MatlabPARAFAC运行办法.pdf

MatlabPARAFAC运行办法.pdf

《MatlabPARAFAC运行办法.pdf》由会员分享，可在线阅读，更多相关《MatlabPARAFAC运行办法.pdf（6页珍藏版）》请在冰点文库上搜索。

Appendix1:

TutorialCharacterizingDOMfluorescenceusingtheDOMFluorToolboxforMATLABInthissectionabrieftutorialinapplyingPARAFACtoDOMfluorescenceEEMswillbegivenwithrecommendationsonasuitableseriesofstagestofollowinordertofacilitatetheprocessbasedontheauthorsownexperience.ThetutorialisbasedonMATLABsoftwareandrequiresthatthisisinstalledonyourcomputer.TheDOMFluortoolboxforMATLABpro-videdisinthecompressedfile“DOMFluorv1_4.zip”（Appen-dix1）andworksindependentofyourcomputersoperatingsystem.ThisisalsofreelydownloadablefromtheChemomet-ricssiteatUniversityofCopenhagen（www.models.life.ku.dk）.ThetoolboxcontainswithinittheN-Waytoolboxv.3.1（AndersonandBro,2000）andadditionalfunctionswhichfacilitaterunningPARAFAConDOMfluorescencedata.TheDOMFluortoolboxiswrittenandtestedinMATLABrelease2007b（v7.5）andisfreesoftwarethatcanbeusedandmodi-fiedunderthetermsoftheGNUGeneralPublicLicense.Withinthetoolboxatestdatasetisprovidedforthetutorial,consistingof65EEMswhichhavebeenspectrallycorrectedforinstrumentbiases,correctedforinnerfiltereffectsandRamancalibrated.Asaresult,alloftheRayleighscatterandmostoftheRamanscatterhasbeenremoved.Theremainingscatterwillberemovedinthepre-processingphase.ThedataconsistsofsamplescollectedfromacruiseonRVGunnarThorsonintheKattegatandBeltSearegion（attheentrancetotheBalticSea）inAugust2001.Three“outlier”sampleswerecreatedandarepresentinthedataforthebenefitofthetutorial.Thetutorialiscomposedofphases;i）datapre-processing,ii）initialexplorativedataanalysis,iii）modelingandmodelvalidation,iv）interpretingresults.Themorethoroughlythefirsttwophasesarecarriedout,theeasierthelattertwowillbe.Partofthepre-processingontheexampledatasethasalreadybeencarriedout（importingthedatatoMATLAB,spec-tralcorrections,calibrations,etc.）.Thisisveryinstrumentspe-cificandthereforenotappropriateforthistutorial,whichfocusonthemodelingofthedata.DuringthistutorialanystatementstobewritteninMAT-LABwillbewritteningreenwithCourierNewfont.WhenwritingintheCommandWindow,notethatMATLABiscasesensitive.A.SettingupMATLABtorunwithN-waytoolboxandDOMFluortoolbox.1.DecideonasensibleplacetolocatetheDOMFluortoolboxonyourcomputer.（Thiscould,forexample,beC:

ProgramFilesMATLABR2007btoolboxbutcanalsobeelsewhere）.CreateafolderandnameitDOMFluor.Extractthecontentsofthezipfile“DOM-Fluor.zip”intotheDOMFluorfolder.2.StartMATLAB.SelectFilemenuthenclickSetpath.ClickAddwithsubfoldersandlocatetheDOMFluorfolderyoucreatedinStep1above.IftheN-Waytool-boxiscurrentlyinstalled,removeitfromthePathlisttoavoidconflicts.TheN-WaytoolboxisincludedaspartoftheDOMFluortoolbox.ClickSavethenClose.YouhavenowassociatedtheDOMFluortool-boxwithMATLABandthismeansthatthefunctionsinthesefolderscanberecognized.3.TocheckthatthetoolboxesarecorrectlyassociatedyoucantypeDOMFluorandpressenterintheCommandWindowinMATLAB.If“Yes”isprintedtotheCommandWindow,thetool-boxisinstalled.Ifredtextappearsthetoolboxhasnotbeencorrectlyinstalled.RedoStep2.Notethatifyouhaveothertoolboxeswithsimilarlynamedfunctions,onlythefunctionupmostinthefolder-listwillworkproperly.4.SectionAwillhavetoberepeatedifanewversionofMATLABisinstalled.B.LoadingthetutorialdataandplottingtheEEMs.1.TypeloadPARAFACexample.matintheCommandWindow.ThiswillloadthetutorialdataintotheMATLABWorkspace.ThisconsistsofadatastructurecalledOriginalData.TrytypingOriginalDataCharacterizingdissolvedorganicmatterfluorescencewithparal-lelfactoranalysis:

atutorialColinA.Stedmon1*andRasmusBro21DepartmentofMarineEcology,NationalEnvironmentalResearchInstitute,AarhusUniversity,Frederiksborgvej399,Roskilde,Denmark2Dept.FoodScience,FacultyofLifeSciences,UniversityCopenhagen,Rolighedsvej30,DK-1958,Frederiksberg,Denmark*Correspondingauthore-mail:

cstdmu.dk,Tel:

+4546301805.Limnol.Oceanogr.:

Methods6,20082008,bytheAmericanSocietyofLimnologyandOceanography,Inc.inthecommandwindow.Thiswilloutputthefollowing:

Thisdetailsthedatathatiscontainedwithinthedatastructure.i.Ex-isalistoftheexcitationwavelengthsmea-sured（innm）.Inthiscase240to450every5nm.ii.Em-isalistoftheemissionwavelengthsmea-sured（innm）.Inthiscase300to600every2nm.iii.X-isthefluorescencedataasathreedimen-sionalarray（65samplesx151emissionwave-lengthsx43excitationwavelengths）.iv.nSample,nExandnEm:

statethenumberofsamples,excitationandemissionwavelengths.v.XBackup-isabackupcopyofthedataandnotusedinthetutorial.2.NowwewillplottheEEMstocheckthatthedataiscorrectlyloaded.Inthetoolboxthereareseveralfunc-tionsthatcanbeusedtoplotEEMs.a.TypingPlotEEMby1（1:

5,OriginalData,R.U.）willplotthefirst5EEMsinthedatasetoneatatimeascontourplots.Thenextgraphcanbeviewedbypressinganykeyonthekeyboard.If1:

5isreplacedwith1:

65all65sampleswillbeplottedoneatatime.TheplottingprocesscanbehaltedatanytimebyenteringCtrl+ConthekeyboardandthenclosingtheFigurewindow.b.TypingPlotEEMby4（1,OriginalData,R.U.）willplotthedatafouratatime.Thisisquickerwhendealingwithlargedatasets.c.TypingPlotEEMby4FixZ（1,OriginalData,R.U.）doesthesameasabovebutplotsthedatawithafixedz-axis（colorbarscale）automaticallyderivedfromtheminandmaxmeasureddata.3.Tryusingthethreefunctions.AlsotrytypinghelpPlotEEMby1,helpPlotEEMby4,andfinallyhelpPlotEEMby4FixZ.SomeinstructionsonhowtousethefunctionsareprintedtotheCommandWindow.ThisistrueforallfunctionsinMATLAB（e.g.trytyp-inghelpload）.4.SurfaceplotscanalsobeplottedinasimilarwayusingthePlotSurfby1,andPlotSurfby4functions.C.Cuttingtheregionofthespectrainfluencedbyscatterpeaks.1.Thisstepcreatesanewcopyofthedatawherethewavelengthsinfluencedbyscatterpeakshavebeencutandreplacedwithmissingvaluesorzeros.TypeCutData=EEMCut（OriginalData,20,20,NaN,NaN,No）thedatawillbecutandthenplottedsothattheEEMsbeforeandafterthecutcanbecompared.Thegraphswillplotautomaticallyfromthefirstsampletothelast.Thefunctiondeletesthedataintheregionofnofluorescence（whereemissionwavelengthislessthanexcitationwavelength）andtheregionsgreatlyinflu-encedbyfirstorderscatter（whereRayleighandRamanpeaksdominatethesignal）andreplacesthemwithmissingvalues（“NaN”（NotANumber）inMAT-LAB）.AdditionallyaregionofzerosisinsertedtoassistthePARAFACmodeling.2.TypehelpEEMCuttoreadanexplanationofwhatthisfunctiondoes.3.Experimentwithchangingtheinputvaluesfor（20,20,NaN,NaN,No）andobservehowthedataiscutdifferently.4.BeforeproceedingwiththetutorialtypeCutData=EEMCut（OriginalData,20,20,NaN,NaN,）sothatthedataisprocessedappropriatelyfortherestofthetutorial.Thiswillcutthedata,butnotplottheresults.D.Initialexplorativedataanalysisandoutlieridentification.Inthisstep,aseriesofPARAFACmodelsareruninordertoexplorethedataforoutliersamples,noisywavelengths,orotherpotentialproblemswiththedatathatarenoteasilyidentifiedbyvisualanalysisoftheEEMplots.Thestepisstruc-turedintotwotests:

oneontheoriginal（complete）dataandasecondonamodifieddatasetwhereoutliershavebeenremoved.ThedatasetcanbereducedinsizebyselectingeveryotherEmissionwavelength,（i.e.300,304,306.nm,insteadofthemeasured300,302,304,306.Thiswillspeedthemod-StedmonandBroFluorescence-PARAFACanalysisofDOM2elingprocessconsiderably,withoutnecessarilynotablyinflu-encingtheoutcome（exceptthespectralresolutionoftheloadings）.Thisisnotdoneherebuttryitoutlater.Theques-tionstobeaddressedhereare;-Aretherespecificsamplesorwavelengthsthatareinflu-encingthemodelfitmuchmorethantheothers?

-Whatseemstobetheappropriate“ballpark”numberofPARAFACcomponents（i.e.thenumberofdistinctfluores-centphenomena）?

3-5components,8-10?

etc.Whendothespectraofthecomponentsfoundbeginnottolooklikeorganicfluorophoresbutmorelikescatterpeaksornoise?

1.TypehelpOutlierTestintotheCommandWindowandreadtheexplanationofthefunction.2.TypeTest1=OutlierTest（CutData,2,1,7,No,No）toperformthefirsttest.Afterpressinganykeythiswillrunaseriesofmodelsfrom2componentsto7components.Theresultsfromtheseinitialfivemod-els（2,3,4,5,6and7componentmodels）canbeeval-uatedinmanywaysbutforthistutorialwewillusetwotypesofplots;loadingsandleverages.Thesetestswilltakeapprox3minutesforyourcomputertorun.3.TypehelpPlotLoadingsandreadaboutthefunctionthentypePlotLoadings（Test1,2）andafigurewillbecreatedshowingthescoresandloadingsofthemodel（a,bandcinEquation1）.4.Plotsbandcshowtheemissionandexcitationload-ingsofthetwocomponents.Plotashowshowtheconcentrationofthetwocomponentsvariesbetweensamples.Checkthattheloadingslookreasonable,i.e.thattheyaresmoothandspectralappearancethatcor-respondswellwithyourunderstandingofthedata.5.Anotherusefulplotisthatofleverages.TypehelpPlotLeverageandreadaboutthefunctionandthentypePlotLeverage（Test1,2）andafigurewillbecreatedshowingtheleverages.Intheplot,lookforsamplesthathaveextremeleveragesindicatingextremeandpotentiallyoutlyingsamples.6.Tip:

UsingthefunctionPlotLLcreatesacombinationofboththeleverageandloadingplots.TrytypingPlotLL（Test1,2）7.Nowtrycreatingthesameplotsfortheothermodels（3,4,5,6and7components）.Afterexaminingtheloadingplotsitshouldbeapparentthatweneedtobegintoconstrainthemodel.Someofthe