HitWalker: variant prioritization for personalized functional cancer genomics.

0.14 MB·English

by  Bottomly, Daniel

BIOINFORMATICS APPLICATIONS NOTE Vol.29no.42013,pages509–510
doi:10.1093/bioinformatics/btt003
Genetics and population analysis
AdvanceAccesspublicationJanuary9,2013
HitWalker: variant prioritization for personalized functional
cancer genomics
Daniel Bottomly1,2,*, Beth Wilmot1,2, Jeffrey W. Tyner1,3, Christopher A. Eide1,4,5,
Marc M. Loriaux1,6, Brian J. Druker1,4,5 and Shannon K. McWeeney1,2
1KnightCancerInstitute,2OregonClinicalandTranslational Research Institute,3Departmentof CellandDevelopmental
Biology, 4Division of Hematology and Medical Oncology, 5Howard Hughes Medical Institute and 6Department of
Anatomic Pathology, Oregon Health and Science University, Portland, OR 97239, USA
AssociateEditor:GunnarRatsch
ABSTRACT and population genetic assumptions are often used to reduce
Summary: Determining the functional relevance of identified se- thesetofvariants(Ngetal.,2009).Forinstance,limitingatten-
quencevariantsincancerisaprerequisitetoultimatelymatchingspe- tion to non-synonymous single-nucleotide variations, as well as
cific therapies with individual patients. This level of mechanistic usingtheuseofvariantdatabases,suchasdbSNP(Sherryetal.,
understandingrequiresintegrationofgenomicinformationwithcom- 2001)andthe1000genomesproject(The1000GenomesProject
plementaryfunctionalanalysestoidentifyoncogenictargetsandrelies Consortium, 2010), enables researchers to focus on low-
onthedevelopmentofcomputationalframeworkstoaidinthepriori- frequency variants that are potentially damaging. However,
tizationandvisualizationofthesediversedatatypes.Inresponseto evenafterthesefilters,itisrelativelyinfrequentthataresearcher
this,wehavedevelopedHitWalker,whichprioritizespatientvariants isleftwithamanageablesetofvariantsforbiologicalvalidation.
relative to their weighted proximity to functional assay results in a Forcancercellsderivedfromagivenpatient,functionalassays
protein–protein interaction network. It is highly extensible, allowing can be performed that allow researchers to determine the rele-
incorporationofdiversedatatypestorefineprioritization.Inaddition vance of genes to cell viability, such as through the use of tar-
toa ranked list of variants, we have also devised a simple shortest getedsiRNAscreens(Tyneretal.,2009).Thesegenetargetscan
path-basedapproachofvisualizingtheresultsinanintuitivemannerto bescoredusingabinaryorquantitativeencodingthatindicates
providebiologicalinterpretation. outlier status relative to other samples. Similarly, screens that
Availabilityand implementation:The program, documentationand measure the sensitivity of a patient’s cells to a panel of small
exampledataareavailableasanRpackagefromwww.biodevlab.org/ moleculescanbescoredrelativetogenesusingtheknowngene
HitWalker.html. targetsofeachcompound(Tyneretal.,2013).Thesefunctional
Contact:[email protected] assaysbythemselvesprovidesomeinformationtoresearcherson
Supplementary information: Supplementary data are available at the identity of the signalling pathways that are required for
Bioinformaticsonline. cancer growth and survival. However, they do not necessarily
indicate the mutated gene(s) leading to dysregulation of these
ReceivedonOctober1,2012;revisedonDecember4,2012;accepted
signalling pathways, as the true causative variant(s) could be
onDecember30,2012
foundinanygenewithcapacitytoregulatethepathway.
Toprioritizevariantsdetectedinourcohortofleukaemiapa-
tientsrelativetoourfunctionalassayresults,wehavedevisedthe
1 INTRODUCTION
RandSQLframeworkHitWalkerusingaprotein–proteininter-
The advent of next-generation sequencing technology such as action(PPI)networkasabackbone.Inadditiontoarankedlist
that available from Illumina provides an unprecedented ability ofvariants,HitWalkeralsoallowsforsimplevisualizationofthe
tointerrogateindividualgenomes(Metzker,2010).Accompany- relevantsubnetwork,aswellasoverlayingrelevantmeta-data.
ingtechnologiessuchastheabilitytomultiplexsamples,aswell
as efficient sequence capture technologies (Ng et al., 2009) fur-
therenabletargetedregionsofthegenometobere-sequencedat 2 DESCRIPTION OF SOFTWARE
a reasonable cost per sample. In this manner, specific genes or
wholeexomescanbeinterrogatedtoidentifyvariantspotentially 2.1 Variant prioritization
having a deleterious impactonproteincoding regions. Efficacy Variants are related to genes implicated from the functional
ofthisapproachforclinicalresearchhasbeenshownthroughthe assays through a user-defined PPI network, such as STRING
discoveryofvariantsunderlyingsimpleMendeliandisorders,as (Szklarczyketal.,2011),whichprovidesknownandhypothetical
wellasmorecomplicatedvariants/mutationsinvolvedincancer links between proteins that are weighted bya confidence score.
and potentially complex traits (Kiezun et al., 2012). Because Variantsandthefunctionalassayresultsaremappedtothepro-
many variants are produced for a given sample, mechanistic teinsofthenetworkusingpre-suppliedmeta-data.Forexample,
variants can be mapped to transcripts, which in turn can be
*Towhomcorrespondenceshouldbeaddressed. mappedtoproteins (as wellas gene symbols).Thishierarchy is
(cid:2)TheAuthor2013.PublishedbyOxfordUniversityPress.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0/), which
permitsunrestrictedreuse,distribution,andreproductioninanymedium,providedtheoriginalworkisproperlycited.
D.Bottomlyetal.
all three genes are candidates based on the observed drug
response.
3 DISCUSSION
Oursoftwareprovidesaflexibleframeworkforprioritizingvari-
antsrelativetofunctionalassayoutcomesandaPPInetworkor
other association graph using an RWR algorithm. In addition,
we include visualization of key subnetwork members to aid in
biological interpretation. Prioritized variants can then be fol-
lowedupusingSanger sequencing,as wellasadditionalexperi-
Fig.1. ModifiedvisualizationoutputfromHitWalkerdisplayingthetop
mentstocategorizetheirphenotypiceffect.Softwaredevelopers
three assay hits (EPHA4, JAK3 and FRK) andvariants (FLT3, ZAK
andPRKCE)foran acutemyeloidleukaemiapatient.Note that other can easily modify queries and R functions for their in-house
hitsarepulledoutandannotated,astheyareontheshortestpath.Gene databases and functional assays. Although MySQL is our
namesareprovidedforeachnode.Fornodescontainingvariants(blue), database of choice, any other R/DBI-supported database
frequencyinformationisreportedintermsofthepatientcohortcounts wouldworkwithminormodificationstotheRcode.
(F),aswellastheRWRrank(R).RedandgreennodesindicatesiRNA
and gene target hits, respectively. Dotted borders indicate absence of
capture probes for a given gene. Dashed borders indicate functional
ACKNOWLEDGEMENTS
assay targets whose inhibition did not significantly alter cell viability.
Confidencescoresfortheinteractionsbetweenthetwogenesarereported The authors acknowledge helpful discussions with Cristina
nearthelinesconnectingtwogivengenes Tognon and the other members of the Leukemia and
Lymphoma Society Specialized Center of Research at the
KnightCancerInstitute.
Funding: Leukemia & Lymphoma Society (award 7005-11);
managedthroughthespecificationof‘core’or‘summary’IDsto
NIH/NCI (5P30CA069533-13, 4R00CA151457-03); NIH/
beusedintheinternalfunctions.
NCATS(5UL1RR024140)VFoundationforCancerResearch;
By default, prioritization is performed using a random walk
WilliamLawrence&BlancheHughesFoundation,HHMI.
withrestartsalgorithm(RWR)(Ko¨hleretal.,2008;Tongetal.,
2008). In HitWalker, RWR provides a measure of weighted ConflictofInterest:nonedeclared.
proximity between a set of proteins associated with functional
assayhitsandasetofproteinscontainingvariants.Variantsare
prioritizedbasedontheresultingRWRassociationscoreattrib-
REFERENCES
uted to the protein. See the Supplementary Material for more
information.Wenotethatanynumberofapproachesrelatedto Erten,S. et al. (2011) DADA: Degree-Aware Algorithms for Network-Based
DiseaseGenePrioritization.BioDataMining,4,19.
ourRWRimplementationcouldbeeasilyappliedaswell,includ-
Forbes,S.A. et al. (2011) COSMIC: mining complete cancer genomes in the
ingcorrectionsfordegree(Ertenetal.,2011).
CatalogueofSomaticMutationsinCancer.NucleicAcidsRes.,39,D945–D950.
Kiezun,A.etal.(2012)Exomesequencingandthegeneticbasisofcomplextraits.
2.2 Visualization Nat.Genet.,44,623–630.
Ko¨hler,S.etal.(2008)Walkingtheinteractomeforprioritizationofcandidatedis-
The relevant subnetworks can be visualized by approximating easegenes.Am.J.Hum.Genet.,82,949–958.
theRWRalgorithmusingtheunweightedshortestpathsbetween Metzker,M.L. (2010) Sequencing technologies: the next generation. Nat. Rev.
the top queries and seeds. Shortest paths connecting the func- Genet.,11,31–46.
Ng,S.B. et al. (2009) Targeted capture and massively parallel sequencing of 12
tional hit and variant genes are also displayed for additional
humanexomes.Nature,461,272–276.
context. For the shortest path calculations, the path with the Sherry,S.T.etal.(2001)dbSNP:theNCBIdatabaseofgeneticvariation.Nucleic
highestoverallpathscoreischosenifmultipleequivalentshortest AcidsRes.,29,308–311.
pathsexist.Theusercandefinefunctionsmappingmeta-datain Szklarczyk,D.etal.(2011)TheSTRINGdatabasein2011:functionalinteraction
thedatabasetoattributesofthenetwork.Forinstance,genescan networksofproteins,globallyintegratedandscored.NucleicAcidsRes.,39,
D561–D568.
berepresentedusingdifferentshapesandcolours.Anexampleof
Tong,H.etal.(2008)Randomwalkwithrestart:fastsolutionsandapplications.
suchafigureisshowninFigure1andSupplementaryFigureS1. Knowl.Inf.Syst.,14,327–346.
Inapatientwithacutemyeloidleukaemia,anS451Faminoacid Tyner,J.W.etal.(2009)RNAiscreenforrapidtherapeutictargetidentificationin
changeintheFLT3genewasthetoprankingvariant,whichhas leukemiapatients.Proc.NatlAcad.Sci.USA,106,8695–8700.
Tyner,J.W.etal.(2013)Kinasepathwaydependenceinprimaryhumanleukemias
been previously characterized and catalogued in COSMIC
determinedbyrapidinhibitorscreening.CancerRes.,73,285–296.
(Forbesetal.,2011).Non-synonymoussingle-nucleotidevariants
The1000GenomesProjectConsortium.(2010)Amapofhumangenomevariation
inZAKandPRKCEwerethenexthighest-rankedvariants,and frompopulation-scalesequencing.Nature,467,1061–1073.
510