Optimization based automated curation of metabolic reconstructions.docx

Optimization based automated curation of metabolic reconstructions.docx

《Optimization based automated curation of metabolic reconstructions.docx》由会员分享，可在线阅读，更多相关《Optimization based automated curation of metabolic reconstructions.docx（24页珍藏版）》请在冰点文库上搜索。

Optimizationbasedautomatedcurationofmetabolicreconstructions

Optimizationbasedautomatedcurationofmetabolicreconstructions

VinaySatishKumar,1MadhukarSDasika,2andCostasDMaranas

2

1DepartmentofIndustrialandManufacturingEngineering,ThePennsylvaniaStateUniversity,UniversityPark,PA16802,USA

2DepartmentofChemicalEngineering,ThePennsylvaniaStateUniversity,UniversityPark,PA16802,USA

Correspondingauthor.

VinaySatishKumar:

vsk111@psu.edu;MadhukarSDasika:

msd179@psu.edu;CostasDMaranas:

costas@psu.edu

ReceivedDecember14,2006;AcceptedJune20,2007.

ThisisanOpenAccessarticledistributedunderthetermsoftheCreativeCommonsAttributionLicense（http:

//creativecommons.org/licenses/by/2.0）,whichpermitsunrestricteduse,distribution,andreproductioninanymedium,providedtheoriginalworkisproperlycited.

Top 

Abstract 

Background 

Results 

Discussionandconclusion 

Methods 

Authors'contributions 

SupplementaryMaterial 

References 

Abstract

Background

Currently,thereexiststensofdifferentmicrobialandeukaryoticmetabolicreconstructions（e.g.,Escherichiacoli,Saccharomycescerevisiae,Bacillussubtilis）withmanymoreunderdevelopment.Allofthesereconstructionsareinherentlyincompletewithsomefunctionalitiesmissingduetothelackofexperimentaland/orhomologyinformation.Akeychallengeintheautomatedgenerationofgenome-scalereconstructionsistheelucidationofthesegapsandthesubsequentgenerationofhypothesestobridgethem.

Results

Inthiswork,anoptimizationbasedprocedureisproposedtoidentifyandeliminatenetworkgapsinthesereconstructions.Firstweidentifythemetabolitesinthemetabolicnetworkreconstructionwhichcannotbeproducedunderanyuptakeconditionsandsubsequentlyweidentifythereactionsfromacustomizedmulti-organismdatabasethatrestorestheconnectivityofthesemetabolitestotheparentnetworkusingfourmechanisms.Thisconnectivityrestorationishypothesizedtotakeplacethroughfourmechanisms:

a）reversingthedirectionalityofoneormorereactionsintheexistingmodel,b）addingreactionfromanotherorganismtoprovidefunctionalityabsentintheexistingmodel,c）addingexternaltransportmechanismstoallowforimportationofmetabolitesintheexistingmodelandd）restoreflowbyaddingintracellulartransportreactionsinmulti-compartmentmodels.Wedemonstratethisprocedureforthegenome-scalereconstructionofEscherichiacoliandalsoSaccharomycescerevisiaewhereincompartmentalizationofintra-cellularreactionsresultsinamorecomplextopologyofthemetabolicnetwork.Wedeterminethatabout10%ofmetabolitesinE.coliand30%ofmetabolitesinS.cerevisiaecannotcarryanyflux.Interestingly,thedominantflowrestorationmechanismisdirectionalityreversalsofexistingreactionsintherespectivemodels.

Conclusion

Wehaveproposedsystematicmethodstoidentifyandfillgapsingenome-scalemetabolicreconstructions.Theidentifiedgapscanbefilledbothbymakingmodificationsintheexistingmodelandbyaddingmissingreactionsbyreconcilingmulti-organismdatabasesofreactionswithexistinggenome-scalemodels.Computationalresultsprovidealistofhypothesestobequeriedfurtherandtestedexperimentally.

Top 

Abstract 

Background 

Results 

Discussionandconclusion 

Methods 

Authors'contributions 

SupplementaryMaterial 

References 

Background

Thegenomeofseveralmicroorganismshasbeencompletelysequencedandannotatedinthepastdecade[1-4].Thisinformationhasaidedthemetabolicreconstructionsofseveralmicrobialandeukaryoticorganismsusingexperimentalevidenceandbioinformaticsbasedtechniquesprovidingsinglecompartment（e.g.,Escherichiacoli[5]）andmulti-compartmentmodels（e.g.,Saccharomycescerevisiae[6]）.Allofthesereconstructionsareinherentlyincompletewithsomefunctionalitiesmissingduetothelackofexperimentaland/orhomologyinformation.Thesemissingreactionstepsmayleadtothepredictionoferroneousgeneticinterventionsforatargetedoverproductionortheelucidationofmisleadingorganizationalprinciplesandpropertiesofthemetabolicnetwork.Akeychallengeintheautomatedgenerationofgenome-scalereconstructionsistheelucidationofthesegapsandthesubsequentgenerationofhypothesestobridgethem.Thischallengehasalreadybeenrecognizedandanumberofcomputationalapproacheshavebeenunderdevelopmenttoresolvethesediscrepancies[7-11].

Mostoftheaforementionedeffortsarebasedontheuseofsequencehomologytouncovermissinggenes.Specifically,sequencehomologyisusedtopinpointgenesinrelatedspeciesthathavesignificantsimilaritywithanunassignedORFofthecuratedmicroorganism[12].GreenetalformalizedandfurtherextendedthisconceptbyintroducingamethodthatidentifiedmissingenzymesinametabolicnetworkusingsequencehomologyrelatedmetricswithinaBayesianframework[11].Alternatively,non-homologybasedreconstructionshavebeenimplementedbyidentifyingcandidategenesbymeasuringthesimilaritywithmetricssuchasmRNAcoexpressiondata[8]andphylogeneticprofiles[10]whilealsotakingintoaccountthelocalstructureoftheexistingpartiallyreconstructedmetabolicnetworks.Arecentadvancementinthisdirectionusesmultipletypesofassociationevidenceincludingclusteringofgenesonthechromosomeandproteinfusioneventsinadditiontophylogeneticprofiles[9].Allmethodsdescribedabovepostulateasetofcandidategenesandthenevaluatethelikelihoodthatanyofthesecandidategenesispresentinthemicroorganism'smetabolicnetworkofinterestusingavarietyofscoringmetrics.Inadditiontotheseapproaches,variousgenomiccontextanalyseshavealsobeenusedtoidentifymissingmetabolicgenes[7,13-16].Specifically,arecentstudyexploitstheavailabilityofhighlycuratedmetabolicnetworkstohypothesizegenereactioninteractionsinlesscharacterizedorganisms[16].Theseaforementionedmethodspredictmissingenzymesinthemetabolicnetworkbyconductingsequencebasedcomparisonsofentiregenomesandinferringpossiblemetabolicfunctionsacrossdifferentmicroorganisms.

Alternatively,arecentsystemsbasedcomputationalapproachidentifiesthelocationofmissingmetabolicfunctionsintheE.coliiJR904modelbypinpointingdiscrepanciesbetweeninsilicomodelpredictionsandknowninvivogrowthphenotypes[17].Subsequently,anoptimizationbasedalgorithmisusedtoresolvethesediscrepanciesbysearchingformissingmetabolicfunctionsfromacandidatedatabaseofreactions.Inthispaperinstead,wepinpointmetabolitesthatcannotcarryanyfluxthroughthemandsubsequentlygeneratehypothesestorestoreconnectivity.Tothisend,weintroduceanoptimizationbasedprocedure（GapFind）tofirstidentifysuchgapsinbothsingleandmulti-compartmentmetabolicnetworksandsubsequentlyusinganoptimizationbasedprocedure（GapFill）restoretheirconnectivityusingseparatepathologyresolvinghypotheses.Incontrasttothepreviousmethodswhichfillgapsonlybyidentifyingmissingenzymes[8-11,17]oraddingtransportreactions[17],wealsoexplorewhetherthesegapscanbefilledbymakingintramodelmodifications.Figure1pictoriallyillustrateshowsuchgapsariseinmetabolicreconstructionsandintroducesthedefinitionsproposedinthispapertopreciselydescribethesepathologies.

Figure1

Characterizationofproblemmetabolitesinmetabolicnetworks.MetaboliteAisdefinedasarootno-productionmetabolitebecausethereisno-productionortransportmechanismforitinthenetwork.MetaboliteCisadownstreamno-productionmetabolite（more...）

Gapsinmetabolicreconstructionsaremanifestedas（i）metaboliteswhichcannotbeproducedbyanyofthereactionsorimportedthroughanyoftheavailableuptakepathwaysinthemodel;or（ii）metabolitesthatarenotconsumedbyanyofthereactionsinthenetworkorexportedbasedonanyexistingsecretionpathways.Werefertothesemetabolitesasrootno-production（e.g.,metaboliteA）androotno-consumptionmetabolites（e.g.,metaboliteB）respectively.Atsteady-stateconditionsnoflowcanpassthroughthemduetotheincompleteconnectivitywiththerestofthenetwork.Clearly,suchpathologiesarenotphysiologicallyrelevantandthusmustbecausedbyomissionand/orerrorsinthemodelreconstructionprocess.Notably,thelackofflowinrootno-productionmetabolitesandrootno-consumptionmetabolitesispropagateddownstream/upstreamrespectivelygivingrisetoadditionalmetabolitesthatcannotcarryanyflow.Werefertothesemetabolitesthatareindirectlypreventedfromcarryingflowasdownstreamno-production（e.g.,metaboliteC）metabolitesandupstreamno-consumptionmetabolites（e.g.,metaboliteD）respectively.Itisimportanttonotethatbyrestoringconnectivityfortherootproblemmetabolitesallupstream/downstreammetabolitesarealsoautomaticallyfixed.Weconcentrateonresolvingonlyno-productionmetabolitesinthecaseofcytosolicmetabolites.Inthecaseofnon-cytosolic（i.e.,presentininternalcompartments）metabolites,weidentifymechanismstoresolvebothno-productionandno-consumptionmetabolites.

Forsinglecompartmentmetabolicnetworks（wherewehaveonlycytosolicmetabolites）,wepostulatethreeseparatemechanismsforfixingno-productionmetabolites（seealsoFigure2）.Weexplorewhether（i）reversingthedirectionalityofexistingreactionsinthemodel（Mechanism1）,（ii）addingnewreactionsfromamulti-speciesdatabase（e.g.,MetaCyc[18]）（Mechanism2）orfinally（iii）allowingforthedirectimportationoftheproblemmetaboliterestoresflowintotheno-productionmetabo