admixture-manual

admixture软件使用手册

Admixture1.23SoftwareManualDavidH.AlexanderJohnNovembre

August22,2013KennethLange

admixture软件使用手册

Contents1Quickstart

2Reference

2.1HowdoIchoosethecorrectvalueforK?..........

2.1.1Background.......................

2.2HowdoIplottheQestimates?................

2.3DoIneedtothinthemarkersetforlinkagedisequilibrium?

2.4HowmanymarkersdoIneedtosupplytoADMIXTURE?.

2.5HowdoIchangetherandomseed?’.............

2.6OptimizationMethods.....................

2.7Terminationcriteria......................

2.8Accelerationmethods.....................

2.9Bootstrapping..........................

2.10Supervisedanalysis.......................

2.11Multithreadedmode......................

2.12Penalizedestimation......................

3Citationandfurtherinformation.....................................................................................................................1334556777888101010Notetousers:thematrixformerlyreferredtoasF,containingthepopulationallelefrequencies,isnowknown(asofv1.20)asP,andisoutputtoa.P?le.ThisistobringournotationinlinewiththatoftheSTRUCUTREpapers.

1Quickstart

ADMIXTUREisaprogramforestimatingancestryinamodel-basedmannerfromlargeautosomalSNPgenotypedatasets,wheretheindividualsareunrelated(forexample,theindividualsinacase-controlassociationstudy).

ADMIXTURE’sinputisbinaryPLINK(.bed),ordinaryPLINK(.ped),orEIGENSTRAT(.geno)formatted?lesanditsoutputissimplespace-delimited?lescontainingtheparam-eterestimates.

TouseADMIXTURE,youneedaninput?leandanideaofK,yourbeliefofthenumberofancestralpopulations.Youshouldalsohavetheassociatedsupport?lesalongsideyourmaininput?le,inthesamedirectory.Forexample,ifyourprimaryinput?leisa.bed?le,youshouldhavetheassociated.bim(binarymarkerinformation?le)and.fam(pedigreestub?le)?lesinthesamedirectory.Ifyourprimaryinput?leisa.pedor.geno?le,acorrespondingPLINKstyle.map?leshouldbeinthesamedirectory.

1

admixture软件使用手册

IfyouhaveanbinaryPED(.bed)formatted?leinyourcurrentdirectory

%ls

hapmap3.bedhapmap3.bimhapmap3.fam

andyoubelievethattheindividualsinthesamplederivetheirancestryfromthreeancestralpopulationsthenrunadmixturelikethis:

%admixturehapmap3.bed3

ADMIXTUREwillstartrunning.Hopefullyitwill?nishsoon,anditwillthenoutputsomeestimates:

%ls

hapmap3.bedhapmap3.bimhapmap3.famhapmap3.3.Qhapmap3.3.P

Thereisanoutput?leforeachparameterset:Q(theancestryfractions),andP(theallelefrequenciesoftheinferredancestralpopulations).Notethattheoutput?lenameshave‘3’inthem.Thisindicatesthenumberofpopulations(K)thatwasassumedfortheanalysis.This?lenameconventionmakesiteasytorunanalysesusingdi?erentvaluesofKinthesamedirectory.

IfyouhaveaPLINK.ped“12”coded?le—generatedbyacommandlike

%plink--filehapmap--recode12--outhapmap

—thenyoufollowbasicallythesameinstructions:

%admixturehapmap3.ped3

NotethatADMIXTUREinferstheformatofyourinput?lebasedonthe?leextension(.bed,.ped,or.geno).

YoucanrunADMIXTUREonainput?lelocatedinanotherdirectory,butitsoutputwillalwaysbeputinthecurrentworkingdirectory.Forexample:

%pwd

/home/dalexander/AdmixtureRuns

%ls

(nooutput—currentdirectoryisempty)

%ls~/Data

hapmap3.pedhapmap3.map

%admixture~/Data/hapmap3.ped3

(waitforittorun)

2

admixture软件使用手册

%ls

hapmap3.3.Qhapmap3.3.P

%ls~/Data

hapmap3.pedhapmap3.map

Ifyoualsowantedstandarderrors,theninsteadofthelastcommandyoushouldhaveused

%admixture-B~/Data/hapmap3.ped3

Thiswillperformpointestimationandthenwillalsouseabootstrappingproceduretocalculatethestandarderrors.Notethat(point-estimation&bootstrapping)takesconsid-erablylongerthanpoint-estimationalone,soyouwillhavetobepatient.Eventuallyitwill?nish,yieldingpointestimatesandstandarderrors:

%ls

hapmap3.3.Qhapmap3.3.Phapmap3.3.Q_se

The“se”?leisinthesameunadorned?leformatasthepointestimates.

Ifyouranalysesaretakingalongtime,?rstconsiderwhy.Isyourdatasethuge?Doyoureallyneedtoanalyzeallthemarkers,orcanyouthinthemarkerset(asdescribedinSection2.3)?Ifyoureallydoneedtoanalyzeallthemarkers,orifyouranalysisisslowishforotherreasons(verylargeK,extensivecross-validationorbootstrapping)youmightconsiderrunningADMIXTUREinmultithreadedmode.Ifyourcomputerhasfourprocessors,forexample,youcoulduseacommandlike

%admixture~Data/huge_dataset.bed3-j4

tosplitADMIXTURE’sworkamongfourthreads—whichcanmakeADMIXTURErunalmostfourtimesasfast.

2

2.1ReferenceHowdoIchoosethecorrectvalueforK?

UseADMIXTURE’scross-validationprocedure.AgoodvalueofKwillexhibitalowcross-validationerrorcomparedtootherKvalues.Cross-validationisenabledbysimplyaddingthe--cv?agtotheADMIXTUREcommandline.Inthisdefaultsetting,thecross-validationprocedurewillperform5-foldCV—youcanget10-foldCV,forexample,using--cv=10.Thecross-validationerrorisreportedintheoutput.Forexample,ifinourbashshellweran

3

admixture软件使用手册

%forKin12345;\

doadmixture--cvhapmap3.bed$K|teelog${K}.out;done

(i.e.,ranADMIXTUREwithcross-validationforKvalues1,2,3,4and5),thenwecouldquicklyviewtheCVerrors:

%grep-hCVlog*.out

CVerror(K=1):0.55248

CVerror(K=2):0.48190

CVerror(K=3):0.47835

CVerror(K=4):0.48236

CVerror(K=5):0.48985

wherethenumberinparenthesesisthestandarderrorofthecross-validationerrorestimate.Wecaneasilyplotthesevaluesforcomparison,asinFigure

1,whichmakesitfairlyclearthatK=3isasensiblemodelingchoice.

Figure1:Cross-validationplotforthehapmap3dataset

2.1.1Background

ThosefamiliarwithSTRUCTUREknowthatitprovidesameansofidentifyingthe“best”valueforK,thenumberofpopulations,basedoncomputingthemodelevidenceforeachpossibleKvalue.ThemodelevidenceforKisde?nedas

Pr(G|K)=f(G|Q,P,K)π(Q,P|K)dQdP

4