下载

1下载券

加入VIP
  • 专属下载特权
  • 现金文档折扣购买
  • VIP免费专区
  • 千万文档免费下载

上传资料

关闭

关闭

关闭

封号提示

内容

首页 Relevance network

Relevance network.pdf

Relevance network

15874117534
2013-09-11 0人阅读 举报 0 0 0 暂无简介

简介:本文档为《Relevance networkpdf》,可适用于医药卫生领域

DiscoveringfunctionalrelationshipsbetweenRNAexpressionandchemotherapeuticsusceptibilityusingrelevancenetworksAtulJButte†‡,PabloTamayo§,DonnaSlonim§,ToddRGolub§¶,andIsaacSKohane††Children’sHospitalInformaticsProgramandDivisionofEndocrinology,DepartmentofMedicine,Children’sHospital,LongwoodAvenue,Boston,MA§WhiteheadInstituteforBiomedicalResearch,CambridgeCenter,Cambridge,MAand¶DanaFarberCancerInstitute,BinneyStreet,Boston,MACommunicatedbyLouisMKunkel,HarvardMedicalSchool,Boston,MA,August,(receivedforreviewMay,)Inanefforttofindgeneregulatorynetworksandclustersofgenesthataffectcancersusceptibilitytoanticanceragents,wejoinedadatabasewithbaselineexpressionlevelsof,genesmeasuredbyusingmicroarraysincancercelllines,toadatabasewiththeamountsof,anticanceragentsneededtoinhibitgrowthofthosesamecelllinesComprehensivepairwisecorrelationswerecalculatedbetweengeneexpressionandmeasuresofagentsusceptibilityAssociationsweakerthanathresholdstrengthwereremoved,leavingnetworksofhighlycorrelatedgenesandagentscalledrelevancenetworksHypothesesforpotentialsinglegenedeterminantsofanticanceragentsusceptibilitywereconstructedTheeffectofrandomchanceinthelargenumberofcalculationsperformedwasempiricallydeterminedbyrepeatedrandompermutationtestingonlyassociationsstrongerthanthoseseeninmultiplypermuteddatawereusedinclusteringWediscusstheadvantagesofthismethodologyoveralternativeapproaches,suchasphylogenetictypetreeclusteringandselforganizingmapsWiththeincreasingavailabilityofRNAexpressionmicroarrays,thecurrentfocusisnowonelucidatingnetworksofgenomicregulationhiddeninthelargeamountsofdataTherehavebeenfourgeneraltechniquesusedtoascertainthefunctionsofgenesfromexpressiondataOnewayistolistgenesbyfoldincreaseordecreaseafteraninterventionThismethodhasbeenusedtoanalyzegeneexpressionpatternsinhumancancerandtofindinflammatorydiseaserelatedgenes(,)Althoughimportant,thismethodtypicallyelucidatesonlytheoneregulatorynetworkexaminedAsecondmethodinvolvestheassignmentofeachgenetoamultidimensionalpointwithcoordinatesequaltoexpressionlevelsatvarioustimepointsorexperimentsEuclideandistancesbetweenpointsarecalculated,thengraphedbyusingphylogenetictypetreesRelatedgenesarethoughttobeclosertoeachotherinthemultidimensionalspaceThistechniquehasbeenusedtopredictfunctionalrelationshipsbetweengenesthoughttobeinvolvedincentralnervoussystemdevelopment(,)Thethirdmethodinvolvestakingthesametypeofmultidimensionalspaceandconstructingselforganizingmapstofindclustersofpoints()However,thereareproblemswithbothofthesemethodsusingEuclideandistances,mostnotablythedifficultyinfindinggenesnegativelyassociatedwitheachotherFinally,afourthmethodinvolvesphylogenetictypetreeclusteringusingbranchlengthsproportionaltothecorrelationcoefficientcalculatedbetweengeneexpressionlevels()Thismethodologyhasbeenusedtohierarchicallyclusterchemotherapeuticagentsbymechanismofaction(,)andtoclusterthegenesinvolvedinthefibroblastresponsetoserum()Aproblemwiththismethodisthatitclustersgenesintoasinglestructureandpairseachgenewithoneother,whenseveralregulatorypathwaysmaybepresentinbiologicalsystemsandexpressedgenescanparticipateinmorethanonepathwayOurpurposeherewastodevelopamethodologythatdistinguishestruebiologicalassociationsfromnoise,generatinghypothesesofputativefunctionalrelationshipsbetweenpairsofgenesSpecifically,weusedbaselineRNAexpressionlevelsmeasuredfromtheNCI,asetofhumancancercelllinesusedbytheNationalCancerInstituteDevelopmentalTherapeuticsProgramtoscreenanticanceragentssince()Wejoinedthegeneexpressionlevelstoadatabasewithmeasuresofcancersusceptibilitytoanticanceragents,toseehowthebaselineRNAexpressionlevelsinthecelllinescorrelatedwiththeinhibitionofgrowthofthesesamecelllinestothousandsofanticanceragentsTobeclear,RNAexpressionlevelsweremeasuredwithoutanyexposuretoanticanceragentsAsshownbelow,thismethodology,termedrelevancenetworks,isabletoformclusterswithouthavingtheproblemslistedabovethatareinherentinothermethodologiesAfeatureofaclusteringtechniquesuchasrelevancenetworks,isthatitallowsustofindcorrelationsacrossdisparatebiologicalmeasures,suchasRNAexpressionandsusceptibilitytopharmaceuticalsMethodsGeneExpressionDataRNAexpressionwasmeasuredatbaselineintheNCIcelllinesDetailsofthestepsneededtomeasureRNAexpressionlevelsincellshavebeendescribed()HUarrays(Affymetrix,SantaClara,CA)wereused,containingprobesetsfor,humangenes(,knowngenesand,expressedsequencetags)Thedetailsoftheexpressiondatasetaredescribedelsewhere(TRGolub,personalcommunication),andtheexpressiondataareavailableinitsentiretyathttp:yywwwgenomewimiteduyMPRBecauseprobesetsforsomegenesarepresentmorethanonceonthearray,thetotalnumberonthearrayis,AffymetrixsoftwarewasusedtocalculatetherelativeabundanceofeachgenefromtheaveragedifferenceofintensitiesbetweenmatchingandmismatchedprobepairsdesignedtohybridizeaparticularsequenceAnticancerAgentSusceptibilityDataWeusedavalidatedsubsetoftheNationalCancerInstituteHumanTumorCellLineScreencontaining,anticanceragentstestedagainsttheNCIpanel(,,)TheamountofgrowthinhibitioncomparedwithcontrolwasmeasuredatseveraldosagesforeachchemicalandcancercelllineFromthis,theGI,ordoseneededtocausegrowthinhibition,wascalculatedForthisanalysis,susceptibilitywasexpressedasthenegativelogarithmofGIAlthoughthesusceptibilitydataandbaselineRNAexpressiondatawerenotmeasuredsimultaneously,thesewerebothcharacteristics,orfeatures,ofthesamecelllinesThedatawere‡Towhomreprintrequestsshouldbeaddressedat:Children’sHospital,LongwoodAvenue,Boston,MAEmail:atulbutteharvardeduThepublicationcostsofthisarticleweredefrayedinpartbypagechargepaymentThisarticlemustthereforebeherebymarked“advertisement”inaccordancewithUSC§solelytoindicatethisfactArticlepublishedonlinebeforeprint:ProcNatlAcadSciUSA,ypnasArticleandpublicationdateareatwwwpnasorgycgiydoiyypnas–uPNASuOctober,uvolunoAdministrator高亮Administrator高亮Administrator高亮Administrator高亮concatenatedtogether,makingatotalof,featuresmeasuredonthecelllines,orcasesTheNationalCancerInstitutedatasetwasnotcompletelycomprehensive,inthattherewere,missinganticanceragentsusceptibilityvaluesRemovingFeatureswithLowEntropyEachfeatureinthedatasetwasfirstanalyzedtoascertainwhetheritcontainedasufficientrangeofvaluesOutliervaluesinafeaturewillbiasthecorrelationcoefficientwhenthatfeatureiscomparedwithothersthus,wedesiredtominimizespuriouslyhighcorrelationcoefficientsTodothis,wecalculatedtheentropyofeachfeaturebyusingHOxp~x!log~p~x!!,wherelogisbaselogarithm,andp(x)istheprobabilityavaluewaswithindecilexofthatfeatureForexample,agenewiththeexpressionamounts,,,,andwouldhavedecilesunitswide,withtwovaluesinthefirstdecile,oneinthesixthdecile,andoneintheninthandthdecile,makingHWeexcludedfromfurtheranalysistheoffeaturesthathadthelowestentropy(ie,theleastuniformlydistributedvalues)andwerelikelytobiasthecorrelationcoefficient,eventhoughthismeantwewereunabletoconstructhypotheseswiththesefeaturesOftheoriginal,features,weexcludedRNAmeasurementsandanticanceragents,leaving,RNAexpressionlevelsand,measuresofanticanceragentsusceptibilityThegenesandanticanceragentsremovedarelistedathttp:yywwwchiporgygenomicsRelevanceNetworksWeevaluatethesimilarityoffeaturesbycomprehensivelycomparingallfeatureswitheachotherinapairwisemanneroverthesamecasesSeveralsimilaritymetricshavebeenpreviouslyusedinthismethodology,includingmutualinformation(,)Inthisexperiment,weratethesimilarityofpatternsoffeaturesbyusingrˆrabs~r!r,whereabsistheabsolutevaluefunctionandristhesamplecorrelationcoefficientIneffect,rˆisthesameasr,thepairwisecorrelationcoefficientaroundwhichalargestatisticalliteraturehasbeenbuilt,butretainingtheoriginalpositiveornegativesignofrAllfeaturesareconnectedtoallotherfeatureswithanrˆWehypothesizethatfeatureswithahighabs(rˆ)representhypothesesofabiologicalrelationshipWechooseathresholdabs(rˆ),thendisplayonlythefractionofrelationshipsatorabovethatthresholdGroupsoffeaturesthatareconnectedtoeachotherwithabs(rˆ)higherthanthethresholdwillaggregateandformacluster,orarelevancenetworkBychangingthethresholdabs(rˆ),onecantunetherelevancenetworkstoincludeorexcludehypotheticalrelationshipsAtlowerthresholds,thehypothesesgeneratedmayrepresentnovelandtruefunctionalrelationships,butalsowillbehardertodistinguishfromrandomnoiseResultsUsingthismethodology,relevancenetworkswereconstructedfromthe,features(baselineexpressionof,genesandmeasuresofsusceptibilityto,anticanceragents)intheNCIcelllinesTherewere,,pairwisecomparisonsbetweenfeatures,ofwhichroughlymillionrelationshipswerebetweenapairofgenes,millionrelationshipswerebetweentwoanticanceragents,andmillionwerebetweenageneandananticanceragentThedistributionofrˆisshowninFigOverall,thedistributionhadamodeatzeroandwasskewedsuchthatthereweremorepositivecorrelationsthannegativeonesFivepercentofassociationshadabs(rˆ)aboveForeachgeneandanticanceragent,measurementswererandomlypermutedindependenttimesTheaveragedistributionofrˆwithstandarddeviationsforthesepermuteddatasetsalsoisplottedinFigPermutationwasunabletocreateanyassociationswithrˆatoroverorunderThus,associationsfoundintheoriginaldatasetwithabs(rˆ)atwerereproducedbypermutationinlessthanoftrialsandwereviewedashighlyunlikelytobegeneratedthroughrandomchance(ie,asignalsubstantiallystrongerthannoise)Thiswasusedtodeterminethethresholdabs(rˆ),inthatthethresholdneededtobeatorabovetomaximizethesignalstrengthovernoiseWefeelthisuseofpermutationtoguidetheanalysiswascriticalForexample,previouslypublishedreportsonalternativeanalysesofthisdatahighlightedassociationswithrat,whichiswellwithintheattainablerangethroughrandompermutation()Withthethresholdabs(rˆ)setto,therewereconstructedrelevancenetworks,containingfeaturesand,associations(Fig)ThemajorityofassociationswerebetweenpairsofmeasuresofanticanceragentsusceptibilityDespitethelargenumberofassociationsshown,thisrepresentsfewerthanofthe,,totalpairwisecomparisonsThediagramistoosmalltomakeoutspecificdetailsandenlargedversionsofallofthefiguresalongwiththedescriptionsforeachaccessionnumberareavailableathttp:yywwwchiporgygenomicsTherelevancenetworksweregraphicallydisplayedbyusingnodestorepresentgenesandanticanceragents,andlinksbetweennodestorepresenthypotheticalfunctionalrelationshipsbetweenfeaturesThegraphicallayoutofrelevancenetworkswasautomaticallygeneratedbyusingtheGRAPHEDITORTOOLKIT(TomSawyerSoftware,Berkeley,CA)SevenspecificnetworksareshowninTableFigAdatabaseofbaselineexpressedlevelsof,genesincancercelllineswasjoinedwithadatabasecontainingtheamountsof,anticanceragentsneededtoinhibitgrowthofthosesamecelllinesThejoineddatabasecontained,featuresmeasuredincelllinesThefeaturesthatdidnotcontainasufficientrangeofvalueswereremoved,usinganentropybasedmethoddescribedinthetextTheremaining,featureswerecomparedagainsteachotherinapairwisemannermaking,,pairs,inanefforttofindanticanceragentsusceptibilitypatternsandgeneexpressionpatternsthatwerecorrelatedwitheachotherThedistributionofcorrelationcoefficientsisshownhere(rˆsignifiesrretainingthesign,positiveornegative,ofr)Foreachfeature,geneandsusceptibilitymeasurementswererandomlypermutedtimesTheaveragedistributionofrˆforeachpermutedsetisshownwitherrorbarscoveringtwostandarddeviationsRandompermutationwasunabletocreateanassociationwithrˆatoroverorlowerthanButteetalPNASuOctober,uvolunouGENETICSWecategorizedtheassociationswefoundinthenetworksintoataxonomyofthreetypes:identityorsynonymy,derivation,andbiologicrelationshipSpecificClustersFoundThroughAnalysisofRNAExpressionandAnticancerAgentSusceptibilityFifteenofthenetworksdemonstratedsynonymytypeassociationsoftheselinkedtheexpressionofRNAusedasendogenousorspikedcontrolsintheAffymetrixHUarrayFourofthesenetworkslinkedgenesthatwerelistedundermultipleGenBankaccessionnumbers:SRP(LandD),tropomyosinalphachain(MandZ),smallnuclearribonucleoproteinB(XandX),andnicotinamideNmethyltransferase(UandU)Onenetworklinkedexpressionlevelsoflamininreceptorprecursor(M)andlamininreceptormRNA(U)Thesesynonymynetworksactasapositivecontrol,inthatmeasurementsfromsimilarsetsofprobepairsshouldbesimilar,andtheexpressionpatternsshouldbehighlycorrelatedOnehundredseventyeightofthenetworkslinkedanticanceragentsexclusively,oneofwhichisshownasnetworkinTableThemajorityoftheseassociationswerebetweenoneanticanceragentandanothercompoundchemicallyrelatedtoorderivedfromthefirstThelargernetworkshadassociationsbetweenmanycompoundswithsimilarmechanismsofbiologicalaction(forexample,thealkylatingagents)Theremainingnineofthenetworksshowedassociationsofthethirdtype:thosesuggestingpotentialbiologicalrelationshipsSixofthesearelistedinTableasnetworks–Onenetwork(notshown)linkedmelanomaassociatedantigens,,andThesethreegenesareexpressedinmelanomaandseveralothermalignanttumorsandshareahighdegreeofsequencesimilarity()Anothernetwork(notshown)linkedcaldesmonandalternativesplicingproductsandandathird(notshown)linkedtworelatedsequencesfrommajorhistocompatibilityclassI(DandX)InTable,networkcorrectlylinkedkeratinand,twointermediatefilamentproteinsKeratinisatypeI(acidic)keratinandkeratinisatypeII(neutralybasic)keratin(),whichareknowntobecoexpressedandfunctiontogethertostabilizeeachotherfromdegradation(,)KeratinsanddonothaveasignificantlysimilarsequenceNetworknegativelylinkedglycoproteinIbbeta,whichisacomponentoftheplateletreceptorforvonWillebrandfactor,andpsd,whichcontainsSecandpleckstrinhomologydomainsGlycoproteinIbbetaisacomponentofareceptorinvolvedintheearlystagesofhemostasisandisknowntointeractwithsignalingprotein–zeta,whichalsocontainspleckstrinhomologydomains(,)ThislinkrepresentsahypothesisthatpsdrepresentsanotherproteininvolvedinthesignalingcascadefromthevonWillebrandfactorreceptorPutativeLinkBetweenaSingleGeneandAnticancerAgentSusceptibilityAtathresholdabs(rˆ)of,onlyonenetworkcontainsanassociationbetweenageneexpressionandameasureofanticanceragentsusceptibility,andthisnetworkislabeledinTableTheassociationisbetweenthegenecodingforlymphocytecytosolicprotein(LCP,pp,orLplastin,UniGeneHs),andtheanticanceragentNSC(thiazolidinecarboxylicacid,(oxo(phenylthio)ethylcyclohexenylacetylthioxo,methylester,Ra(R*),a(CI))LCPisanactinbindingproteininvolvedinleukocyteadhesion()whoseregulationissteroidhormonereceptordependent()AspecificroleforLplastinintumorogenicityhasbeenpostulatedlowlevelexpressionofLplastinisthoughttooccurinmosthumancancercelllines()Itishypothesizedthatphosphorylationofthisproteinmayregulatelymphokineactivatedkillercelladhesiontotumors()ProstatecarcinomainvasionisdecreasedwhenlevelsofLplastinaresuppressed()ExpressionofTplastin,arelatedgene,isincreasedincisplatinresistantcelllines()AlthoughthereisnoknownrelationshipbetweenthisspecificanticanceragentandgeneFigRelevancenetworksconstructedfromthejoineddatabasesofbaselinegeneexpressionincancercelllinesandmeasuresofsusceptibilityofthesamecelllinestoanticanceragentsThepairsoffeatures(anticanceragentsingreenboxes,genesinwhiteboxes)withrˆatorgreaterthanweredrawnwithlinethicknessproportionaltorˆFeatureswithoutanassociationatwereremovedAssociationswithnegativerˆareinredSevennetworksarehighlightedinorangeandareinTableLargeversionsofallfiguresanddescriptionsforeachaccessionnumbermaybefoundathttp:yywwwchiporgygenomicsuwwwpnasorgButteetalinthebiomedicalliterature,otherthiazolidinecarboxylicacidderivativesareknowntoinhibittumorcellgrowth()TheGIofagentNSCwasfoundtoincreaseincellsexpressingmoreLCPAscatterplotoftheRNAexpressionofLCPversustheGIofcelllinesagainstagentacrosscelllinesisshowninFigthecalculatedrˆwasDiscussionUsingrelevancenetworks,agenecanbedirectlyorindirectlylinkedtoseveralgenesaswellasphenotypicmeasurements,suchasmeasuresofanticancersusceptibilityRelevancenetworksdisplaynodeswithvaryingdegreesofcrossconnectivityIntheextreme,thesearecliques,whereeverynodeiscrossconnectedwitheveryothernodeinanetworkAnexampleisinnetworkinTable,wherefiveanticanceragentswithsimilarmechanismsofactionwerehighlycrossconnectedThesehighlycrossconnectednetworksofnodesrepresentfeaturesthatarenotonlyassociatedpairwise,butalsoinaggregateTheyrepresentthemosttrustedassociationsPhylogenetictypetreescanonlylinkeachfeaturetooneotherfeature,typicallytheoneitismoststronglycorrelatedwith,anddonotdisplayadditionallinks()Inaddition,phylogenetictypetreescannoteasilyclusterdisparatetypeso

用户评价(0)

关闭

新课改视野下建构高中语文教学实验成果报告(32KB)

抱歉,积分不足下载失败,请稍后再试!

提示

试读已结束,如需要继续阅读或者下载,敬请购买!

评分:

/5

VIP

在线
客服

免费
邮箱

爱问共享资料服务号

扫描关注领取更多福利