuserifavarianthasalready
of the project management. Every project contains one or beenfoundinadifferentproject.TheadditionoftheArtefact
morepatientandeachcanhaveseveralrawsequencingdata andFalsereferencecategoriesallowstheusertoquicklyfilter
associated.Therawsequencingdatacaninturnhaveseveral outcommonsequencingerrorsinhisparticularsequencing
alignments associated, for example, using different aligners setup.Thehierarchicalorganizationofthedataisintendedto
oralignmentconfigurations.Projectspecificsettingssuchas guidetheuserthroughtheanalysisprocesswhilekeepingit
specificregionsofintereststobeusedduringtheanalysiscan clear where the data comes from. An example of this is the
alsobeconfigured.ThoseregionscanbedescribedbyBED sequencealignment.Itisdisplayedasachildelementofthe
filesto,forexample,limittheanalysistoregionsrelevantto raw sequencing data, making the relation between the raw
thetypeofanalysisperformedinthecurrentproject.Those dataandthealignmentimmediatelyobvious.
BEDfilescaneitherbeprovidedbytheuserorautomatically
GensearchNGS does not provide the user with a single
generated based on a list of gene names. Projects also
buttonwhichgoesfromtherawsequencingdatatothefinal
associate the specific reference sequence to be used during
report.Insteadoffullyautomatingthepipeline,Gensearch-
the analysis. The user can automatically download them
NGS leads the user from one processing step to the next.
fromUCSC,whichprovidesreferencesequencesofthemost
Before each step the user is given the choice to change
commonhumanreferenceassemblies,hg18,hg19,andhg38.
parametersorusethepreviouslyused(orstandard)parame-
Theusageofcustomreferencessequencesisalsosupported.
ters.Aftertheinitialimportoftherawsequencingdata,the
The patient centric database integrated into GensearchNGS
user no longer has to handle the data files. GensearchNGS
allows the user to associate variants with patients, a feature
performs the required data file handling and conversions
very important in diagnostics. The variants associated with
betweentheprocessingsteps.
patients are shared between projects, allowing the user to
While there is no fully automated way to go from the
identify variants found in different projects on previously
rawsequencingdatatothefinalvariantreport,thereisstill
analysedpatients.
awaytoautomateabigpartoftheanalysis.Whenimporting
To perform the NGS data analysis, GensearchNGS uses
rawsequencingdata,theuserhasthepossibilitytoperform
thegenesdescribedinEnsemblasthegenemodel.Inapro-
sequencealignmentandvariantcallingautomatically.Those
cesstransparenttotheuser,thegenemodelcorrespondingto
two additional analysis steps are performed using either
thereferencesequencedefinedfortheanalysisisdownloaded
previouslyusedconfigurationparametersorthedefaultones.
automaticallyfromEnsembl.Additionalmetainformationis
Only the final variant validation has then to be performed
attached to the genes, such as the associated phenotypes
by the user to create the final report. Furtherdetails of this
as described by HPO [18] or OMIM [30]. Figure 2 shows
processfollowinthenextchapter.
the main interface of GensearchNGS, with the patient list
ontheleftandtheinformationaboutthecurrentlyselected
patient in the center. The central area of the image shows 3.2.DataAnalysis. Toperformtheinitialqualitycontrolof
the information about the currently selected patient. This the raw sequencing data, GensearchNGS integrates its own
includes his name, internal ID, associated phenotypes, and tools allowing the user to import the raw data available
allvariantswhichhavemanuallybeenconfirmedbytheuser. in different file formats. The compatible data formats are
Whenvalidatingvariantstheusercanclassifythemmanually describedinSection2.Thevariousformatsareconvertedin
into different categories. The categories are based on the thestandardFASTQformatduringthedataimportstep.For
6 BioMedResearchInternational
Figure3:Overviewofvariousstatisticsthattheusercaninspecttoverifythedataquality.
the quality control of the raw sequencing data, the user is overrepresented K-mers at the start and the end of the
presentedwithvariousstatisticsasseeninFigure3. sequencedreads.
The statistics allow the user to judge the quality of the Thisrawdataimportprocessreplacesexistingtoolslike
sequencing data. This includes statistics such as the length, FastQC to check the quality, FASTX to filter the data, and
averagequality,andGCcontentdistributionofthesequenced the various utilities that transform different sequencing file
reads, as well as the distribution of the sequenced bases formatsintothestandardFASTQfileformat.Transparently
and their quality for every read position and detection of for the user, the custom implementations of those tools,
BioMedResearchInternational 7
whichproducethesameresults,havebeenintegratedintoa be inspected, giving detailed coverage information of the
singleprocessingstep.Thisincludesthepossibilitytoremove previously defined regions of interest. The collected quality
adapter sequences and split the raw sequencing data into controlinformationcanbeexportedasareportforarchiving
multiple files if they were produced through multiplexed purposesusingoneofvariousformatssuchasPDF,HTML,
sequencing. Based on the reported quality information, the orPNG.
usercanchoosetofiltertherawdatafollowingseveralfilters, WhileinitiallyGensearchNGSintegratedVarscan2[14]
such as maximum and minimum length of the sequences asitsvariantcaller,areplacementwascreatedtoreducethe
󸀠
and their minimal average quality. He can also trim the 5 external dependencies and to improve the performance of
󸀠
and 3 parts of all reads or trim their excess length if they variant calling. A variant scanner based on the model of
arelongerasauserdefinedlength.Thetrimmingisdonein Varscan2hasbeenimplementedandisintheprocessofbeing
additiontotheadapterandprimerremoval.Thepossibilityto publishedasaseparatetoolcalledGNATY,availableforfree
splittherawsequencingdatabasedonbarcodesintheread, fornoncommercialusage(http://gnaty.phenosystems.com/).
allowingtheimportofrawdatafrommultiplesamplesthat Our custom variant caller shows important performance
were sequenced at the same time, is also integrated in the advantagesovertheexistingVarscan2implementationofthe
importprocess. samevariantcallingmodel,whileproducingthesameanaly-
Toalignthesequenceddataagainstareferenceassembly, sisresults.Indeed,GensearchNGSshowsspeedimprovement
GensearchNGSprovidestheuserwithmanyoptions.Several of up to 18 times for variant calling compared to Varscan
existingalignmentalgorithmsareintegratedintotheapplica- 2, providing a similar range of features, such as filters that
tionthroughplugins,allowingtheirusagewithouttheneed excludesequencingartifacts.
of advanced technical know-how. The currently integrated To be able to further filter down the detected variants
alignment algorithms are BWA [7], Bowtie 1 + 2 [8], and based on their relevancy to the sample’s phenotype, several
Stampy[29]. sourcesareusedtoannotatethemwithvariousinformation.
Most of those alignment algorithms are developed in a This includes the prediction of the variant effect on the
waythatdonotallowthemtobeusedonmultipleplatforms. different transcripts it affects, a feature closely modeled
Infact,mostofthemonlyrunonLinux,whichisproblematic aftertheVariantEffectPredictor(VEP)fromEnsembl[17].
foramultiplatformapplicationlikeGensearchNGS.Forthis GensearchNGS implements its own version of the the VEP
reason, a custom alignment algorithm written in Java has algorithm to be able to perform the effect prediction in a
beenintegrateddirectlyintoGensearchNGS.Thisallowsthe fractionofthetimecomparedtotheonlineVEPservice.This
usertoperformthefullNGSdataanalysis,evenonoperating isismainlyachievedbydoingallcalculationslocallywithout
systems,likeWindows,notsupportedbymostbioinformatics having to transfer the data to a remote server. Predictions
tools. suchastheadditionorremovalofastopcodoninanaffected
Thealignersupportspairedendreadsandgappedalign- transcript, if a splice site is affected and many others, can
ment and has been optimized to use less than 5GB of be made with this algorithm. Additional information, such
RAM to perform a full genome alignment. This makes it asifthevariantisknowninanypublicdatabase,associated
suitable for most of today’s workspace desktops. It is fully clinical significance, and SIFT [33] and PolyPhen 2 [34]
multithreaded to support all cores of the machine and is predictions are retrieved through the Ensembl web service.
able to offload calculations, as described in Section 3.4, to Other services, like the Human Phenotype Ontology [18],
multiple computers. The aligner uses a hash based index, providetheinformationaboutthephenotypeassociatedwith
a concept made popular through the BLAST aligner [9], aspecificgene,acrucialinformationindiagnostics.
withacustomheuristicbasedcandidatealignmentpositions To filter the annotated variants, GensearchNGS takes
selection [37]. The final alignment is being done using the an interactive approach which allows the user to modify
Gotoh [38] algorithm, a dynamic programming algorithm differentbasicfilters,suchasminimumormaximumsample
usedforgappedalignment. frequency,coverage,readbalance,orquality.Therearealso
The custom alignment algorithm being implemented in advanced filters, such as the predicted consequence of the
Javadoesnotachievequitethesameperformanceasnatively variant, as determined by the annotation process, or if a
runningalignerslikeBWA.ForWGSdatasetsitisgenerally certainphenotypeisassociatedwiththeaffectedgenes.This
50% slower. The advantage of the GensearchNGS aligner includes the possibility to filter by population frequency of
over external aligners is that it is available on all supported avariant.Thefilteringbypopulationfrequencydoesnotyet
platforms, allowing the user to perform the complete NGS takeintoaccounttheethnicityofthepatient.Figure4shows
data analysis on all platforms. But when running on a anexampleofthislist,withseveraluserselectedinteractive
platform which supports the external aligners and speed is filtersapplied.
adecidingfactor,theusersareencouragedtousethealigners Additional features are available to perform a more in-
availableasplugins. depthanalysisofvariants,suchasfamilyanalysiswherethe
Thequalityofresultingalignmentcanbeverifiedthrough variants of multiple patients can be compared to find those
various metrics by the user. He can easily verify how many only present in a certain group of patients. The usage of
sequences have been discarded during the alignment pro- pedigreeinformationisofgrowingimportanceinNGSdata
cess and how many sequences have been aligned to the analysis, as it allows the user to filter sequencing artifacts
differentchromosomesinthereferencesequence.Advanced and noncausative variants. This is a very important feature,
statistics,similartotheonescreatedbybedtools2,canalso especiallywhenusingWGSwheretheamountofvariantscan
8 BioMedResearchInternational
Figure 4: Integrated variants list with several interactive filters
applied.
Figure 5: Example view of the genome browser, showing several
availabletracks.
be very large. Thanks to the usage of standard file formats
likeVCF,externaltoolslikeexomewalker[39]canbeused
to perform more advanced family analysis for features not withallrelevantinformationwhenneeded.Theusercanopen
covered in GensearchNGS. In contrast to many other NGS external resources linked from many features in the visual-
data analysis software packages, GensearchNGS does not ized data, such as genes, transcripts, or variants, to retrieve
focusontheautomationofeveryaspectoftheanalysis.While additionalinformation.Thegenomebrowsertakesadvantage
certainpartscanbeautomated,suchasperformingsequence oftheintegrateddatasources,liketheEnsemblgenemodel,
alignment and variant calling on raw sequencing data, the to display and interactively update encoded proteins, based
philosophy of our software is to let the user to control the on the variants present in the visualized data. The genome
different analysis steps. This gives him enough flexibility to browser is tightly integrated into GensearchNGS, allowing
adapt to the data he is currently analysing without slowing the user to easily visualize the alignment data belonging to
himdown,byhavingprojectspecificanalysisoptionswhich featuresofthesample.
arepreselectedforeveryanalysisstep.Thismakesitpossible
to perform most types of analysis with the project specific
3.4. Distributed Computing. The lack of computing power
defaultoptionsthroughasimplepressofabutton.
to perform NGS data analysis is a very common issue in
manylaboratories.Toaddressthisproblem,GensearchNGS
3.3. Genome Browser. A central tool of GensearchNGS is doesnotonlyoptimizethealgorithmsforspeedandmemory
the integrated genome browser to visualize the aligned usagebutalsotrytoexploitthealreadyexistingcomputing
sequencingdata.Thegenomebrowser,asshowninFigure5, infrastructure. One way to do this is to use distributed
allows to interactively browse the sequence alignments. It computing, a technique which allows the user to combine
is possible to display the aligned sequences and additional multiple computers to perform computing intensive tasks.
genomic data, such as the gene locations and the encoded Those computers can be in the same local network or at a
proteinsbydifferenttranscripts. remote location. One of the most demanding steps in NGS
The genome browser allows the user to display sev- dataanalysisissequencealignment.Forthisreasonitisthe
eral tracks, containing information which helps to analyse firstfeaturetotakeadvantageofthistechniqueinGensearch-
the sequencing data. The predefined tracks, which can be NGS,allowingtheusertojoinmultiplecomputersinthesame
activated or deactivated, allow the user to display coverage networktoperformsequencealignment.Theapproachused
information, GC content of the reference sequence, the byGensearchNGSistotransparently,onrequestbytheuser,
qualityofthesequencedreads,orcustomannotationfilesin useallcomputerswithinthesamenetworkwhereGensearch-
standardfileformatssuchasBED,WIG,andGFF.Thefocus NGSisinstalled.Additionally,theuserhasthepossibilityto
was put on optimizing the display of the data, allowing the startupGensearchNGSinstancesinacloud.Currently,only
usertovisualizelargedatafileswithoutrequiringapowerful theamazoncloudissupported.Thisprocessworksregardless
computer.Thisisachievedbystreamingthedataondemand, of the operating system used by the different installations
only loading the parts of the data currently needed for of GensearchNGS. By using an automatic service discovery
the visualization.The informationabout which variants are protocol,everyinstallationofGensearchNGSlocatedinthe
presentinthealignmentisavailabletotheuserinthesame same network is discovered without the need for manual
formasthevariantlistinthemainapplication.Itcanbeseen configuration.Thisallowstheusertousemultiplecomputers
inFigure4,allowingtheusertoquicklyjumptothevariants at the same time or to completely offload calculations to a
of interest. The genome browser, like all of GensearchNGS, differentcomputer,makingitpossibletousecomputersnot
is heavily connected with external data sources (Ensembl, suitedforfullgenomealignmentsasananalysisworkstation.
dbSNP,Entrez,OMIM,Blast,andHPO)providingtheuser The distributed sequence aligner used in GensearchNGS
BioMedResearchInternational 9
currentlyonlysupportstheGensearchNGSaligneralgorithm NGS data analysis pipeline. We then presented Gensearch-
andnotexternalalignerssuchasBWAorBowtie2.Itusesa NGS, a software suite integrating various algorithms to
streambasedapproach,wheretheclientmachinestreamsthe perform the discussed workflow. GensearchNGS is a com-
datatobealignedtoallcomputersinvolvedinthealignment mercially available software, distributed by Phenosystems
process.Thisincludestheclientmachinewhichreceivesthe SA. A demo version of GensearchNGS is freely available
aligned sequences back and combines them into the final athttp://www.phenosystems.com/.Thetwomainissues,the
alignment file. The process of distributing the alignments technical complexity of the analysis and the computing
overmultiplecomputersandthecloudistransparenttothe infrastructurerequirements,havebeenaddressedasfollows.
user, but it is optional, because the usage of an external To reduce the technical complexity of the analysis, the
cloudisnotcompatiblewiththesecurityguidelinesofsome userisprovidedwithanintuitiveuserinterfacewhichguides
laboratories.GensearchNGSisnottheonlyNGSdataanalysis himthroughtheprocessofthedataanalysis.Heisprovided
software using distributed sequence alignment. It is this withtherequiredtoolstogofromrawsequencingdatathethe
ability to combine the computing resources of the local final visualization and report generation for the sequenced
computer,multiplecomputersinthesamenetworkaswellas samples, without having to learn complex bioinformatics
thecloud,whichmakesitsapproachuniqueandveryflexible. tools. The interface abstracts the technical complexity from
theuserwhilekeepingenoughflexibilitytoadapttodifferent
4.Discussion analysis protocols. By lowering the technical complexity of
NGS data analysis compared to existing individual analysis
GensearchNGSaddressesproblemsthatsmallerlaboratories
tools,GensearchNGSexpandsthepotentialuserbaseofthe
are facing in various ways. The issue of the technical com-
software.
plexityoftheanalysisisapproachedthroughanintuitiveuser
To address the issue of the computing infrastructure
interface. The user interface abstracts the complexity of the
requirements, GensearchNGS focuses on the optimal usage
underlying NGS data analysis toolchain, which is a major
of existing computing resources. This is primarily achieved
usability improvement over the usage of individual tools.
through the development of analysis algorithms that run
Nevertheless,itgivestheuserfullcontrolovertheindividual
on a minimal amount of resources while producing the
analysis steps, instead of focusing on the creation of auto-
sameanalysisresultsasstandardtools.Anexamplehasbeen
matedanalysispipelines.Italsoprovidesanoriginalapproach
describedwiththevariantcallingalgorithmwhichproduces
by providing a patient centric user interface. This user
the same results as Varscan 2 but is 18 times faster. The
friendlyinterfaceallowsusersunfamiliarwiththeunderlying
secondapproachistheintegrationofdistributedcomputing,
software tools to fine tune the analysis parameters, without
allowing the distribution of computing intensive tasks like
overwhelmingthemwithfeaturesorrarelyusedoptions.This
alignmenttobespeedupbytheusageofmultiplecomputers.
increases the number of people able to perform NGS data
With a solid framework built for DNAseq analysis,
analysis. This is particularly interesting for geneticists that
GensearchNGSwillexpandintofurthertypesofanalysisin
mightnothavespecialbioinformaticstraining.Theissueof
thefuture.OneexampleistheintegrationofRNAseqanalysis,
increasinginfrastructurerequirementsforNGSdataanalysis
as the combination of OMICS data for a single sample is
is being addressed twofold. First, more efficient algorithms
of increasing importance to understand rare diseases [44].
arebeingdeveloped,liketheheavilyoptimizedvariantcalling
In fact, many parts of GensearchNGS, such as the genome
algorithm included in GensearchNGS. Secondly, the ability
browser,havealreadybeenadaptedtosupportRNAseqdata.
to distribute heavy calculations over several computers or
Theintegrationofmultipledatasourceswillnotonlyallow
even a cloud service has been developed. GensearchNGS
theusertointegratedifferenttypesofOMICSdatabutalso
hasauniqueapproachtothisproblembyenablingtheuser
allow multiple views on the same sample. One example of
to combine the various distribution methods to perform
this is the integration of Sanger sequencing to NGS data,
sequence alignment. It provides the flexibility needed to
a very useful feature for diagnostics where it is common
choose the best compromise between data privacy and
practice to validate variants identified through NGS with
alignment speed depending on the laboratory policies and
Sangersequencing.
needs. The resulting software enabled various laboratories
Further optimizations of the algorithms used in the
to perform NGS data analysis and publish their results in
analysiswillalsobeevaluated,includingoptimizationforthe
various fields. The topics range from muscular dystrophies
[40,41]tohearingloss[42]andmalignanthyperthermia[43] underlyinglibraries.OneexampleisHTSJDK,whereseveral
but also yet unpublished work about CNV analysis and the performance patches have been submitted and included,
detectionofsomaticmutationsincancerscreening. improvingNGSdataanalysisperformancenotonlyforusers
of GensearchNGS, but also forthe complete bioinformatics
community.
5.Conclusion
In this paper we reviewed the workflow used to analyze Disclosure
NGS data in a diagnostics environment. The reviewed
workflow includes the detection of SNPs and short indels PierreKuonenhasanonpaidadvisorypositionatPhenosys-
from panel, WES or WGS DNAseq data. We also pre- temsSA.DavidAtlanisworkingfulltimeforPhenosystems
sentedissuesfacedbysmallerlaboratoriestryingtosetupa SA.
10 BioMedResearchInternational
ConflictofInterests [15] J.O’Rawe,T.Jiang,G.Sunetal.,“Lowconcordanceofmultiple
variant-callingpipelines:practicalimplicationsforexomeand
ThomasDandekarhasnoconflictofinterestsregardingthe genomesequencing,”GenomeMedicine,vol.5,no.3,article28,
publicationofthispaper. 2013.
[16] S.T.Sherry,M.-H.Ward,M.Kholodovetal.,“dbSNP:theNCBI
Acknowledgments databaseofgeneticvariation,”NucleicAcidsResearch,vol.29,pp.
308–311,2001.
GensearchNGS is a commercially available software dis- [17] W. McLaren, B. Pritchard, D. Rios, Y. Chen, P. Flicek, and F.
tributed by Phenosystems SA. Beat Wolf’s Ph.D. thesis is Cunningham,“Derivingtheconsequencesofgenomicvariants
withtheEnsemblAPIandSNPEffectPredictor,”Bioinformatics,
partiallyfinanciallysupportedbyPhenosystemsSA.
vol.26,no.16,ArticleIDbtq330,pp.2069–2070,2010.
[18] P. N. Robinson and S. Mundlos, “The human phenotype
References
ontology,”ClinicalGenetics,vol.77,no.6,pp.525–534,2010.
[19] P.Danecek,A.Auton,G.Abecasisetal.,“Thevariantcallformat
[1] E. L. van Dijk, H. Auger, Y. Jaszczyszyn, and C. Thermes,
and VCFtools,” Bioinformatics, vol. 27, no. 15, pp. 2156–2158,
“Tenyearsofnext-generationsequencingtechnology,”Trends
2011.
inGenetics,vol.30,no.9,pp.418–426,2014.
[20] J.T.Robinson,H.Thorvaldsdo´ttir,W.Winckleretal.,“Integra-
[2] B.Sikkema-Raddatz,L.F.Johansson,E.N.deBoeretal.,“Tar-
tivegenomicsviewer,”NatureBiotechnology,vol.29,no.1,pp.
getednext-generationsequencingcanreplacesangersequenc-
24–26,2011.
inginclinicaldiagnostics,”HumanMutation,vol.34,no.7,pp.
[21] I. Milne, G. Stephen, M. Bayer et al., “Using tablet for visual
1035–1042,2013.
explorationofsecond-generationsequencingdata,”Briefingsin
[3] H. P. J. Buermans and J. T. den Dunnen, “Next generation Bioinformatics, vol. 14, no. 2, Article ID bbs012, pp. 193–202,
sequencingtechnology:advancesandapplications,”Biochimica 2013.
etBiophysicaActa—MolecularBasisofDisease,vol.1842,no.10,
[22] T.Abeel,T.vanParys,Y.Saeys,J.Galagan,andY.vandePeer,
pp.1932–1941,2014.
“GenomeView: a next-generation genome browser,” Nucleic
[4] FastQC,“Aqualitycontroltoolforhighthroughputsequence AcidsResearch,vol.40,no.2,articlee12,2012.
data,”BabrahamBioinformaticsWebsite,http://www.bioinfor-
[23] W.J.Kent,C.W.Sugnet,T.S.Fureyetal.,“Thehumangenome
matics.babraham.ac.uk/projects/fastqc/.
browser at UCSC,” Genome Research, vol. 12, no. 6, pp. 996–
[5] W.R.Pearson,T.Wood,Z.Zhang,andW.Miller,“Comparison 1006,2002.
ofDNAsequenceswithproteinsequences,”Genomics,vol.46, [24] F.Cunningham,M.RidwanAmode,D.Barrelletal.,“Ensembl
no.1,pp.24–36,1997. 2015,” Nucleic Acids Research, vol. 43, no. 1, pp. D662–D669,
[6] H. Li and N. Homer, “A survey of sequence alignment algo- 2015.
rithms for next-generation sequencing,” Briefings in Bioinfor- [25] NGS Data Analysis Software List, SEQanswers, 2012, http://
matics,vol.11,no.5,pp.473–483,2010. seqanswers.com/wiki/Software/list.
[7] H.LiandR.Durbin,“Fastandaccuratelong-readalignment [26] CLCGenomicsWorkbench,CLCBio,Aarhus,Denmark,2014,
withBurrows-Wheelertransform,”Bioinformatics,vol.26,no. http://www.clcbio.com.
5,pp.589–595,2010.
[27] NextGENe, Softgenetics, State College, Pa, USA http://www
[8] B.LangmeadandS.L.Salzberg,“Fastgapped-readalignment .softgenetics.com.
withBowtie2,”NatureMethods,vol.9,no.4,pp.357–359,2012. [28] J. Goecks, A. Nekrutenko, J. Taylor et al., “Galaxy: a com-
[9] S. F. Altschul, T. L. Madden, A. A. Scha¨ffer et al., “Gapped prehensive approach for supporting accessible, reproducible,
BLASTandPSI-BLAST:anewgenerationofproteindatabase and transparent computational research in the life sciences,”
search programs,” Nucleic Acids Research, vol. 25, no. 17, pp. GenomeBiology,vol.11,no.8,articleR86,2010.
3389–3402,1997. [29] G.LunterandM.Goodson,“Stampy:astatisticalalgorithmfor
[10] M. A. DePristo, E. Banks, R. Poplin et al., “A framework for sensitiveandfastmappingofIlluminasequencereads,”Genome
variationdiscoveryandgenotypingusingnext-generationDNA Research,vol.21,no.6,pp.936–939,2011.
sequencingdata,”NatureGenetics,vol.43,no.5,pp.491–498, [30] Online Mendelian Inheritance in Man, OMIM, McKusick-
2011. NathansInstituteofGeneticMedicine,JohnsHopkinsUniver-
[11] P.Carnevali,J.Baccash,A.L.Halpernetal.,“Computational sity,Baltimore,Md,USA,2015,http://omim.org/.
techniques for human genome resequencing using mated [31] CafeVariome,http://www.cafevariome.org.
gappedreads,”JournalofComputationalBiology,vol.19,no.3, [32] TheGenomeSequencingConsortium,“Initialsequencingand
pp.279–292,2012. analysisofthehumangenome,”Nature,vol.409,pp.860–921,
[12] H. Li, B. Handsaker, A. Wysoker et al., “The sequence align- 2001.
ment/mapformatandSAMtools,”Bioinformatics,vol.25,no.16, [33] P.Kumar,S.Henikoff,andP.C.Ng,“Predictingtheeffectsof
pp.2078–2079,2009. codingnon-synonymousvariantsonproteinfunctionusingthe
[13] A. R. Quinlan and I. M. Hall, “BEDTools: a flexible suite of SIFTalgorithm,”NatureProtocols,vol.4,no.7,pp.1073–1082,
utilities for comparing genomic features,” Bioinformatics, vol. 2009.
26,no.6,pp.841–842,2010. [34] I. A. Adzhubei, S. Schmidt, L. Peshkin et al., “A method and
[14] D.C.Koboldt,Q.Zhang,D.E.Larsonetal.,“VarScan2:somatic server for predicting damaging missense mutations,” Nature
mutation and copy number alteration discovery in cancer by Methods,vol.7,no.4,pp.248–249,2010.
exomesequencing,”GenomeResearch,vol.22,no.3,pp.568– [35] AlamutVisual,InteractiveBiosoftware,Rouen,France,http://
576,2012. www.interactivebiosoftware.com/alamut-visual.