|
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Full Text | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
PAGE 1 MICRORNAREGULATORYNETWORKS: ANALYZINGSTRUCTURALCHANGESIN TIMESERIESDATA BY RAYMONDV.ROBERTSIII AThesis SubmittedtotheDivisionofNaturalSciences NewCollegeofFlorida InpartialfulllmentoftherequirementsforthedegreeofBachelor ofArts UnderthesponsorshipofDr.ChristopherHart Sarasota,FL May,2011 PAGE 2 Acknowledgements IwouldliketothankLovelaceRespiratoryResearchInstituteLRRIforprovidingthedatauponwhichmanyofmyideasweretestedandforfundingmy researchinNewMexico.WithoutthepartnershipbetweenLRRIandNewCollegeofFloridaIwouldnothavebeenabletowritethisthesis.Ofcourse,that partnershipwouldnothavebeenpossiblewithoutDr.RobertRubinandDr. KevinHarrod.Dr.Rubin,asthepresidentofLRRIhasbenetedbothmyself andNewCollegeofFloridawiththispartnership. Dr.PatrickMcDonaldwasmyadvisorformuchofmytimeatNewCollege,his supportandguidanceduringthattimewasinstrumentalinhelpingmediscover mypassionfordata,graphs,andalgorithms.ItwashisclassonNetworks,Graphs, andAlgorithmsthatrstallowedmetothinkaboutproblemsintermsofgraphs and,mostimportantly,intermsofcomputationalcomplexity. IoweDr.KatherineWalstromagreatdealforagreeingtositonmythesis committeeatthelastmoment.Theexpertiseshebringstothecommitteeis greatlyappreciated. Myparentshavebeensupportiveandunderstandingthroughoutmycollege career.Thereisnotmuchmoreasoncanaskfor. DavidWydehasbeenawonderfulfriendandimpromptuprogrammingmentor. Ihavelearnedagreatdealaboutprogrammingandespeciallypythonfromhim. ii PAGE 3 ShellyDuFordhasbeenawonderful,supportivepartnerandhasbeenkindenough tolistentomeworkthroughmanyofmyideasatalltimesofthedayandnight.I wouldalsoliketothankallofthefriendsIhavemadealongtheway,Zeke,Chase, Justin,Wyatt,Kathleen,Datev,Alli,Ned,Alberto,Jack,Geo,andmanyothers. YouhaveallmademytimeatNewCollegeatrulyamazingexperience. Finally,I'dliketothankDr.ChrisHart,mythesisadvisor.Dr.Harthas beenimmenselysupportivethroughouttheprocessofresearchingandwriting thisthesis.TheenthusiasmandpositivityDr.Hartbroughttoeverymeeting werekeyinhelpingmeproducethisthesis.IneachnewndingDr.Hartsees morequestionsandopportunitiesforexpandingourunderstanding. iii PAGE 4 MICRORNAREGULATORYNETWORKS: ANALYZINGSTRUCTURALCHANGESINTIME SERIESDATA RAYMONDV.ROBERTSIII NewCollegeofFlorida,2011 Abstract MicroRNAsmiRNAsare 22nucleotide,non-coding,single-strandedRNA molecules.MicroRNAsarepost-transcriptionalregulatorsofmessengerRNAs mRNAsthatcontroltranslationthrougheithertranslationalrepressionordirectingmRNAdegradation.TheexpressionpatternsofmicroRNAsareofinterest tobiologistsastheseregulatorymoleculeshavebeenfoundtobesignicantin awiderangeofbiologicalprocesses,includingdevelopment,celldierentiation, anddiseaseprocesses.Onecomputationalapproachtoexaminingthisregulatory networkisthroughtheuseofadirectedgraphmodelofmicroRNAtomRNA targeting.Inthisapproach,agraphisconstructedsuchthatitisbothdirected andbipartite,limitingtheapplicationofmanyalgorithmsdevelopedtoanalyze complexrealworldnetworks.WeproposeanewrepresentationofmicroRNA regulatorynetworkswhichallowsfortheexpressionofcooperativemicroRNA regulationofmRNAsbymappingmRNAnodestoedges.Thiscreatesanundiiv PAGE 5 rectedgraphwithcyclesandallowsforeasierdetectionof"hubs",whichhasbeen showntobeusefulinunderstandingfunctionalaspectsofvariousotherbiomolecularinteractionnetworks.Inthisthesisweexplorethisnewrespresentationof themiRNAregulatorynetworkbyapplyingbetweennesscentralitytodetermine therelativesignicanceofeachmiRNAineachnetworkfromatimecourseexperiment.WecomparethechangesinbetweennesscentralitywiththemiRNA expressiondatagatheredfromanH1N1infectiontimecourse. ProfessorC.Hart DissertationCommitteeChair v PAGE 6 Contents Acknowledgementsii Abstractiv ListofFiguresviii ListofTablesx 1Introduction1 1.1Background :::::::::::::::::::::::::::::: 1 1.1.1Biogenesis ::::::::::::::::::::::::::: 2 1.2PredictingmiRNATargets :::::::::::::::::::::: 6 1.2.1MicroRNASeeds ::::::::::::::::::::::: 7 1.3Motivation ::::::::::::::::::::::::::::::: 9 1.3.1NetworkStructure :::::::::::::::::::::: 10 1.3.2Scale-FreeNetworks ::::::::::::::::::::: 11 2Methods14 2.1BuildingGraphsusingPython :::::::::::::::::::: 14 2.1.1DatastructuresandParallelism ::::::::::::::: 15 2.2Inuenzaexperiment ::::::::::::::::::::::::: 17 2.3AnalysisofthemiRNA:miRNAGraphs ::::::::::::::: 18 2.3.1GraphDensity :::::::::::::::::::::::: 18 2.3.2ShortestPath ::::::::::::::::::::::::: 19 2.3.3BetweennessCentrality :::::::::::::::::::: 21 2.3.4PeaktoPeakscore :::::::::::::::::::::: 22 2.4RandomizationAnalysis ::::::::::::::::::::::: 23 2.5TargetScan :::::::::::::::::::::::::::::: 23 vi PAGE 7 3ResultsandDiscussion26 3.1ExperimentalAnalysis :::::::::::::::::::::::: 26 3.2NetworkAnalysis ::::::::::::::::::::::::::: 28 3.2.1AverageDegree :::::::::::::::::::::::: 30 3.2.2GraphDensity :::::::::::::::::::::::: 32 3.2.3BetweennessCentrality :::::::::::::::::::: 35 3.2.4BetweennessVersusDegreeChange ::::::::::::: 38 3.2.5ExplorationofthePCTandContextGraphSpace :::: 39 3.2.6ConclusionandFutureResearch ::::::::::::::: 41 Bibliography43 vii PAGE 8 ListofFigures 1.1Thestructureofvepri-miRNAs :::::::::::::::::: 3 1.2MiRNAbiogenesis :::::::::::::::::::::::::: 4 1.3MicroRNAseedbindingtypes :::::::::::::::::::: 7 3.1DegreeanalysisofmiRNAandmRNAnodesthroughoutthetimecourse.TheinformationisdividedintomiRNAandmRNAwhichhave anticorrelatedexpressionvalues.Topa:out-degreeofmiRNAwithan increaseinexpressionvalue.Bottoma:out-degreeofmiRNAwitha decreaseinexpression.Topb:in-degreeofmRNAwithanincreasein expression.Bottomb:in-degreeofmRNAwithadecreaseinexpression.27 3.2changesinmiRNAdegreeT6toT24 :::::::::::::::: 28 3.3changesinmiRNAdegreeT6toT72 :::::::::::::::: 29 3.4changesinmiRNAdegreeT24toT72 :::::::::::::::: 30 3.5anti-correlationdirectedbipartitenetworkmapformiRNAupat time6 :::::::::::::::::::::::::::::::::::: 31 3.6AveragedegreeofeachgraphatthegivenPCTandContext. ::: 32 3.7ConsequencesofPCTandContextonAveragedegreeofthegraphs33 3.8Densityofthegraphsw.r.tpctandcontext ::::::::::::: 34 3.9MicroRNABetweennessCentralityPeaktoPeakScores :::::: 36 3.10Percentoverlapofthetop20%ofmiRNAnodesbybetweenness shiftwiththesetofmiRNAfromthemicroarraywithap-valueofless than0.05 :::::::::::::::::::::::::::::::::: 37 3.11PercentoverlapofallthemiRNAnodesfromthemiRNA:miRNA networkwiththesetofmiRNAfromthemicroarraywithap-valueof lessthan0.05asPCTandContextvary ::::::::::::::::: 38 3.12PTPofBetweennessCentralityVersusPTPDegree :::::::: 39 3.13Comparisonofthedatafromgure3.12withrandomizednetworks. Hereaplus+denotesrandomizeddataandaredcircleisactualdata.40 viii PAGE 9 3.14BinaryEdgeGraphDegreeStatistics :::::::::::::::: 40 3.15MultigraphDegreeStatistics ::::::::::::::::::::: 41 ix PAGE 10 ListofTables 2.1Samplebetweennesscentralitydataasstoredinastructurednumpy array ::::::::::::::::::::::::::::::::::::: 22 x PAGE 11 Chapter1 Introduction 1.1Background MicroRNAsmiRNAsaresmall,19-25nucleotide,non-coding,single-stranded RNAmoleculesgeneratedfromendogenoushairpin-shapedtranscripts[1,5,9]. ThroughinteractionswithspecicsequencesonmRNAs,miRNAscanpost-transcriptionally regulatetranslationofthemRNAsintoproteinbytranscriptdestabilization, translationalrepression,orboth[13].MicroRNAshavesignicantregulatory eectsinanimalsandplants.Recentstudieshaverevealedthesignicanceof miRNAsindevelopmentaltiming,haematopoieticcelldierentiation,apoptosis, cellproliferation,andorgandevelopment[22].Withhundredstothousandsof distinctmiRNAsexperimentallyidentiedacrossallmetazoanlife,thesesmall RNAsareclearlyanimportantandabundantclassgene-regulatorymoleculesin mammals[39,13,5].AsinglemiRNAcanpotentiallybindtomanymRNAas 1 PAGE 12 Chapter1.Introduction wellastargetasinglemRNAmultipletimes.Itisalsothecasethatmultiple miRNAscancooperativelyregulateasinglemRNA[31,13].Thiscooperative regulationcanaecttheextentandnatureoftranslationalrepression. 1.1.1Biogenesis Here,weattempttoprovideabriefoverviewofmiRNAbiogenesis. EarlyanalysisofannotatedmiRNAsrevealedthatmostarelocatedinintergenicregionsi.e., > 1kilobaseawayfromannotatedorpredictedgenes,however, asubstantialminoritywaspresentintheintronicregionsofannotatedgenesin thesenseoranti-senseorientation[26,25]. ThisndingledtothepostulationthatthemajorityofmiRNAgenesare transcribedasautonomoustranscriptionunits[22].Approximately50%were foundtobeincloseproximitytoothermiRNAs[26,25,35]. MicroRNAtranscriptioncanbecontrolledfromtheirownpromoters[29,8], individuallyoraspolycistronicprimarytranscriptspri-miRNAs[30],oraspart oftheintronicregionofanothertranscriptionalunit,suchasaproteincoding gene.Forexamplesofpri-miRNAsseegure1.1 MicroRNAcontainingtranscriptsarereferredtoaspri-miRNAs.Pri-miRNAs arerelativelylarge,sometimesseveralkilobasesinlength,poly-adenylated,with a5'cap.Oneormorehairpin-loopstructuresmaybepresentinapri-miRNA 2 PAGE 13 Chapter1.Introduction Figure1.1:Thestructureofvepri-miRNAs,fromV.N.Kim;MicroRNABiogenesis:CoordinatedCroppingandDicing;NatureReviews;2005. a :miRNAsin theexonicregionsofnon-codingtranscripts. b :miRNAsintheintronicregions ofnon-codingtranscripts. c :miRNAsintheintronicregionofaproteincoding transcript. gure1.1.MicroRNAgenesarebelievedtobetranscribedbypolymeraseIIpol II[29,8],thisiscontrarytowhatwasoriginallyexpected.PolymeraseIIIwas originallybelievedtoberesponsiblefortranscriptionofmiRNAgenesasmany othersmallRNAsaretranscribedbypolIII,e.g.,tRNAs.Theevidenceagainst polIIItranscriptioncamewhenitwasdiscoveredthatmiRNAsaretranscribedas 3 PAGE 14 Chapter1.Introduction partofapri-miRNA,sometimesseveralkilobasesinlength,makingitlikelythat fourUracilsinarowwouldoccur,astopsignalforpolIII[30].Furtherevidencefor Figure1.2:MiRNABiogenesisfromV.N.Kim;MicroRNABiogenesis:CoordinatedCroppingandDicing;NatureReviews;2005. 4 PAGE 15 Chapter1.Introduction transcriptionbypolIIcamefromexpressionanalysisofmiRNAs.Theanalysis revealedthatmiRNAsareunderelaboratecontrolduringdevelopmentandin varioustissues.Thisndingalignswithwhatisknownaboutothergenesthatare transcribedbypolII[27,7,24,2,20].Itisalsopossiblethatafullyfunctional miRNAcanbegeneratedfromaplasmidthatcontainsapri-miRNAsegment underthecontrolofaheterologouspolIIpromoter[22].Threemorepiecesof directevidencewerereported: 1.Pri-miRNAscontainbothCAPSTRUCTURESandPOLYATAILS[29,8]. 2.TranscriptionofmiRNAsissensitivetoalpha-amanitinatconcentrations thatspecicallyinhibitpolIIbutnotpolIorpolIII[29]. 3.Usingchromatinimmunoprecipitationanalysesthephysicalassociationof polIIwiththepromoterofmiR-23a 27a 24-2wasdemonstrated[29] Processingofpri-miRNAsoccursinthenucleuswhereDiGeorgeSyndromeCriticalRegion8DGCR8associateswiththeRNaseIIItypeenzymeDrosha.The pri-miRNAprocessingcomplexreliesprimarilyonthetertiarystructureofthe pri-miRNAforsubstrate-specicity[28,11,45].Seegure1.2. AfterprocessingbytheDGCR8-Droshacomplexweareleftwithpre-miRNAs whichconsistoftheprocessedstem-loopstructure.Thepre-miRNAsareexported tothecytoplasmbyexportin-5[34,43,6].Experimentalanalysisindicatesthat 5 PAGE 16 Chapter1.Introduction pre-miRNAsdonotaccumulateinthenucleusafterdepletionofexportin-5,suggestingthatpre-miRNAsmightberelativelyunstableandhencestabilizedby exportin-5[43]. Oncethepre-miRNAhasbeenexportedtothecytoplasm,processingbythe RNaseIIIenzymeDicertransformsthepre-miRNAintoa 22ntmiRNAduplex. Strandselectionisbelievedtooccurbasedonthethermodynamicstabilityofthe twoendsofthemiRNAduplex,leavingthestrandwithrelativelystablepairing onthe5'endbehind[40,21]. MaturemiRNAarethenincorporatedintoeectorcomplexescalledmiRNPs miRNA-containingribonucleoproteincomplex.Othernamesinclude'mirgonaute' thecombinationofmiRNAwithargonaute,theproteinatthecoreofthemiRNP and'miRISC'miRNA-containingRNA-inducedsilencingcomplex[22].Once themiRNAisincorporatedintotheeectorcomplexbindingtothe3'UTRofan mRNAcanoccur,leadingtotranslationalrepressionormessagedegradation. 1.2PredictingmiRNATargets ThepredictionofmiRNAbindingsiteshasproventobearatherdicult problem.However,sometechniqueshavebeendevelopedwhichprovidereasonabletoolsforndingandrankingmiRNAbindingsitesintermsofseveralkey 6 PAGE 17 Chapter1.Introduction Figure1.3:MicroRNAseedbindingtypesfromFriedman,et.al.;MostmammalianmRNAsareconservedtargetsofmicroRNAs;GenomeResearch;2009 parameters.ThemostimportantparameterinthepredictionofmiRNAtargeting isthe5'endofthemiRNA,knownastheseed.Theseedconsistsofnucleotides 2-6gure1.3. 1.2.1MicroRNASeeds Thesignicanceofthe6nucelotideregiononthe5'endofthemiRNAwas determinedthroughcomparativesequenceanalysis.Thiswasdonebyexamining 7nucelotidesegmentsspanningthelengthofthemiRNA.Thesegmentscorrespondingtotheseedregionnucleotides2-7onthe5'endweretheonlyones conservedmorethanexpectedbychance[31]. 7 PAGE 18 Chapter1.Introduction Theabilitytodowhole-genomealignmentshasmadeitpossibletodetectsite conservationinorthologouslocationsofgenes,however,itisamoredicultproblemtodistinguishbetweensitesthatareconservedbychanceversusconserveddue toselectivepressures.Z-scoretestswereusedtodetectpreferentialconservation ofmotifs,howeveritdidnotcontrolforgenomiclocationorsequencecharacteristics[42].Analternativeapproach,developedfordetectingmaintenanceof miRNAsites,hasbeentogeneraterandomsetsofmiRNA-likesequencesand thendeterminethenumberofconservedsitesmatchingthesecontrolsequences andusethisasanestimateofchanceconservation[31].Furtheranalysisofthe seedregionuncoveredthesignicanceofnon-Watson-CrickrecognitionofanA attherstnucleotideaswellasanAorUacrossfromnucleotide8[32,36]. Fourtypesofseed-matchedsitesareknowntobeselectivelyconserved[32] 1.6mersite:Perfectlymatchesthe6ntmiRNAseed. 2.7mer-m8site:SeedmatchplusaWatson-Crickpairingtotheeighthnucleotide. 3.7mer-A1site:TheSeedmatchplusanAdenosineacrossfromtherstnucleotide. 4.8mersite:Thecombinationofallthree;aseedmatchplusm8plusA1.... Thefollowinghierarchydescribesthebindingecacyofeachseed-matchtype: 8mer > 7mer-m8 > 7mer-A1 > 6mer[15,36] 8 PAGE 19 Chapter1.Introduction 1.3Motivation IntheclassofRNAbasedgene-regulators,miRNAsareamongthemostabundantmolecules[39,13].Theextensiveresearchconductedoverthelast20years onmiRNAshasbeguntorevealthesignicanceoftheseregulatorymolecules inmanyimportantcellprocesses.However,asinmanybiologicalinquiries,the abilitytounderstandthesystemasawholeremainselusive.Currentbiological andcomputationalapproaches,whileimproving,failtocapturethecomplexity ofmiRNAregulatorynetworks.Inordertobetterunderstandkeycellprocesses, particularlydiseaseanddevelopmentalprocesses,wemustdevelopbettermodels andtechiniquesforanalyzinganddecomposingthesenetworks. Fortunately,advancesoverthelastdecadeinmeasuringexpressionlevelsof miRNAandmiRNAtargetshaveresultedinexperimentswhichprovideever wideningsnapshotsoftheprocessesoccurringinthecell.Withtheabilityto measureexpressionlevelsofmostknownandpredictedmiRNAs,aswellasthousandsofmRNAs,wecannowexaminethedynamicsofthesenetworksasawhole atseveraltimepointsacrossvariedbiologicalprocesses.Webelievethatbyexamingthechangesinthestructureofthesenetworksovertime,itwillallowusto identifyfeatureswhichhavepreviouslygoneunnoticed. 9 PAGE 20 Chapter1.Introduction Inparticular,wewillinvestigateaspartofthisworkthedynamicstructural changestothemiRNA:mRNAregulatorynetworkwhenhumanlungepthelial cellsareexposedto2009H1N1inuenza.Thisispartiallymotivatedbyastudy ofreconstructed1918inuenzavirusinfectioninmouse.Itwasfoundthatan atypicalgeneexpressionprole,ascomparedtoTx/91infection,maybecorrelated withtheextremevirulenceofthevirus[33].Inalaterstudybythesamegroup, usingarchivedmouselungtissue,ananalysisofthemicroRNAexpressionprole suggestedthatdierentialexpressionofmicroRNAsplayedaninterestingand signicantfactorinthephysicalmanifestationofthediseasesstudied.These observationssuggestthatinuenzainfectionmaydirectlyorindirectlyuseor manipulatethecellsendogenousmiRNAregulatorymechanisms. Wepresentherethedevelopmentofanewgraphabstractionthatallowsfor theapplicationofseveralcommongraphmetricstothesenovelrepresentationsof miRNAregulatorynetworksthatmayhelpelucidatefunctionalinteractionsthat areperturbedduringinuenzainfection. 1.3.1NetworkStructure Networks,or,moreprecisely,graphs,areabstractionsofwhatarebelievedto beinteractionsedgesbetweenobjectsnodes.Formanyyearstheunderstandingofrealworldnetworkswasbasedonthesimplerandomgraphmodel[38]. 10 PAGE 21 Chapter1.Introduction Therandomnetworkmodelexplainedmanyoftheobservedpropertiesofreal worldnetworkssuchasthesmallworldpropertytheshortestpathbetweenany twonodesinthenetworkislessthanlength6.Onlyrecentlywasitdiscovered thatmanyrealworldnetworksareinfactnotrandom,theyarescale-free.See section1.3.2.InourabstractionofmiRNAregulatorynetworkswedesignedour structurestobeeasilyanalyzedusingexistinggraphalgorithms.Forourgraph, G = V;E ,where V isthevertexsetofmiRNAsand E istheedgesetof sharedtargetsweconstructitasfollows 1.deneaone-to-manymapping, T : m i M where m i ismiRNA i andMisthesetofmRNAwhicharepotentialmiRNA targets.Then T m i M 2. m i isadjacentto m j ifandonlyif T m i T T m j 6 = ; Thesignicanceofinteractionsinbiologicalnetworksisastrongmotivatorfor examiningthestructuralchangesinthesegraphsovertimeseriesdata. 1.3.2Scale-FreeNetworks Thediscoveryofthescale-freeSFnetworkstructure[4]drasticallyaltered ourunderstandingofrealworldcomplexnetworks.TheInternet,socialnetworks, 11 PAGE 22 Chapter1.Introduction protein:proteininteractionnetworks,scienticpaperslinkedbycitations,andweb pagelinksareallscale-freenetworksasaremanyothers.SFnetworkshave degreedistributions P k ck )]TJ/F22 7.9701 Tf 6.587 0 Td [( where c issomescalarand P k isthefractionofnodesinthegraphwith k connections.With typicallyintherange2 3. Thehubinascale-freenetworkisdenedasthenodewiththelargestdegree. However,itiscommontorefertomultiplehubsinscale-freenetworks,wherea hubcanmeananynodeinthetailofthedegreedistributionthosenodeswith exceptionallyhighdegree.Scale-Freenetworksarepaticularlyrobustagainst randomnoderemoval,however,targetedhubremovalisextremelydisruptive[19]. Forinstance,theknockoutofhubsintheyeastprotein-proteininteractionnetwork resultsinanapproximatelythreefoldincreaseinlethalityovernon-hubs[19,41, 14].ThesendingssuggestthatifmiRNAnetworksareSFthehubsshould beofsignicantinterestasanydierentialexpressionofahubislikelytohave signicanteectsontheregulationofmRNAs. Inadditiontothebroaddenitionofahubinascalefree-network,thereare twospecichubtypeswhichhavebeenfoundinotherbiologicalnetworks[17]. Partyhubsinteractwiththeiradjacentnodesinthesametimeandspace.Date 12 PAGE 23 Chapter1.Introduction hubsinteractwiththeirneighborsindierenttimeand/orspace[17].Partyhubs inbiologicalnetworkstendtohavesimilarexpressionpatterns,whereasdatehubs exhibitlimitedco-expression,suggestingtheinteractionsoccuratdierenttimes and/orplaces[17]. 13 PAGE 24 Chapter2 Methods 2.1BuildingGraphsusingPython Aswiththemajorityoftheanalysis,thecodetocreatethesegraphsiswritten inpythonandisavailablewiththesupplementarymaterial.Thenetworkx[16] module,acollectionofgraphstructuresandalgorithms,isusedastheprimary computationaltool.AsanoteforthosefamiliarwithPython,graphsarebasedon thebuilt-indictionarystructure,ahashtablebasedcollectionofkey:valuepairs. Networkximagesarerenderedusingthepygraphvizmodule,apythonwrapper forthegraphvizpackagedevelopedatAT&Tlabsthatprovidesacollectionof graphvisualizationalgorithmsandtools. MostofthepreliminaryanalysisandconstructionofthemiRNA:miRNA graphswasdoneusingnumpyarrays[3,37]. 14 PAGE 25 Chapter2.Methods 2.1.1DatastructuresandParallelism DuetothehighcomputationtimeinvolvedinbuildingmiRNA:miRNAgraphs, O N 2 whereNisthenumberofmiRNAnodesinthegraph,wewilllayoutthe methodusedforconstructingtheadjacencymatrices. Themostimportantpackageusedinthecomputationisnumpy,anumerical computationpackagewithbuilt-inarraydatatypesthatcancontainmixeddata suchastextandoats.ThecoreofnumpyiswritteninCandisthereforeusefulin reducingtheleadingcoecientontheruntime.AreferencemiRNAdictionary isconstructedthatkeepstrackofthemiRNAassociatedwitheachrowinthe adjacencyarray. TherstarrayweconstructedwasastructuredarrayTheideabehindthis arrayissimilartoarelationaldatabaseofallmiRNA:mRNArelationshipswith ProbabilityofConservedTargetingPCTandcontextscorestakenfromthe targetscandata.Thiswasan mxn array,miRNAintherows,mRNAinthe columns.IfmiRNA i targetsmRNA j weinsertthecorrespondingPCTand contextscoresintothearrayatpositioni,j.Thisstructureallowedustoquickly querythedata,creatingabooleanarraycorrespondingtothesetofedgeswhich passourthresholdvalues.Usingthenumpystructuredarrayswehavetheability tomakelogicalstatementsusing&andand j orrelations. Array['pct'] PctThreshold&Array['context'] > ContextThreshold 15 PAGE 26 Chapter2.Methods ByiteratingovertherangeofPCTandcontextscoreswewereabletoconstruct amappingoftheseconnectionsforresolutionsaslowas0 : 02.Thereturnof eachiterationisabooleanarraywhereaTruevalueindicatesthatthegiven miRNA:mRNAconnectionmetourthresholdrequirements.Usingthefactthat arrayoperationsinnumpyareelement-wiseandmultiplicationofbooleanvalues isequivalenttoalogical'and'operationwewereabletoquicklydopairwisetarget comparisonsforthemiRNA.TakethebooleanarrayreturnedbythePCTand contextqueryandforeachrowiterateovereveryotherrowanddoanelementwise multiplicationoftheTrue/Falsestatements.Ifthesumofthismultiplication, whichisequivalenttoan"and"logicaloperation,thenthemiRNAcorresponding toeachrowareconnectedbyanedgeoredgesifdoneforamultigraph.To furtherspeeduptheanalysiswetookadvantageoftheinherentparallelismofthe forloopsanddistributedtheproblemacrossthemaximumnumberofavailable cores.Theuseofthesedatastructures,alongwithparallelprocessing,ultimately reducedthecomputationtimefromoneweektolessthananhour. Usingastructurednumpyarraywewereabletoquicklyquerythetargetscan data.Thiswasdoneusingnumpy'sbuiltinstructuredarrayandthe&operator. 16 PAGE 27 Chapter2.Methods 2.2Inuenzaexperiment Theexperimentaldatausedherewasgatheredfromnormalhumanbronchial epithelialNHBEcellswereculturedanddierentiatedinvitro,usinganairliquidinterfaceculturesystem.Dierentiatedcellswerethenexposedtothe2009 H1N1swineorgininuenzavirus.TotalRNAwasextractedfromcellsat0,6,24, and72hourspostinfection.Small < 200ntRNAandlargeRNAwereseparated andpreparedindependentlyformicroarrayanalysis. MeasurementswereperformedintriplicateusingIlluminabeadarrays.The triplicateswerethensummarizedusingthreedierentsummarizationalgorithms, RMA16,PLIER16,anditerativePLIER16.Valueswerethenlog 2 transformed anddividedbythemedianofthecorrespondingcontrolsample.Probeswerelteredbasedontheirrawexpressionvaluesandp-valuescoreinaonewayANOVA. Ifanexpressionvalueof30wasseenin3of12samplesandthep-valuewas 0 : 05, theprobewasretained.Finally,iftheprobeswereintheintersectionofatleast twoofthelteredsetsbasedonthedierentsummarizationalgorithms,they wereretained. 17 PAGE 28 Chapter2.Methods 2.3AnalysisofthemiRNA:miRNAGraphs WeconstructedmiRNAtomiRNAgraphsbymappingmRNAtargetstoedges, asdescribedinsections1.3.1and3.2.Forexample,iflet-7bandmiR-210both potentiallybindtoEFNA3wedrawanedgebetweenlet-7bandmiR-210.This transformationallowsamoredirectevaluationofcooperativeregulationbymiRNAs.Binaryandmultigraphswereconstructed.Binarygraphsaresimplythe multigraphswithedgescompressedtoasinglerepresentativeedge. 2.3.1GraphDensity ToexaminetheconnectednessofthemiRNA:miRNAgraphswemeasurethe densityofeachgraph, G = V;E .Thedensity, Den G ,isdenedas Den G = 2 j E j j V j j V j)]TJ/F16 11.9552 Tf 17.933 0 Td [(1 Where j E j isthesizeoftheedgeset,orthenumberofmRNAswithunique miRNAregulationand j V j isthesizeofthevertexset,orthenumberofmiRNA whichareadjacenttoatleastoneothermiRNAdegreegreaterthanzero.Densityrangesbetweenzeroandone,whereavalueofzeroindicatesacompletely disconnectedgraphandavalueofoneindicatesacompletelyconnectedgraph. Multigraphscanhavedensityhigherthanone. 18 PAGE 29 Chapter2.Methods 2.3.2ShortestPath Givenagraph, G = V;E ,wedeneaweightfunction w : E R .IfGisan unweightedgraph, w : E !f 1 g .Deneapathfrom s 2 V to t 2 V asasequence ofvertices f v 0 = s;v 1 ;v 2 ;:::;v n = t g andthereisanedge e i;i +1 = v i ;v i +1 2 E connectingthe i thvertexinthesequencetothe i +1thvertex.Thelength,or weight,ofthispathisdenedas n )]TJ/F19 7.9701 Tf 6.587 0 Td [(1 X i =0 w v i ;v i +1 Giventhedenitionofapath,canwendtheshortestpathconnectingany twonodes.Let d G s;t betheshortestpathsbetween s and t .Then d G s;s =0, d G s;t = d G t;s ,and,given v 2 V d G s;t = d G s;v + d G v;t ifandonlyif v liesonashortestpathbetween s and t .Wecanenumerateallpossiblepathsin thegraph,alongwiththeirweights.However,weonlywishtondthepaththat takesusfromsomestartingnode, s toaterminalnodet.Thus,wecanelminatea signicantnumberofthepathsfromthesolutionspacebyimposingtherestricion thatthestartnodebe s andtheterminalnodebe t .However,thisstillrequires thatwendallpossiblepathsbetween s and t .Wenotethatgivenasolutionto theshortestpathproblemfortwonodes, s and t ,thereisarecursiverelationship tobefoundinthedenitionof d G ,theshortestpath. 19 PAGE 30 Chapter2.Methods Givenashortestpathd G s;t = f v 0 = s;v 1 ;:::;v n = t g d G s;t = d G s;v n )]TJ/F19 7.9701 Tf 6.587 0 Td [(1 + d G v n )]TJ/F19 7.9701 Tf 6.586 0 Td [(1 ;v n =d G s;v n )]TJ/F19 7.9701 Tf 6.586 0 Td [(2 + d G v n )]TJ/F19 7.9701 Tf 6.587 0 Td [(2 ;v n )]TJ/F19 7.9701 Tf 6.586 0 Td [(1 + d G v n )]TJ/F19 7.9701 Tf 6.587 0 Td [(1 ;v n = ::: =d G s;v 1 + ::: + d G v n )]TJ/F19 7.9701 Tf 6.586 0 Td [(1 ;v n Wecannowseethatthereisasimplepartitioningoftheproblemwhich allowsustoeasilyprovideasolution.Ifwesolvetheshortestpathproblem bydecomposingitintosmallersubproblemsforwhichwecanprovideoptimal solutionswecanprovideasolutionfortheactualproblem.Thisisadynamic programmingsolution. TherstandmostwellknownsolutiontotheshortestpathproblemisDijkstra'sAlgorithm[10],namedafterit'sauthor,EdsgerW.Dijkstra.Theworstcase performanceofDijkstra'salgorithmis O j E j + j V j log j V j [12]whenimplemented usingamin-priorityqueue.Dijkstra'soriginalimplementationhasaworstcase performaneof O j V j 2 .Dijkstra'salgorithmbelongstotheclassofbreadth-rst searchalgorithms. 20 PAGE 31 Chapter2.Methods 2.3.3BetweennessCentrality TomeasuretherelativesignicanceofagivenmiRNAinournetworkwe proposeusingthebetweennesscentralityofeachnodemiRNA v C B v ,which isdenedas C B v = X s 6 = v 6 = t st v st where st denotesthetotalnumberofshortestpathsbetween s and t and st v denotesthenumberofshortestpathsbetweensandtwhichpassthrough nodev.InnetworkxshortestpathsarecomputedusingDijkstra'salgorithm,a branchandboundmethod.Thismeasureofsignicanceisusedofteninanalyzing socialnetworksandcitationsofpapers.Betweennesscentralityhasevenbeenused inmodelstounderstandthespreadofdiseasessuchastuberculosis[23]. Itisimportanttonotethatinacompletelyconnectedgraph,wherethedistancefromanynodetoanyothernodeisexactlyone,thebetweennesscentrality iszero.Inouranalysis,wetreatednodeswhichwerenotconnectedatsometime pointbutwereatothersashavingabetweennesscentralityofzero,astheyare disconnectedfromthegraph,thusappearingonzeroshortestpaths. 21 PAGE 32 Chapter2.Methods miRNA T6 T24 T72 hsa-miR-221 0.0 1.521472859e-05 0.000361309 hsa-miR-103 0.001248625 0.004269102 0.000403236 hsa-miR-373 0.000790656 0.002003904 0.001616997 hsa-miR-372 0.000790656 0.002734683 0.002022384 hsa-miR-107 0.001248625 0.004269102 0.000403236 . . . Table2.1:Samplebetweennesscentralitydataasstoredinastructurednumpy array 2.3.4PeaktoPeakscore AftercomputationofthebetweennesscentralityforeverynodeineverynetworkateverytimepointweareleftwithanarraystructuredlikeTable2.1.One foreachcombinationofPCTandContext. OncewehavethisinformationwecomputethePeaktoPeakscoreptpfor eachmiRNAnodesvectorofbetweennesscentralityscores, n i .Thisisasimple measurewhichisdenedas ptp n i = max n i )]TJ/F21 11.9552 Tf 11.956 0 Td [(min n i ThemaximumbetweennesscentralityscoreattainedbymiRNA i minusthe minimumbetweennesscentralityscoreattainedbymiRNA i .Thismeasureallows ustoeasilydeterminethemiRNAwiththegreatestcentralityshiftsoverthetime course. 22 PAGE 33 Chapter2.Methods 2.4RandomizationAnalysis TocomparesomeofourresultsweconstructedmiRNA:mRNAnetworksidenticaltotheonesusedintheconstructionofthemiRNA:miRNAnetworks.However,beforethetransformationintothemiRNA:miRNAnetworksweshued edges.ThismethodallowedustopreservethemiRNAtargetdistributions.The actualimplementationwasdonebyshuingtheelementsineachrowoftheadjacencymatrixforthedirectedmiRNA:mRNAgraph.Eachrowwasshuedten times.Therandomizednetworkswerecompute100timesandthenthesame analysiswasperformed.Betweennesscentralityanddegreeshiftsweremeasured andthen"scattered". 2.5TargetScan WeutilizethemiRNAbindinginformationgeneratedbytheTargetscanalgorithm.TheTargetscanmethodwasthersttoaccountforthevariablerates ofevolutionarychangeacrossthegenome[13].Mutation,gene-conversionand crossoverratesvarythroughoutthegenome,thusaectingthe3'UTRsinvastly dierentways.TheGCcontent,dinucleotidecontent,theinterrelationofmiRNA seed-matchtypes,genomealignmentquality,andthelocalconservationratewere allfactorsaccountedforintheTargetscanmethodology.Theeectsofthesefac23 PAGE 34 Chapter2.Methods torsonthebackgroundconservationslevelsofshortk-merswereaccountedfor usingempiricalobservationsratherthantheoreticalcalculations[13].Thereare threekeyfactorsusedbyTargetscantodeterminemiRNAtargeting.Therstand mostsignicantistheseed-match.Usingonlythe5'regionofmiRNAs,known astheseed,targetscanbepredictedabovethenoiseoffalse-positives[31].Second,ProbabilityofConservedTargetingPCT,ameasureofhowwellaspecic miRNAaswellasitspotentialbindinglocationsareconserved.Asignaltobackgroundratio S B wascalculatedforeachpotentialbindingsiteateachsite'sbranch length,wherebranchlengthisdenedasacomputedmeasureofevolutionary distanceandbranchesarewherethesequencesaredisparateenoughaccording toathresholdtowarrantanewbranch.PCTisapprximatelyequalto S B )]TJ/F16 11.9552 Tf 11.955 0 Td [(1 S B Third,context,ameasureofthebindingqualityoftheremainingmiRNA sequencetothetargetsite.Contextisbasedonfourfeatures: 1.site-typecontribution 2.3'pairingcontribution 3.localAUcontribution 4.positioncontribution AmajorassumptionuponwhichweoperateintheanalysisofmiRNAregulatorynetworksisthatthetargetingispredictedfairlywell.Thisislikelynotas 24 PAGE 35 Chapter2.Methods accurateaswewouldlike.ForthisreasonweperformedouranalysisofthenetworksacrossvariousrangesofPCTandcontextwhichhopefullyleadstoamore robustanalysisthansimplysettingahardthresholdforeitheroftheparameters. 25 PAGE 36 Chapter3 ResultsandDiscussion 3.1ExperimentalAnalysis InthestudyofmiRNA:mRNAinteractionsdirected,bipartitegraphsareone methodusedtomodelmiRNAbindingtothe3'-UTRofmRNA,asweourselves did,e.g.,seegure3.5.WesawapronouncedincreaseinthenumberofmiRNA undergoingsignicantregulationthroughouttheinuenzatimecourse.ThiseffectwasespeciallynoticeableinthemiRNAsundergoingdown-regulation,see gure3.1a. Weexaminedthechangesindegreeofthe52miRNAwhichpassedthrough themoststringenttestsofsignicance.Apairwisecomparisonwasdoneforthe datafromeachtimepoint. 26 PAGE 37 Chapter3.ResultsandDiscussion amiRNAdegreehistograms bmRNAdegreehistograms Figure3.1:DegreeanalysisofmiRNAandmRNAnodesthroughoutthetimecourse.TheinformationisdividedintomiRNAandmRNAwhichhaveanticorrelatedexpressionvalues.Topa:out-degreeofmiRNAwithanincreasein expressionvalue.Bottoma:out-degreeofmiRNAwithadecreaseinexpression. Topb:in-degreeofmRNAwithanincreaseinexpression.Bottomb:in-degree ofmRNAwithadecreaseinexpression. Justaswesawinthehistograms3.1athedegreeofthemiRNAnodesgrows intheoverwhelmingmajorityofsignicantlyexpressedmiRNAsthroughoutthe timecourse. 27 PAGE 38 Chapter3.ResultsandDiscussion Figure3.2:Comparisonofdegreechangefromhour6tohour24post-infection. Top :Thechangesaresortedbasedonpercentchangeindegree. Bottom :The sameorderingisappliedtothemiRNAsabsolutechangeindegree. 3.2NetworkAnalysis AsdiscussedinChapter2,themiRNA:miRNAnetworksweconstructedare basedontheTargetscandatasetandthemeasuresusedbytheTargestcanalgorithmtoassessthequalityofaconnection.ToexploretheeectsofthePCT andContextscoresonthegraphsweperformedsomebasicanalysisofeachgraph acrosstherangeofPCTandcontextscores,witharesolutionof0 : 02. 28 PAGE 39 Chapter3.ResultsandDiscussion Figure3.3:Comparisonofdegreechangefromhour6tohour72post-infection. Top :Thechangesaresortedbasedonpercentchangeindegree. Bottom :The sameorderingisappliedtothemiRNAsabsolutechangeindegree. Todothisanalysisrequiredthatwedecomposethespaceofpossibletargets. Recallinsection1.3.1wedenedthespaceofpotentialmiRNAtargetsasM.Using themeasuresprovidedbytheTargetscanalgorithmwemodifyourmapping T s.t. T pct;context : m i M .Where 8 mRNA 2 T pct;context m i thebindingof m i to eachmRNAisassociatedwith pct and context scoresexceedingtheminimum thresholds.Theremainderofthegraphisconstructedinanidenticalmanner, where m i isadjacentto m j ifandonlyif T pct;context m i T T pct;context m j 6 = ; 29 PAGE 40 Chapter3.ResultsandDiscussion Figure3.4:Comparisonofdegreechangefromhour24tohour72post-infection. Top :Thechangesaresortedbasedonpercentchangeindegree. Bottom :The sameorderingisappliedtothemiRNAsabsolutechangeindegree. 3.2.1AverageDegree Giventhegraphstructuredescribedin3.2and1.3.1 Foreachgraph, G = V;E ,wecomputedtheaveragedegreeofanodein G Denedas A G = 2 j E j j V j 30 PAGE 41 Chapter3.ResultsandDiscussion Figure3.5:anti-correlationdirectedbipartitenetworkmapformiRNAupat time6.WeseethemiRNAhererepresentedbycirclesandtheirtargets,mRNA, representedbyrectangles.Thecolorintensitycorrespondstotheexpressionvalue ofthatmiRNAormRNA. Weseeingure3.6thattheaveragedegreedecreasesasthePCTandcontext thresholdsincrease.Thisndingalignswithwhatwewouldexpectofthegraphs asthenumberofviabletargetsdeclines. 31 PAGE 42 Chapter3.ResultsandDiscussion Figure3.6:AveragedegreeofeachgraphatthegivenPCTandContext.The originisinthetopleft,wherethegraphiscompletelyconnected. Wethencomputedthedierencebetweenthetotalnumberofnodesinthe graphandtheaveragedegreeofthegraph,seegure3.7c. 3.2.2GraphDensity Thedensityofagraph, G ,providesausefulmeasureofhowconnected G is.GiventhatourtransformationtomiRNA:miRNAinteractionscanbemost 32 PAGE 43 Chapter3.ResultsandDiscussion aAverageDegree bTotalNodes cTotalminusAverageDegree Figure3.7:WeexaminedtheaveragedegreeofeachmiRNA:miRNAgraphgeneratedforagivenProbabilityofConservedTargetingandContextscoreandthen askedhowmuchtheaveragedegreeofanodeinthegraphdiersfromthetotal numberofnodes,i.e.,howcloselyitistobeingcompletelyconnectedthedistance fromonenodetoanyothernodeis1 easilyinterpretedasanapproximationofpossibledirectandindirectmiRNA cooperation,wehypothesizethatthehighdensityofmanyofournetworksreveals thatmiRNAcooperationisveryimportantandmayevenbenecessarytoproper regulatorycontrol.Alternatively,wemayinterprettherelativelyhighdensity asanindicationofevolvedrobustness,anextremelyimportantaspectofcomplex networks.Thishighconnectivitymaybeimportanttobiologicalsystemsasfailure ofmiRNAregulationmayunderscoresomecancers[18,44]anddisease. Weobservedthatovertheinfectiontimecoursethedensityofthegraphs tendedtoincreaseFigure3.8.Thiscorrespondswiththeobservationthatthe 33 PAGE 44 Chapter3.ResultsandDiscussion aT6Density bT24Density cT72Density Figure3.8:Thedensityofeachgraphiscomputedandvisualizedasaheatmap withrespecttopctandcontext numberofmRNAexpressedaboveourthresholdcutoincreasesoverthetime course.However,weshouldnotethattheincreaseindensity,whilearesultof theincreaseinedgesmRNA,alsoindicatesthatthenumberofnodesmiRNA beingconnectedtothegraphisgrowingmuchslowerthantheratewhichedges arebeingadded. TheincreaseindensityoverthetimecourseFigures3.8a,3.8b,and3.8c correspondswithourobservationsintheanalysisoftheexperimentaldataandthe directedmiRNA:mRNAnetworksthatthenumberofmRNAexpressedaboveour thresholdincreasesoverthetimecourse,thusresultinginanincreasedlikelihood thatedgeswillbeaddedtographs.HigherdensityimpliesmiRNAnodesare becomingbetterconnectedasmRNAedgesareadded.Itismorelikelythata newedgewillconnecttwomiRNAalreadypresentconnectedthanthatanew edgewillconnectapreviouslydisconnectedmiRNAtotherestofthegraph. 34 PAGE 45 Chapter3.ResultsandDiscussion 3.2.3BetweennessCentrality ThebetweennesscentralityanalysisofthemiRNA:miRNAgraphsrevealed manymiRNAnodeswithrelativelylargeshiftsincentrality.BecauseweperformedthisanalysisonnearlytheentirespaceofPCTandcontextscores,more than2400plotswererenderedgeneratedfrom7500miRNA:miRNAgraphs.All code,data,andguresareavailableuponrequestandincludedasastatic"media appendix".Wewillonlypresentafewoftheavailablegureshere. ThePeaktoPeakptpscorewascomputedforeachmiRNAbetweenness centralityvectore.g.,intable2.1,theptpscorecorrespondstothebetweennesscetralityscoresattainedbyasinglemiRNAacrossthetimecourseandthe miRNAweresortedonthisscore.Wethentookoutthetop20%ofmiRNAs basedontheirptpvaluesandcomparedthemtothemiRNAfromthemicroarray experimentwhichpassedthemoststringenttests,describedinsection2.2,aswell asafoldchangethresholdof1 : 8.Theoverlapneverexceeded8%.Seegure3.10. TogaugetheoverlapofallmiRNApresentinthemiRNA:miRNAnetworkwith themiRNApresentinthedatasetfromthemicroRNAarrayweperformedthe samecomputationwithoutlteringthemiRNAfromthenetwork,asthemiRNA expressiondatahasbeendeterminedtobeofquestionablequalityinotherinvestigationsnotdiscussedherepersonalcommunicationsHart.Thisisseenin 35 PAGE 46 Chapter3.ResultsandDiscussion Figure3.9:MicroRNABetweennessCentralityPeaktoPeakScore.HerePCT= Context=0.6.MicroRNAarealongthex-axisandthecorrespondingptpscores onthey-axis. gure3.11.Theoverlapofthesetwosetspeakedatapproximately13%ofthe totalmiRNAs. Itispossibletocalculatebetweennesscentralityonmultigraphs,however,the toolsweusedintheanalysis,partofthenetworkxpackage,arenotdesigned todothis.Inaddition,thecomputationsonthebinarygraphswerealreadya 36 PAGE 47 Chapter3.ResultsandDiscussion Figure3.10:Percentoverlapofthetop20%ofmiRNAnodesbybetweennessshift withthesetofmiRNAfromthemicroarraywithap-valueoflessthan0.05 signicantcomputationalendeavorand,asnoted,thecomputationtimeincreases withthenumberofedges. Wemeasuredoverlapbymakingthefollowingcomputation Over X;Y = j X T Y j j X S Y j foreachsetofmiRNAwithXspecictothegraphresultingfromthegiven PCTandcontext,andYthemiRNAwithap-value 0 : 05fromthemicroarray. 37 PAGE 48 Chapter3.ResultsandDiscussion Figure3.11:PercentoverlapofallmiRNAnodesfromthemiRNA:miRNAnetworkwiththesetofmiRNAfromthemicroarraywithap-valueoflessthan0.05 asPCTandContextvary 3.2.4BetweennessVersusDegreeChange AscatterplotofbetweennesscentralitychangeversusdegreechangewasrenderedforeveryPCTandcontextcombination.Again,wepresentonlyasampling oftheavailablegureshereFigure3.12. Asweseeinthecomparisonoftherandomizednetworksgure3.13our actualdataisfardierentfromwhatwegetwhenweshuetheedges. 38 PAGE 49 Chapter3.ResultsandDiscussion Figure3.12:ThePeaktoPeakscoreformiRNAbetweennesscentralityversus thedegreechangeforthatmiRNAwasscattered.Here,weseethescatterplot forPCT=Context=0.6. 3.2.5ExplorationofthePCTandContextGraphSpace Examiningsecond,third,andfourthmomentsaboutthemeanforthedegree distributionofeachinstanceofagraphaparticularPCTandContextreveals asignicantlyhighervariancewhenbothPCTandcontextarerelativelylow. VariancedropsosignicantlywhenPCTisgreaterthan0.1orContextisgreater than0.3. 39 PAGE 50 Chapter3.ResultsandDiscussion Figure3.13:Comparisonofthedatafromgure3.12withrandomizednetworks. Hereaplus+denotesrandomizeddataandaredcircleisactualdata. aVariance bSkew cKurtosis Figure3.14:BinaryEdgeGraphDegreeStatistics 40 PAGE 51 Chapter3.ResultsandDiscussion aVariance bSkew cKurtosis Figure3.15:MultigraphDegreeStatistics 3.2.6ConclusionandFutureResearch AnalysisofmiRNAregulatorysignicancefromtheperspectiveofbetweenness centralityallowstheselectionofmiRNAsbasedonlyonmRNAexpressiondata. However,aswehaveseen,thismethodofanalysisdidnotalignwellwiththe expressiondatagatheredusingthemicroarrayassay.Thediscrepanciesbetween thebetweennesscentralityanalysisandthemicroarraydatadoesnotnecessarilymeanthatthecentralityanalysisyieldslesssignicantorinterestingresults. MicroRNAnodesundergoinglargeshiftsincentralityoveratimecoursemaybe signicantintermsofeithercooperativeregulationofmRNAorasinhibitorsof othermiRNAbinding.ItisoftenthecasethatmiRNAbindingsitesoverlapon 3'UTRs.This,alongwiththefactthatasinglemiRNAis,insomecases,not enoughtoinhibittranslationordirectthemRNAfordegradationcouldmean 41 PAGE 52 Chapter3.ResultsandDiscussion thatmiRNAsarecapableofpreservingmRNAsignalfromothermiRNAswhich wouldotherwisebindanddegradethemessage. Thereareseveralmethodsforidentifyingtheskeletonofanetwork,aspecial caseofaspanningtree.Tocreatetheskeleton,edgesareremovedbasedon theiredgebetweennessscore,ameasurewhichisidenticaltothebetweenness centralityscorebutcomputedforedgesratherthannodes.Thismethodwould beespeciallyusefulinidentifyingthehubsofthesemiRNAnetworkswhenusing expressiondatatoweightedges.Theskeletonisusuallycomputedasameansof removing"shortcuts"fromhighlyconnectedclusters.Thisissometimesusedto identifypotentiallyscalefreenetworksandimprovegraphvisualization. Deninganedgeweightfunctionforthesenetworksisanotherproblemunto itself.WedidnotwishtopartitionthenetworksbysplittingthemRNAintoup ordownsetsateachtimepoint,thismeanswecanhavenegativeedgeweights resultinginthepossibilityofacyclewithnegativetotalweight.Normalizing theexpressiondatawasnotagoodsolutioneitherastheedgeweightsarethen highlydependentonthemRNAwiththelargestfoldchange,yieldingdatathat isdiculttocompare. 42 PAGE 53 Bibliography [1] Ambros,V. Auniformsystemformicrornaannotation. RNA9 277279. [2] Aravin,A.A.,etal. Thesmallrnaproleduring Drosophilamelanogaster development. Dev.Cell5 ,337350. [3] Ascher,D.,Dubois,P.F.,Hinsen,K.,Hugunin,J.,andOliphant, T. NumericalPython ,ucrl-ma-128569ed.LawrenceLivermoreNational Laboratory,Livermore,CA,1999. [4] Barabasi,A.-L.,andAlbert,R. Emergenceofscalinginrandomnetworks. Science286 [5] Bartel,D.P. Micrornas:genomics,biogenesis,mechanism,andfunction. Cell116 ,281297. [6] Bohnsack,M.T.,Czaplinski,K.,andGorlich,D. Exportin5is arangtp-dependentdsrna-bindingproteinthatmediatesnuclearexportof pre-mirnas. RNA10 ,185191. [7] Brennecke,J.,Hipfner,D.R.,Stark,A.,Russel,R.B.,andCohen,S. M.bantam encodesadevelopmentallyregulatedmicrornathatcontrolscellproliferationandregulatestheproapoptoticgene hid in Drosophila Cell113 ,2536. [8] Cai,X.,Hagedorn,C.H.,andCullen,B.R. Humanmicrornasare processedfromcapped,polyadenelatedtranscriptsthatcanalsofunctionas mrnas. RNA10 ,19571966. [9] Cullen,B.R. Transcriptionandprocessingofhumanmicrornaprecursors. Mol.Cell16 [10] Dijkstra,E. Anoteontwoproblemsinconnexionwithgraphs. Numerische Mathematik1 ,269271. 43 PAGE 54 Bibliography [11] Flippov,V.,Solovyev,V.,Flippova,M.,andGill,S.S. Anovel typeofrnaseiiifamilyproteinsineukaryotes. Gene245 ,213221. [12] Fredman,M.L.,andTarjan,R.E. Fibonacciheapsandtheirusesin improvednetworkoptimizationalgorithms. FoundationsofComputerScience,1984.25thAnnualSymposiumon ,338346. [13] Friedman,R.C.,Burge,C.B.,andBartel,D.P. Mostmammalian mrnasareconservedtargetsofmicrornas. GenomeResearch19 92105. [14] Giaever,G.,etal. Functionalprolingofthesaccharomycescerevisiae genome. Nature418 [15] Grimson,A.,Farh,K.K.,Johnston,W.K.,Garrett-Engele,P., Lim,L.P.,andBartel,D.P. Micrornatargetingspecicityinmammals: Determinantsbeyondseedpairing. MolecularCell27 ,91105. [16] Hagberg,A.A.,Schult,D.A.,andSwart,P.J. Exploringnetwork structure,dynamics,andfunctionusingNetworkX.In Proceedingsofthe7th PythoninScienceConferenceSciPy2008 Pasadena,CAUSA,Aug.2008, p.1115. [17] Han,J.-D.J.,Bertin,N.,Hao,T.,Goldberg,D.S.,Berriz,G.F., Zhang,L.V.,Dupuy,D.,Walhout,A.J.M.,Cusick,M.E.,Roth, F.P.,andVidal,M. Evidencefordynamicallyorganizedmodularityin theyeastproteinproteininteractionnetwork. Nature430 ,8893. [18] Hon,L.S.,andZhang,Z. Therolesofbindingsitearrangementand combinatorialtargetinginmicrornarepressionofgeneexpression. Genome Biology [19] Jeong,H.,Mason,S.P.,Barabasi,A.-L.,andOltvai,Z.N. Lethalityandcentralityinproteinnetworks. Nature411 [20] Johnson,S.M.,Lin,S.Y.,andSlack,F.J. Thetimeofappearance ofthe C.eleganslet-7 micrornaistranscriptionallycontrolledutilizinga temporalregulatoryelementinitspromoter. Dev.Biol.259 ,364379. [21] Khvorova,A.,Reynolds,A.,andJayasena,S.D. Functionalsirnas andmirnasexhibitstrandbias. Cell115 ,209216. [22] Kim,V.N. Micrornabiogenesis:Coordinatedcroppinganddicing. Nature Reviews6 May2005,376385. 44 PAGE 55 Bibliography [23] Klovdahl,A.S.,Graviss,E.A.,Yaganehdoost,A.,Ross,M.W., Wanger,A.,Adams,G.J.,andMusser,J.M. Networksandtuberculosis:anundetectedcommunityoutbreakinvolvingpublicplaces. Soc.Sci. Med.52 ,681694. [24] Lagos-Quintana,M.,etal. Identicationoftissuespecicmicrornas frommouse. Curr.Biol.12 ,735739. [25] Lagos-Quintana,M.,Rauhut,R.,Lendeckel,W.,andTuschi,T. Identicationofgenesencodingforsmallexpressedrnas. Science294 [26] Lau,N.C.,Lim,L.P.,Weinstein,E.G.,andBartel,D.P. An abundantclassoftinyrnaswithprobableregulatoryrolesincaenorhabditis elegans. Science294 [27] Lee,R.C.,Feinbaum,R.L.,andAmbros,V. The C.elegans heterochronicgene lin-4 encodessmallrnaswithantisensecomplimentarityto lin-14 Cell75 ,843854. [28] Lee,Y.,etal. Thenuclearrnaseiiidroshainitiatesmicrornaprocessing. Nature425 ,415419. [29] Lee,Y.,etal. Micrornagenesaretranscribedbyrnapolymeraseii. EMBO23 ,40514060. [30] Lee,Y.,Jeon,K.,Lee,J.T.,Kim,S.,andKim,V. Micrornamaturation:stepwiseprocessingandsubcellularlocalization. EMBOJ.21 [31] Lewis,B.P.,Shih,I.H.,Jones-Rhoades,M.W.,Bartel,D.P., andBurge,C.B. Predictionofmammalianmicrornatargets. Cell115 ,787798. [32] Lewis,B.P.,Shih,I.H.,Jones-Rhoades,M.W.,Bartel,D.P., andBurge,C.B. Conservedseedpairing,oftenankedbyadenosines, indicatesthatthousandsofhumangenesaremicrornatargets. Cell120 ,1520. [33] Li,Y.,Chan,E.Y.,Li,J.,Peng,C.N.X.,Rosenzweig,E., Tumpey,T.M.,andKatze,M.G. Micrornaexpressionandvirulence inpandemicinuenzavirus-infectedmice. JournalofVirology84 ,6March 2010,30233032. [34] Lund,E.,Guttinger,S.,Calado,A.,Dahlberg,J.E.,andKutay, U. Nuclearexportofmirnaprecursors. Science303 ,9598. 45 PAGE 56 Bibliography [35] Mourelatos,Z.,etal. mirnps:anovelclassofribonuceoproteinscontainingnumerousmicrornas. GenesDev.16 [36] Nielsen,C.B.,Shomron,N.,Sandberg,R.,Hornstein,E.,Kitzman,J.,andBurge,C.B. Determinantsoftargetingbyendogenousand exogenousmicrornasandsirnas. RNA13 ,18941910. [37] Oliphant,T.E. GuidetoNumPy .Provo,UT,2006. [38] os,P.E.,andRenyi,A. Onrandomgraphs. Publ.Math.6 [39] Rusu,P.,Sheridan,M.,Sewer,R.,Iovino,A.,Aravin,N.,Pfeffer,A.,Rice,S.,Kamphorst,A.,Lanthaler,A.O.,andLandgraf,M. Amammalianmicrornaexpressionatlasbasedonsmallrnalibrary sequencing. Cell129 ,14011414. [40] Schwarz,D.,etal. Asymmetryintheassemblyofthernaienzyme complex. Cell115 ,199208. [41] Winzeler,E.A.,etal. Functionalcharacterizationofthes.cerevisiae genomebygenedeletionandparallelanalysis. Science285 [42] Xie,X.,Lu,J.,Kulbokas,E.,Golub,T.,Mootha,V.,LinbladToh,K.,Lander,E.S.,andKellis,M. Systematicdiscoveryofregulatorymotifsinhumanpromotersand3'utrsbycomparisonofseveral mammals. Nature434 ,338345. [43] Yi,R.,Qin,Y.,Macara,I.G.,andCullen,B.R. Exportin-5mediates thenuclearexportofpre-micrornasandshorthairpinrnas. GenesDev17 ,30113016. [44] Yu,Z.,Li,Z.,Jolicouer,N.,Zhang,L.,Fortin,Y.,Wang,E., Wu,M.,andShen,S.H. Aberrantallelefrequenciesofthesnpslocatedin micrornatargetsitesarepotentiallyassociatedwithhumancancers. Nucleic AcidsResearch35 ,45354541. [45] Zeng,Y.,Yi,R.,andCullen,B.R. Recognitionandcleavageorprimary micrornaprecursorsbythenuclearprocessingenzymedrosha. EMBOJ.24 ,138148. 46 | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||