meta analysis first stepsnemo.nic.uoregon.edu/wiki/images/c/c0/presentation...2009/09/23 · factor...
TRANSCRIPT
MetaAnalysisFirstSteps
DataAnalysis,MetricGenera4onandExtractedPa:ernAnnota4on
ProjectGoalsforMeta‐Analysis
• Subgoal#1:Completefirststa4s4calmeta‐analysisofERPpa:ernsfromNEMOconsor4umdatasets– TargetfirstpapersubmissionbyMay2009
• Subgoal#2:Comparepa:ernmappingsfromdifferentmeta‐analyses&establishfunc4onallyrelevantlinksbetweenpa:erns– Lexical,seman4c,&memory‐relatedERP
– Establishmeaningfulpa:ernclasses,hierarchiesbasedonmetaanalysesresults
MetaSummaryofERPDataMetaAnalysis
AnalyzeMark‐upLabelClusterLink
LabelLinkedClustersPublish
(rinseandrepeat)
MetaAnalysisSteps• ObtainERPdatasetswithcompa4blefunc4onalconstraints
– NEMOconsor4umdata
• Decompose/segmenttheERPdataintodiscretespa4o‐temporalpa:erns– PCA/ICA/MicrostateSegmenta4on
• Mark‐uppa:ernswiththeircategorical,func4onalandspa4o‐temporalcharacteris4cs– NEMOautolabel
• Labelpa:erns
• Clusterpa:ernswithindatasets
• Linklabeledclustersacrossdatasets
• Labellinkedclusters
• Publish
DatasetsforMeta‐analysis#1
DatasetsforMeta‐analysis#2
TechniquesforDecomposing/Segmen4ngERPDataIntoDiscreteSpa4o‐TemporalPa:erns
• ComponentSepara4on– PCA:PrincipalComponentsAnalysis
• Establishedprotocolwithsuppor4ngliterature– Dien,Frishkoff,Kayser&Tenke
• Appliedto9consor4umdatasetsfrom3separatelabs
– ICA:IndependentComponentsAnalysis• Establishedprotocolwithsuppor4ngliterature,thoughlessextensivethanPCAERPresearch– Makeigetal
• Mixedresults/interpreta4ondifficul4esw.r.t.consor4umdata
• AutomatedWindowing/MicrostateSegmenta4on• Establishedprotocolwithsuppor4ngliterature
– Lehman,Koenig,Murray• Inprogress:currentlyaddingtoNEMOautolabel
MicrostateSegmenta4onOverviewSimulatedMicrostates
• Simulateddatasetof4dis4nct,finitedura4on,topographies(microstates).Notetopographiesarepar4allyoverlapping
MicrostateSegmenta4onOverviewMicrostateBoundaries
• Microstateborderprobabilityfunc4on(MSBPF):Quan4fiesprobabilityoftopographicchangeasafunc4onof4me
OverviewofPCA
• TemporalPCA– Variables:Timesamples– Observa4ons:Channelwaveformsacrossconditons+subjects– Rela4onshipmatrixquan4fiestemporalcorrela4ons
Basisofapproachfordecomposingandsta2s2callyquan2fyingNEMOconsor2umdata
• Spa4alPCA– Variables:Channelloca4ons– Observa4ons:Spa4altopographiesacrossconditons+subjects– Rela4onshipmatrixquan4fiesspa4alcorrela4ons– Problema4cduetohighspa4aloverlapofpa:ernsfromvolumeconduc4on
Notusedduetoconcernsofmisalloca2onofvariance/factorspli=ng
PCADecomposi4onProtocolforAnalysisofConsor4umData
• DienPCAToolbox/ERPToolkit• ReadSegRaw/PCAtoRawtoimportfromandexportto
EGIsegmentedsimple‐binaryfiles
• TemporalPCAalgorithms– Covariancerela4onshipmatrix
– Kaiserfactorloadingnormaliza4on
– Retainallfactorspriortoanyandallrota4ons– Varimaxrota4onfollowedbyPromaxrelaxa4on
– Sta4s4callyanalyze25post‐rota4onfactors,sortedinorderofdecreasingprojectedvariance(basedonFacVar)
PCAtoRaw/ERPPCAToolbox• PCAtoRawrun‐4meparameters
• PCAtoRawinvokesDienERPPCAToolbox
PCAtoRaw/ERPPCAToolbox• NprawdataPCAdecomposi4onsummary
PCADecomposi4onProtocolFactorReten4on–PartI
• Pre‐rota4onfactorreten4on– “Theproblemofthenumber ofcomponents”
• Screetest– Linearscalemayunderes4matefactor
reten4on
• Paralleltest– Comparescreeofexperimentaldatato
screeofrandomdataofequaldimensions
• Fullpre‐rota4onreten4on– Kayser&Tenkeproposal– Factorreten4on,pre‐rota4on,affectsbothexplainedvarianceand
rota4onoutcome– Fullpre‐rota4onreten4oneliminateseffectofreten4onsubjec4vityon
rota4onoutcome
PCADecomposi4onProtocolFactorReten4on–Part2
• Post‐rota4onfactorreten4on– Determinenumberofretained
componentsforadequatereconstruc4onofscalprecordedERP
• RetainedcomponentsrepresentmajorityofERPvariance– FactorsaresortedonFacVar,thefrac4on
ofdatavarianceaccountedforbyeachindividualfactor– Defaultpost‐rota4onsortorderofERPPCAToolkit– Sta4s4calanalysis,viaNEMOautolabel,performedonretainedfactors– Flagretainedfactorswith“robust”varianceorhighrela4veGlobalFieldPower– Flagretainedfactorscontainingspa4otemporalcharacteris4csoftargetpa:erns
PCAtoRaw/ERPPCAToolboxNPraw.rawTestDatasetResults
• PCAtoRawoutputfiles:– .log: Summaryrunsta4s4cs
– .mat: MATLABworkspacevariables
– .fig: Pre‐andpost‐rota4onfactor screeplots
– .raw: Factorloadingsprojectedbackto
channelspace Onesetforeachconditon/cell Grandaverage(_G.raw)orsubject‐specific(_S##.raw)
PCAtoRaw/ERPPCAToolboxNPraw.rawTestDatasetResults
• ExamineNPraw_G.rawfactorwaveforms,inscalp‐surfacespace,ateachchannelacrosscondi4ons(TopoPlotMode)
Factor1waveforms(0‐900ms;0.1uv/mm).Notecondi4oneffects/factorsepara4onatcentropariertalandanteriorventralsites
PCAtoRaw/ERPPCAToolboxNPraw.rawTestDatasetResults
• ExamineNPraw_G.rawfactortopographies,inscalp‐surfacespace,atpeakintensityacrosscondi4ons(TopoMapMode)
Factor1scalp‐surfacetopographies,600mspost‐s4mulus,forthe4NPrawexperimentalcondi4ons(LtoR):ConFinal,ConMid,InconFinal,InconMid
NEMOautolabelMarkingupERPComponents/Microstates:NEMO_data
• Mark‐upobservedpa:erns(components/microstates)withuser‐specifiedinforma4onontheexperimentalprocedureandsubjectgroup
• Eachmark‐upelement(NEMOautolabellabel)hasauniqueNEMOautolabelIDandwillmaptoacorrespondingelementintheNEMOontology
ERP_CompAnalysisMethod
EEG_Montage
ExptID
SessID
CellNo
CellLabel
ERP_ObsID
SubjID
Subject_Group
ERP_ObservedPattern
Cond_Stan
Event_Modality
Event_Type
Stim_Type
ExptID ExptIDrepresents"experimentID"andspecifiestheexperimentalprocedure
andsubjectgroup.
AL:0000003 experiment_id NM:0000059
NEMOautolabel_Name NEMOautolabel_Def NEMOautolabel_ID NEMOlex_Name NEMOlex_ID
NEMOautolabelMarkingupERPComponents/Microstates:NEMO_data
• Mark‐upobservedpa:erns(components/microstates)withtheirtemporalcharacteris4cs
• MATLAB‐basedfunc4onsextracttemporalmetricsforeachcondi4on,subjectandcomponent/microstate– Datadriven– Harnessesexpert‐knowledge:Domainexpertsspecifythetemporalcharacteris4csofinterest
Ti_Max
Ti_Max_round
TI_Begin
TI_End
TI_Dur
TI_Dur_round
Ti_Max Ti_Maxspecifiesforeachtemporalcomponentthe4mepointofitspeakabsoluteintensity,in
milliseconds.
AL:0000019 ERP_pa:ern_peak_latency NM:0000047
NEMOautolabel_Name NEMOautolabel_Def NEMOautolabel_ID NEMOlex_Name NEMOlex_ID
NEMOautolabelMarkingupERPComponents/Microstates:NEMO_data
• Mark‐upobservedpa:erns(components/microstates)withtheirspa4alcharacteris4cs• MATLAB‐basedfunc4onsextractspa4almetricsforeachcondi4on,subjectand
component/microstate– Datadriven– Harnessesexpert‐knowledge:Domainexpertsspecifythespa4alcharacteris4csofinterest
COP_X2d
COP_Y2d
CON_X2d
CON_Y2d
COP_X3d
COP_Y3d
COP_Z3d
CON_X3d
CON_Y3d
CON_Z3d
ITTCh_COP Interna4onal10‐10electrodeloca4onclosesttothe
componentpair'scenter‐of‐posi4vityxy‐coordinatepair(COP_X2d,COP_Y2d),inL2‐norm,onamontage‐specific2‐Dflatmapofscalp‐surface
electrodeloca4ons.
AL:0000036 TBA TBA
NEMOautolabel_Name NEMOautolabel_Def NEMOautolabel_ID NEMOlex_Name NEMOlex_ID
EGICh_COP
ITTCh_COP
ROI_COP
EGICh_CON
ITTCh_CON
ROI_CON
LatIndex_Threshold
Laterality_COP
LatIndex_COP
ROInolat_COP
Laterality_CON
LatIndex _CON
ROInolat_CON
ReferencesPCA
Dien,J.(1998).Addressingmisalloca4onofvarianceinprincipalcomponentsanalysisofevent‐relatedpoten4als.BrainTopogr,11(1),43‐55.
Dien,J.,&Frishkoff,G.A.(2005).Introduc4ontoprincipalcomponentsanalysisofevent‐relatedpoten4als.InT.Handy(Ed.),Event‐RelatedPoten4als:AMethodsHandbook.(pp.189‐208).Cambridge,MA:MITPress.
Dien,J.,Beal,D.J.,&Berg,P.(2005).Op4mizingprincipalcomponentsanalysisofevent‐relatedpoten4als:matrixtype,factorloadingweigh4ng,extrac4on,androta4ons.ClinNeurophysiol,116(8),1808‐1825.
Dien,J.(2006).ProgressingtowardsaconsensusonPCAofERPs.ClinNeurophysiol,117(3),699‐702;authorreply703‐697.
Dien,J.,Khoe,W.,&Mangun,G.R.(2007).Evalua4onofPCAandICAofsimulatedERPs:Promaxvs.Infomaxrota4ons.HumBrainMapp,28(8),742‐763.
Dien,J.(2009).Evalua4ngtwo‐stepPCAofERPdatawithGeomin,Infomax,Oblimin,Promax,andVarimaxrota4ons.Psychophysiology.
Kayser,J.,&Tenke,C.E.(2003).Op4mizingPCAmethodologyforERPcomponentiden4fica4onandmeasurement:theore4calra4onaleandempiricalevalua4on.ClinNeurophysiol,114(12),2307‐2325.
Kayser,J.,&Tenke,C.E.(2005).Trus4nginorbreakingwithconven4on:towardsarenaissanceofprincipalcomponentsanalysisinelectrophysiology.ClinNeurophysiol,116(8),1747‐1753.
ReferencesICA
Dien,J.,Khoe,W.,&Mangun,G.R.(2007).Evalua4onofPCAandICAofsimulatedERPs:Promaxvs.Infomaxrota4ons.HumBrainMapp,28(8),742‐763.
MicrostateAnalysis
Michel,C.M.,Murray,M.M.,Lantz,G.,Gonzalez,S.,Spinelli,L.,&GravedePeralta,R.(2004).EEGsourceimaging.ClinNeurophysiol,115(10),2195‐2222.
Murray,M.M.,Brunet,D.,&Michel,C.M.(2008).TopographicERPanalyses:astep‐by‐steptutorialreview.BrainTopogr,20(4),249‐264.
Koenig,T.,Kochi,K.,&Lehmann,D.(1998).Event‐relatedelectricmicrostatesofthebraindifferbetweenwordswithvisualandabstractmeaning.ElectroencephalogrClinNeurophysiol,106(6),535‐546.
Koenig,T.,&Lehmann,D.(1996).Microstatesinlanguage‐relatedbrainpoten4almapsshownoun‐verbdifferences.BrainLang,53(2),169‐182.
Lehman,D.,&Skrandies,W.(1985).Spa4alanalysisofevokedpoten4alsinman‐Areview.ProgressinNeurobiology,23,227‐250.
Pizzagalli,D.,Lehmann,D.,Koenig,T.,Regard,M.,&Pascual‐Marqui,R.D.(2000).Face‐elicitedERPsandaffec4veautude:brainelectricmicrostateandtomographyanalyses.ClinNeurophysiol,111(3),521‐531.
ReferencesAnnota4ngfunc4onala:ributes
Fox,P.T.,Laird,A.R.,Fox,S.P.,Fox,P.M.,Uecker,A.M.,Crank,M.,etal.(2005).BrainMaptaxonomyofexperimentaldesign:descrip4onandevalua4on.HumBrainMapp,25(1),185‐198.
Spa4al&temporalmetricgenera4on
Handy,T.(2005).BasicPrinciplesofERPQuan4fica4on.InT.Handy(Ed.),Event‐RelatedPoten4als:AMethodsHandbook(pp.33–56).Cambridge,MA:MITPress.
Luck,S.(2005).AnIntroduc4ontotheEvent‐RelatedPoten4alTechniqueBoston,MA:TheMITPress.
O:en,L.J.,&Rugg,M.D.(2005).Interpre4ngEvent‐RelatedBrainPoten4als.InT.Handy(Ed.),Event‐RelatedPoten4als:AMethodsHandbook(pp.3–16).Cambridge,MA:MITPress.
Picton,T.W.,Ben4n,S.,Berg,P.,Donchin,E.,Hillyard,S.A.,Johnson,R.,Jr.,etal.(2000).Guidelinesforusinghumanevent‐relatedpoten4alstostudycogni4on:recordingstandardsandpublica4oncriteria.Psychophysiology,37(2),127‐152.