Corresponding author: Frank G. Hillary, Department of ... · Graph theory approaches to functional...
Transcript of Corresponding author: Frank G. Hillary, Department of ... · Graph theory approaches to functional...
Graphtheoryasatooltounderstandbraindisorders
1
Graphtheoryapproachestofunctionalnetworkorganizationinbraindisorders:
Acritiqueforabravenewsmall-world
MichaelN.Hallquist1,2
and
FrankG.Hillary1,2,3
1DepartmentofPsychology,PennsylvaniaStateUniversity,UniversityPark,PA
2SocialLifeandEngineeringSciencesImagingCenter,UniversityPark,PA
3DepartmentofNeurology,HersheyMedicalCenter,Hershey,PA
AcceptedforpublicationinNetworkNeuroscience.
Corresponding author: Frank G. Hillary, Department of Psychology, Penn State University, 140 Moore Building, University Park, PA 16802. Email: [email protected]. This research was supported by a grant from the National Institute of Mental Health to MNH (K01 MH097091). Acknowledgments.WethankZachCeneviva,AllenCsuk,RichardGarcia,MelanieGlatz,andRiddhiPatelfortheirworkcollecting,organizing,andcodingreferencesfortheliteraturereviewandmanuscript.Keywords:graphtheory,braindisorders,networkneuroscience,proportionalthresholding
Graphtheoryasatooltounderstandbraindisorders
2
Abstract
Overthepasttwodecades,resting-statefunctionalconnectivity(RSFC)methods
haveprovidednewinsightsintothenetworkorganizationofthehumanbrain.Studiesof
braindisorderssuchasAlzheimer’sdiseaseordepressionhaveadaptedtoolsfromgraph
theorytocharacterizedifferencesbetweenhealthyandpatientpopulations.Here,we
conductedareviewofclinicalnetworkneuroscience,summarizingmethodologicaldetails
from106RSFCstudies.Althoughthisapproachisprevalentandpromising,ourreview
identifiedfourchallenges.First,thecompositionofnetworksvariedremarkablyintermsof
regionparcellationandedgedefinition,whicharefundamentaltographanalyses.Second,
manystudiesequatedthenumberofconnectionsacrossgraphs,butthisisconceptually
problematicinclinicalpopulationsandmayinducespuriousgroupdifferences.Third,few
graphmetricswerereportedincommon,precludingmeta-analyses.Fourth,somestudies
testedhypothesesatonelevelofthegraphwithoutaclearneurobiologicalrationaleor
consideringhowfindingsatonelevel(e.g.,globaltopology)arecontextualizedbyanother
(e.g.,modularstructure).Basedonthesethemes,weconductednetworksimulationsto
demonstratetheimpactofspecificmethodologicaldecisionsoncase-controlcomparisons.
Finally,weoffersuggestionsforpromotingconvergenceacrossclinicalstudiesinorderto
facilitateprogressinthisimportantfield.
Graphtheoryasatooltounderstandbraindisorders
3
Introduction
Effortstocharacterizea“humanconnectome”havebroughtsweepingchangesto
functionalneuroimagingresearch,withmanyinvestigatorstransitioningfromindicesof
localbrainactivitytomeasuresofinter-regionalcommunication(Friston,2011).Thebroad
goalofthisconceptualrevolutionistounderstandthebrainasafunctionalnetworkwhose
coordinationisresponsibleforcomplexbehaviors(Biswaletal.,2010).Theprevailing
approachtostudyingfunctionalconnectomesinvolvesquantifyingcouplingoftheintrinsic
brainactivityamongregions.Inparticular,resting-statefunctionalconnectivity(RSFC)
methods(Biswal,Yetkin,Haughton,&Hyde,1995)focusoninter-regionalcorrespondence
inlow-frequencyoscillationsoftheBOLDsignal(approximately0.01-0.12Hz).
WorkoverthepasttwodecadeshasdemonstratedthevalueofRSFCapproachesfor
mappingfunctionalnetworkorganization,includingtheidentificationofseparablebrain
subnetworks(Biswaletal.,2010;Lairdetal.,2009;Poweretal.,2011;Smithetal.,2009;
vandenHeuvel&HulshoffPol,2010).BecauseRSFCmethodsdonotrequirethestudy-
specificdesignsandcognitiveburdenassociatedwithtask-basedfMRIstudies,RSFCdata
aresimpletoacquireandhavebeenusedinhundredsofstudiesofhumanbrainfunction.
Nevertheless,therearenumerousmethodologicalchallenges,includingconcernsaboutthe
qualityofRSFCdata(Poweretal.,2014)andtheeffectofdataprocessingonsubstantive
conclusions(Ciricetal.,2017;Hallquist,Hwang,&Luna,2013;Shirer,Jiang,Price,Ng,&
Greicius,2015).
RSFCstudiesofbraininjuryordiseasetypicallyexaminedifferencesinthe
functionalconnectomesofaclinicalgroup(e.g.,Parkinson’sdisease)comparedtoa
matchedcontrolgroup.Therearespecificmethodologicalandsubstantiveconsiderations
Graphtheoryasatooltounderstandbraindisorders
4
thatapplytoRSFCstudiesofbraindisorders.Forexample,differencesintheoveralllevel
offunctionalconnectivitybetweenpatientandcontrolgroupscouldleadtodifferencesin
thenumberofspuriousconnectionsinnetworkanalyses,potentiallyobscuringmeaningful
groupcomparisons(vandenHeuveletal.,2017).Likewise,thereisincreasingconcernin
theclinicalneurosciencesthatanunacceptablysmallpercentageoffindingsarereplicable
(Mülleretal.,2017).Suchconcernsechothegrowingemphasisonopen,reproducible
practicesinneuroimagingmoregenerally(Poldracketal.,2017).
Inthispaper,wereviewthecurrentstateofgraphtheoryapproachestoRSFCinthe
clinicalneurosciences.Basedonkeythemesinthisliterature,weconductedtwonetwork
simulationstodemonstratethepitfallsofspecificanalysisdecisionsthathaveparticular
relevancetocase-controlstudies.Finally,weproviderecommendationsforbestpractices
topromotecomparabilityacrossstudies.
Ourreviewdoesnotdirectlyaddressmanyimportantmethodologicalissuesthat
areactiveareasofinvestigation.Forexample,detectingandcorrectingmotion-related
artifactsremainsacentralchallengeinRSFCstudies(Ciricetal.,2017;Dosenbachetal.,
2017)thatisespeciallyproblematicinclinicalanddevelopmentalsamples(Greene,Black,
&Schlaggar,2016;VanDijk,Sabuncu,&Buckner,2012).Thelengthofresting-statescans
(intermsoftime,numberofmeasurements,andnumberofsessions)alsoinfluencesthe
reliabilityoffunctionalconnectivityestimates(Birnetal.,2013;Gonzalez-Castillo,Chen,
Nichols,&Bandettini,2017)andlikelyhasdownstreamconsequencesonnetworkanalyses
(Abroletal.,2017;Yangetal.,2012).Finally,brainparcellation—definingthenumberand
formofbrainregions—isoneofthemostimportantinfluencesonthecompositionof
RSFCnetworks.Therearenumerousparcellationapproaches,includinganatomicalatlases,
Graphtheoryasatooltounderstandbraindisorders
5
functionalboundarymapping,anddata-drivenalgorithmsbasedonvoxelwiseBOLDtime
series(Goñietal.,2014;Honnoratetal.,2015).InordertofocusonRSFCgraphtheory
researchintheclinicalliterature,whererelevant,wereferreaderstomorefocused
treatmentsofimportantissuesthatarebeyondthescopeofthispaper.
ALiteratureReviewofClinicalNetworkNeuroscienceStudies
Graphtheoryisabranchofdiscretemathematicsthathasbeenappliedinnumerous
studiesofbrainnetworks,bothstructuralandfunctional.Agraphisacollectionofobjects,
calledverticesornodes;thepairwiserelationshipsamongnodesarecallededgesorlinks
(Newman,2010).GraphscomposedofRSFCestimatesamongregionsprovideawindow
intotheintrinsicconnectivitypatternsinthehumanbrain.Figure1providesasimple
schematicofthemostcommongraphtheoryconstructsandmetricsreportedinRSFC
studies.Formorecomprehensivereviewsofgraphtheoryapplicationsinnetwork
neuroscience,seeBullmoreandSporns(2009,2012),orFornito,Zalesky,andBullmore
(2016).
Graphtheoryasatooltounderstandbraindisorders
6
Figure1.Atoygraphandrelatedgraphtheoryterminology.AdaptedwithpermissionfromHillary&Grafman,2017.
Thegoalofourliteraturesearchwastoobtainarepresentativecross-sectionof
graph-theoreticRSFCstudiesspanningneurologicalandmentaldisorders.Wefocusedon
functionalconnectivity(FC)asopposedtostructuralconnectivity,whereadistinctsetof
methodologicalcritiquesarelikelyrelevant.Also,wereviewedfMRIstudiesonly,
excludingelectroencephalography(EEG)andmagnetoencephalography(MEG).Although
thereareimportantadvantagesofEEG/MEGinsomerespects(Papanicolaouetal.,2017),
wefocusedonfMRIinpartbecausethevastmajorityofclinicalRSFCstudieshaveusedthis
modality.Inaddition,therearefMRI-specificconsiderationsfornetworkdefinitionand
spatialparcellationinRSFCstudies.WeconductedtworelatedsearchesofthePubMed
database(http://www.ncbi.nlm.nih.gov/pubmed)toidentifyarticlesfocusingongraph-
Clustering coefficient: local density of connections, or in this example (A and B), the number of triangles over the number of connected triples (“how many of my friends are friends with each other”). For (A) the clustering coefficient is 1, for (B) it is 0.
Degree (of a node): number of edges incident to that node.
Density: The density of the graph is the ratio of the number of observed connections in the network to the number of possible connections.
Edges: (pairwise) connections between nodes.
Hub: node with high degree or high centrality. Node b (red) has high degree and betweenness, and node c (blue) has high degree within the network. Betweenness centrality of a node reflects the frequency with which it sits along the shortest path between any two nodes in the graph.
Modules/Communities: clusters of nodes densely linked amongst themselves and more sparsely linked to nodes outside the cluster. Three communities circled in red.
Node: brain region of interest.
Path Length: here, the shortest distance (in number of edges) separating two nodes; e.g., in D from a to d = 5
Weight (of an edge): relationship between corresponding pair of nodes (e.g., correlation). In (A) and (B) the network is unweighted, while (C) is weighted.
A
C
0.4 0.20.6
0.7
D
a
bc
d
B
Graphtheoryasatooltounderstandbraindisorders
7
theoreticapproachestoRSFCinmentalandneurologicaldisorders(fordetailsonthe
queries,seeMethods).
ThesesearcheswereruninApril2016andresultedin626potentialpapersfor
review(281fromneurologicalquery,345frommentaldisorderquery).Studieswere
excludediftheywerereviews,casestudies,animalstudies,methodologicalpapers,used
electrophysiologicalmethods(e.g.,EEGorMEG),reportedonlystructuralimaging,ordid
notfocusonbraindisorders(e.g.,healthybrainfunctioning,normalaging).After
exclusions,thesetwosearchesyieldeddistinct106studiesincludedinthereview(see
Table1).AfulllistingofallstudiesreviewedisprovidedintheSupplementaryMaterials.
Table1.Clinicaldisordersrepresentedinthereviewof106clinicalnetworkneurosciencestudies
Note.ADHD:attentiondeficithyperactivitydisorder,MCI:mildcognitiveimpairment
ClinicalPhenotype n(frequency)Alzheimer’sDisease/MCI 19(17.9%)Epilepsy/Seizuredisorder 13(12.3%)Depression/Affective 12(11.3%)Schizophrenia 11(10.4%)Alcohol/SubstanceAbuse 7(6.6%)Parkinson’sDisease/Subcortical 6(5.7%)TraumaticBrainInjury 6(5.7%)Anxietydisorders 5(4.7%)ADHD 5(4.7%)Stroke 4(2.8%)Cancer 3(2.8%)MultipleSclerosis 2(1.9%)AutismSpectrumDisorder 2(1.9%)Disordersofconsciousness 2(1.9%)Somatizationdisorder 2(1.9%)DualDiagnosis 2(1.9%)OtherNeurologicaldisorder 3(2.8%)OtherPsychiatricdisorder 2(1.9%)
Graphtheoryasatooltounderstandbraindisorders
8
Below,weoutlineimportantgeneralthemesfromtheliteraturereview(ourmajor
concernsaresummarizedinBox1),includingtheheterogeneityofdataanalytic
approachesacrossgraphtheoreticalstudies.Wethenturnourfocustotwocriticalissues
withimportantimplicationsforinterpretingnetworkanalysesincase-controlstudies:1)
networkthresholding(i.e.,determininghowtodefinea“connection”fromacontinuous
measureofFC)and2)matchingthehypothesistothelevelofinquiryinthegraph.Foreach
oftheseissues,weoffernetworksimulationstoillustratetheimportanceoftheseissuesfor
case-controlcomparisons.
Box1:Majorchallengesingraph-theoreticstudiesoffunctionalconnectivityinbrain
disorders
1) There is substantial heterogeneity in defining the nodes and edges of graphs
across resting-state studies that largely precludes quantitative comparisons and formal meta-analyses. Key sources of heterogeneity include: brain parcellations, functional connectivity metrics (e.g., full versus partial correlation), handling of negative FC estimates, and preprocessing decisions (e.g., global signal regression). Furthermore, few graph metrics were reported in common, further undermining an examination of similarities across studies and disorders.
2) Many studies use proportional thresholding to convert a continuous functional connectivity metric (e.g., Pearson correlation) to a binary edge in the graph while controlling for the total number of edges. This approach may obscure the effects of brain pathology in case-control designs, where clinical groups may have changes in the number and strength of functional connections. Proportional thresholding may also identify spurious group differences when groups differ in the connectivity strength of selected brain regions.
3) The link between many graph metrics and neurobiology has been more commonly presumed than established (e.g., neural efficiency).
4) The alignment between study hypotheses and specific graph analyses was often unclear. Moreover, many studies have not considered how findings at one level (e.g., global topology) are contextualized by another (e.g., modular structure).
Graphtheoryasatooltounderstandbraindisorders
9
CreatingComparableNetworksinClinicalSamples
Definingnodesinfunctionalbrainnetworks.GraphtheoryanalysesofRSFCdata
fundamentallydependonthedefinitionofnodes(i.e.,brainregions)andedges(i.e.,the
quantificationoffunctionalconnectivity).Fornetworkanalysestorevealnewinsightsinto
clinicalphenomena,investigatorsmustchooseaparcellationschemethatrobustlysamples
theregionsandnetworksofinterest.Ourliteraturereviewrevealedthat76%ofstudies
definedgraphsbasedoncomprehensiveparcellations(i.e.,samplingmostorallofthe
brain),whereas24%analyzedconnectivityintargetedsub-networks(e.g.,motorregions
only;Table2a).Furthermore,wefoundsubstantialheterogeneityinparcellations,ranging
from10to67632nodes(Mode=90;M=1129.2;SD=7035.9).Infact,whereas25%of
studieshad90nodes(mostoftheseusedtheAALatlas;Tzourio-Mazoyeretal.,2002),the
frequencyofallotherparcellationsfellbelow5%,resultinginatleast50distinct
parcellationsin106studies.
Graphtheoryasatooltounderstandbraindisorders
10
Table2.Networkconstructionandedgedefinition
a.Networkconstruction n(frequency)Comprehensiveregionsampling 80(75.5%)Targetedregionsampling 26(24.5%)
b.Edgedefinition n(frequency)Weightednetwork 48(45.3%)Binarynetwork 42(39.6%)Both 13(12.3%)Unknown/unclear 3(2.8%)
c.EdgeFCstatistic n(frequency)Correlation(typically,Pearson’sr) 82(77.3%)Partialcorrelation 11(10.4%)Waveletcorrelation 6(5.7%)Causalmodeling(e.g.,SEM) 3(2.8%)Other/unclear 4(3.7%)
d.TreatmentofNegativeedges n(frequency)Notreported/unclear 61(57.5%)Discardnegativevalues 23(21.7%)Absolutevalue 10(9.4%)Maintained/analyzed 9(8.5%)Othertransformation 3(2.8%)Note.FC=functionalconnectivity;SEM=structuralequationmodeling
Althoughitmayseemself-evident,itbearsmentioningthatseveralpopular
parcellationschemesprovideabroad,butnotcomplete,samplingoffunctionalbrain
regions.Forexample,recentparcellationsbasedonthecorticalsurfaceofthebrain(e.g.,
Glasseretal.,2016;Gordonetal.,2016)haveprovidedanewlevelofdetailonfunctional
boundariesinthecortex.Yetifaresearcherisinterestedincortical-subcortical
connectivity,itiscrucialthattheparcellationbeextendedtoincludeallrelevantregions.
Thereareadvantagesandchallengestoeveryparcellationapproach(e.g.,Honnorat
etal.,2015;Poweretal.,2011);here,wefocusontwospecificconcerns.First,agoalof
mostclinicalnetworkneurosciencestudiesistodescribegroupdifferencesinwhole-brain
Graphtheoryasatooltounderstandbraindisorders
11
connectivitypatternsthatarereasonablyrobusttothegraphdefinition.Thus,investigators
maywishtouseatleasttwoparcellationsinthesamedatasettodetermineifthefindings
areparcellation-dependent.Becauseofthefundamentaldifficultyincomparingunequal
networks(vanWijk,Stam,&Daffertshofer,2010),onewouldnotexpectidenticalfindings.
Inparticular,globaltopologicalfeaturessuchasefficiencyorcharacteristicpathlength
mayvarybyparcellation,butotherfeaturessuchasmodularityandhubarchitectureare
likelytobemorerobust.Applyingmultipleparcellationstothesamedatasetincreasesthe
numberofanalysesmultiplicatively,aswellastheneedtoreconcileinconsistentfindings.
Cassidyandcolleagues(2018)recentlydemonstratedthatpersistenthomology,a
techniquefromtopologicalanalysis,hasthepotentialtoquantifythesimilarityof
individualfunctionalconnectomesacrossdifferentparcellations.Webelievethatsuch
directionsholdpromiseforsupportingreproduciblefindingsinclinicalRSFCstudies
(Craddock,James,Holtzheimer,Hu,&Mayberg,2012;Royetal.,2017;Schaeferetal.,in
press).
Second,differencesinparcellationfundamentallylimittheabilitytocompare
studies,bothdescriptivelyandquantitatively.Asnotedabove,ourreviewrevealed
substantialheterogeneityinparcellationschemesacrossstudiesofbraindisorders.
Extendingourconcernaboutparcellationdependence,suchheterogeneitymakesit
impossibletoascertainwhetherdifferencesbetweentwostudiesofthesameclinical
populationareanartifactofthegraphdefinitionorameaningfulfinding.Moreover,
whereasmeta-analysesofstructuralMRIandtask-basedfMRIstudieshavebecome
increasinglypopular(e.g.,Goodkindetal.,2015;Mülleretal.,2017),suchanalysesarenot
currentlypossibleingraph-theoreticstudiesinpartbecauseofdifferencesinparcellation.
Graphtheoryasatooltounderstandbraindisorders
12
Toresolvethisissue,weencouragescientiststoreportresultsforafield-standard
parcellation,apointweelaborateinourrecommendationsbelow.Wealsobelievethereis
valueininvestigatorshavingfreedomtotestandcompareadditionalparcellationsthat
mayhighlightspecificfindings.
Definingedgesinfunctionalbrainnetworks.Parcellationdefinesthenodes
comprisingagraph,butanequallyimportantdecisionishowtodefinefunctional
connectionsamongnodes(i.e.,theedges).Thevastmajority(77%)ofthestudiesreviewed
usedbivariatecorrelation,especiallyPearsonorSpearman,asthemeasureofFC.InCritical
Issue1below,weconsiderhowFCestimatesarethresholdedinordertodefinesedgesas
presentorabsentinbinarygraphs.
Quantifyingfunctionalconnectivity.ThestatisticalmeasureofFChasimportant
implicationsfornetworkdensityandtheinterpretationofrelationshipsamongbrain
regions.Theprevailingapplicationofmarginalassociation(typically,bivariatecorrelation)
doesnotseparatethe(statistically)directconnectivitybetweentworegionsfromindirect
effectsattributabletoadditionalregions.Bycontrast,mostconditionalassociation
methods(e.g.,partialcorrelation)relyoninvertingthecovariancematrixamongall
regions,therebyremovingcommonvarianceanddefiningedgesbasedonunique
connectivitybetweentworegions(Smithetal.,2011).
Atthepresenttime,thereisnoclearresolutiononwhetherFCshouldbebasedon
marginalorconditionalassociation(forusefulreviews,seeFallani,Richiardi,Chavez,&
Achard,2014;Varoquaux&Craddock,2013).Nevertheless,wewishtohighlightthatasthe
averagefullcorrelationincreaseswithinanetwork,partialcorrelationvalues,onaverage,
mustdecrease.Furthermore,partialandfullcorrelationsrevealfundamentallydifferent
Graphtheoryasatooltounderstandbraindisorders
13
graphtopologiesinresting-statedata(Cassidyetal.,2018)andmayyielddifferent
conclusionsaboutthesamescientificquestion(e.g.,identifyingfunctionalhubs;seeLiang
etal.,2012).Incasesofneuropathologywhereonemightexpectdistinctgainsorlossesof
functionalconnections,theuseofpartialcorrelationshouldbeinterpretedbaseduponthe
relativeedgedensityandmeanFC.Ofthestudiesreviewedhere,10%usedpartial
correlation,butvirtuallynostudyaccountedforpossibledifferencesinedgedensity(see
Table2c).
Negativeedges.Anotherimportantaspectofdefiningedgesishowtohandle
negativeFCestimates.FullcorrelationsofRSFCdatatypicallyyieldanFCdistributionin
whichmostedgesarepositive,butanappreciablefractionarenegative.Bycontrast,partial
correlationmethodsoftenyieldarelativebalancebetweenpositiveandnegativeFC
estimates(e.g.,Y.Wang,Kang,Kemmer,&Guo,2016).Thereremainslittleconsensusfor
handlingorinterpretingnegativeedgeweightsinRSFCgraphanalyses(cf.Murphy&Fox,
2017).Inourreview,57%ofthestudiesreportedinsufficientornoinformationabouthow
negativeedgeswerehandledingraphanalyses(Table2d).Twenty-onepercentofstudies
deletednegativeedgespriortoanalysis,and9%includedthenegativeweightsaspositive
weights(i.e.,usingtheabsolutevalueofFC).Asdetailedelsewhere,somegraphmetricsare
eithernotdefinedorneedtobeadaptedwhennegativeedgesarepresent(Rubinov&
Sporns,2011).
Importantly,themeanofthemarginalassociationRSFCdistributiondependson
whetherglobalsignalregression(GSR)isincludedinthepreprocessingpipeline(Murphy,
Birn,Handwerker,Jones,&Bandettini,2009).WhenGSRisincluded,thereisoftena
balancebetweenpositiveandnegativecorrelations.IfGSRisincludedasanuisance
Graphtheoryasatooltounderstandbraindisorders
14
regressor,alargefractionofFCestimatesmaysimplybediscardedasirrelevanttocase-
controlcomparisons,whichisamajor,untestedassumption.ThemeaningofnegativeFC,
however,remainsunclear,withseveralinvestigatorsattributingnegativecorrelationsto
statisticalartifactsandGSR(Murphyetal.,2009;Murphy&Fox,2017;Saadetal.,2012).
GiventhatnegativecorrelationsareobservableintheabsenceofGSR,however,
othershaveexaminedwhethernegativeweightscontributedifferentiallytoinformation
processingwithinthenetwork(Parenteetal.,2017).Negativecorrelationsmayalsoreflect
NMDAactionincorticalinhibition(Anticevicetal.,2012).Theseconnectionsbear
considerationgiventhatbrainnetworkscomposedofonlynegativeconnectionsdonot
retainasmall-worldtopology,butdohavepropertiesdistinctfromrandomnetworks
(Parenteetal.,2017;Schwarz&McGonigle,2011).Altogether,theomissionof
methodologicaldetailsaboutnegativeFCinempiricalreportsseverelyhampersthe
resolutionofthisimportantchoicepointindefininggraphs.
Degreedistributionasafundamentalgraphmetric.Afterresolvingthe
questionsofnodeandedgedefinition,wealsobelieveitiscrucialforstudiestoreport
informationaboutglobalnetworkmetricssuchascharacteristicpathlength,clustering
coefficient(akatransitivity),anddegreedistribution.AsnotedinTable3,localandglobal
efficiencywerecommonlyreported(71%and74%,respectively)andtypicallyacross
multipleFCthresholds.However,ourreviewrevealedthatonly27%ofstudiesprovided
cleardescriptivestatisticsformeandegree,and16%plottedthedegreedistribution.In
binarygraphs,thedegreedistributiondescribestherelativefrequencyofedgesforeach
nodeinthenetwork.Asimilarpropertycanbequantifiedbyexaminingthestrength(aka
cost)distributioninweightedgraphs.Wearguethatpresentationofthedegree(strength)
Graphtheoryasatooltounderstandbraindisorders
15
distributioninpublishedreportsisvitaltounderstandinganyRSFCnetworkforafew
reasons.First,itprovidesasanitycheckonthedata.Onecurrentperspectiveisthathuman
brainsareorganizedtomaximizecommunicationwhileminimizingwiringandmetabolic
cost(Bassettetal.,2009;Betzeletal.,2017;Chen,Wang,Hilgetag,&Zhou,2013;Tomasi,
Wang,&Volkow,2013),Thus,whenexaminingwhole-brainRSFCdata,themosthighly
connectedregionsarerareandshouldbeevidentinthetailofthedegree/strength
distribution(Achard,Salvador,Whitcher,Suckling,&Bullmore,2006).Second,reporting
detailsofthedegreedistributionfacilitatescomparisonsofgraphtopologyacrossstudies,
aswellastheimpactofpreprocessingandanalyticdecisionssuchasFCthresholding.
Finally,examiningthedegreedistributionasafirststepindataanalysismayoffer
otherwiseunavailableinformationaboutthenetworktopologyinhealthyandclinical
samples.Forexample,inpriorwork,weisolatededgesfromthetailofthedegree
distributiontounderstandtheimpactofthemosthighlyconnected,andrare,nodesonthe
network(Hillaryetal.,2014).
Table3.GraphmetricscommonlyreportedinclinicalRSFCstudies
Graphmetrics n(frequency)DegreeDistribution(plotted) 17(16.0%)MeanDegree(weightedorbinary) 29(27.4%)Clustering/Localefficiency 76(71.7%)PathLength/Globalefficiency 79(74.5%)Smallworldness 33(31.1%)Modularity(e.g.,Q-value) 21(19.8%)
Note.Totalfrequencyisgreaterthan100%becausesomestudiesreportedmorethanoneofthesemetrics.
Graphtheoryasatooltounderstandbraindisorders
16
CriticalIssue1:EdgeThresholdingandComparingUnequalNetworks
Wenowfocusonedgethresholding—thatis,howtotransformacontinuous
measureofFCintoanedgeinthegraph.Inourreview,39%ofstudiesbinarizedFCvalues
suchthatedgeswereeitherpresentorabsentinthegraph,whereas45%ofstudies
retainedFCasedgeweights(seeTable2b).Regardlessofwhetherinvestigatorsanalyze
binaryorweightednetworks,therearefundamentalchallengestocomparingnetworks
thatdiffereitherintermsofaveragedegree(k)orthenumberofnodes(N)(Fornitoetal.,
2016).Inparticular,comparinggroupsongraphmetricssuchaspathlengthandclustering
coefficientcanbeambiguousbecausethesemetricshavemathematicaldependencieson
bothkandN(vanWijketal.,2010).
Mostbrainparcellationapproachesdefinegraphswithanequivalentnumberof
nodes(N)ineachgroup.Ontheotherhand,connectiondensityisoftenavariableof
interestinclinicalstudieswherethepathologymayalternotonlyconnectionstrength,but
alsothenumberofconnections.IfNisconstant,variationinkbetweengroupsconstrains
theboundariesoflocalandglobalefficiency.Iftwogroupsdiffersystematicallyinedge
density,thisalmostguaranteesbetween-groupdifferencesinmetricssuchasclustering
coefficientandpathlength.Determiningwheretointerveneinthiscircularproblemhas
greatimportanceinclinicalnetworkneuroscience,wherehypothesesoftenfocusonthe
numberandstrengthofnetworkconnections.
Toaddressthisissue,severalinvestigatorshaverecommendedproportional
thresholding(PT)inwhichtheedgedensityisequatedacrossnetworks(Achard&
Bullmore,2007;Bassettetal.,2009;Power,Fair,Schlaggar,&Petersen,2010).
Furthermore,toreducethepossibilitythatfindingsarenotspecifictothechosendensity
Graphtheoryasatooltounderstandbraindisorders
17
threshold,29%ofstudieshavetestedforgroupdifferencesacrossarange(e.g.,5-25%;
Table4).However,wearguethatdefiningedgesbasedonPTmaynotbeidealforclinical
studies,wherethereareoftenregionaldifferencesinfunctionalcouplingorpathology-
inducedalterationsinthenumberoffunctionalconnections(Hillary&Grafman,2017).For
example,indepression,thedorsomedialprefrontalcortexexhibitsenhancedconnectivity
withdefaultmode,cognitivecontrol,andaffectivenetworks(Sheline,Price,Yan,&Mintun,
2010).Asdemonstratedbelow,whenFCdiffersinselectedregionsbetweengroups,PTis
vulnerabletoidentifyingspuriousdifferencesinnodalstatistics(e.g.,degree).The
concernsexpressedhereextendfromvandenHeuvelandcolleagues(2017),who
demonstratedthatPTincreasesthelikelihoodofincludingspuriousconnectionsinthe
networkwhengroupsdifferinmeanFC.
Table4.Thresholdingmethodfordefiningedgesingraphs
Natureofthresholding n(frequency)Allconnectionsretained 6(5.7%)Singlebyvalue 25(23.6%)Multiplebyvalue 28(26.4%)Singlebydensity 5(4.7%)Multiplebydensity 31(29.2%)Bothbyvalueanddensity 7(6.6%)Other(connectionslost) 3(2.8%)Unknown/unclear 1(0.9%)
Note.Value:thresholdingdeterminedbyFCstrength,includingstatisticallycorrectedanduncorrectedvalues;Single:analysesreportedatasinglethresholdvalue;Multiple:networkexaminedacrossmultiplethresholds.
SimulationtodemonstrateaproblemwithPTingroupcomparisons:Whack-
a-node.Inthefirstsimulation—‘whack-a-node’—weexaminedtheconsequencesofPT
onregionalinferenceswhengroupsdifferinFCforselectedregions.Unlikeempirical
Graphtheoryasatooltounderstandbraindisorders
18
resting-statefMRIdata,wheretheunderlyingcausalprocessesremainrelativelyunknown,
simulationsallowonetotesttheeffectofbiologicallyplausiblealterations(e.g.,
hyperconnectivityofcertainnodes)onnetworkanalyses.Simulationscanalsoclarifythe
effectsofalternativeanalyticdecisionsonsubstantiveconclusionsinempiricalstudies.For
thedetailsofoursimulationapproach,seeMethods.Briefly,weusedanetwork
bootstrappingapproachtosimulateresting-statenetworkdataforacase-controlstudy
with50patientsand50controls.Werepeatedthissimulation100times,increasing
connectivityforpatientsinthreerandomlytargetednodes(hereaftercalled‘Positive’)and
decreasingconnectivityslightly,butnonsignficantly,inthreerandomnodes(‘Negative’).
Oursimulationscapturedbothbetween-andwithin-personvariationinFCchangesfor
targetednodes(seeMethods,Equations5-7).WecomparedchangesinPositiveand
NegativenodestothreeComparatornodesthatdidnotdifferbetweengroups.
Resultsofwhack-a-nodesimulation.
Proportionalthresholding.Inamultilevelregressionofgroupdifferencet-statistics
(patient–control)ondensitythreshold,wefoundthatPTwassensitiveto
hyperconnectivityofPositivenodes,reliablydetectinggroupdifferences,averaget=12.4
(SD=1.15),averagep<.0001(Figure2a).Importantly,however,degreewassignificantly
lowerinpatientsthancontrolsforNegativenodes,averaget=-6.52(SD=1.15),averagep
<.001.WedidnotfindanysystematicdifferencebetweengroupsinComparatornodes,
averaget=-.22(SD=.65),averagep=.47.
Thesegroupdifferenceswerequalifiedbyasignificantdensityxnodetype
interaction(p<.0001)suchthatgroupdifferencesforPositivenodeswerelargerathigher
densities(Figure3,toppanel),B=11.32(95%CI=10.59–12.04),t=30.72,p<.0001.
Graphtheoryasatooltounderstandbraindisorders
19
Conversely,wefoundequal,butopposite,shiftsinNegativenodes(Figure3,middlepanel):
groupdifferencesbecameincreasinglynegativeathigherdensities,B=-11.93,(95%CI=-
12.65–-11.2),t=-32.74,p<.0001.However,wedidnotobserveanassociationbetween
densityandgroupdifferencesinComparatornodes,B=.31,p=.40(Figure3,bottom
panel).
Graphtheoryasatooltounderstandbraindisorders
20
Figure2.Theeffectofthresholdingmethodongroupdifferencesindegreecentralitywhenthereisstronghyperconnectivityinthreerandomlyselectednodes(Positive)andweakhypoconnectivityinthreenodes(Negative).Threeunaffectednodesaredepictedforcomparison(Comparator).Forparametersofthissimulation,seeTableS1,Whack-a-nodehyperconnectivity.Thecentralbarofeachrectangledenotesthemediantstatistic(patient–control)across100replicationdatasets(patientn=50,controln=50),whereastheupperandlowerboundariesdenotethe90thand10thpercentiles,respectively.Thedarklineatt=0reflectsnomeandifferencebetweengroups,whereasthelightgraylinesatt=-1.99and1.99reflectgroupdifferencesthatwouldbesignificantatp=.05.a)Graphsbinarizedat5%,15%,and25%density.b)Graphsbinarizedatr={.2,.3,.4}.c)Graphsbinarizedatr={.2,.3,.4},withdensityincludedasabetween-subjectscovariate.d)Strengthcentrality(weightedgraphs).
−505
1015
0.05 0.15 0.25Density
Mea
n t
Density thresholda)
−505
1015
0.4 0.3 0.2Pearson r
FC Thresholdb)
−505
1015
0.4 0.3 0.2Pearson r
Mea
n t
FC threshold, covary densityc)
−505
1015
Node TypePositiveNegativeComparator
Weightedd)
Graphtheoryasatooltounderstandbraindisorders
21
Figure3.Theeffectofdensitythresholdongroupdifferencesindegreecentrality.Dotsdenotethemeantstatisticatagivendensity,whereasverticallinesdenotethe95%confidenceinterval.Allstatisticsreflectgroupdifferencesindegreecentralitycomputedongraphsbinarizedatdifferentdensities.
Functionalconnectivity(FC)thresholding.Ingraphsthresholdedatdifferinglevelsof
FC(rsrangingbetween.2and.5),wefoundreliableincreasesinPositivenodesinpatients,
averaget=6.37,averagep<.0001(Figure2b).Negativenodes,however,werenot
significantlydifferentbetweengroups,averaget=-1.75,averagep=.23.Neitherdid
Comparatornodesdifferbygroup,averaget=.05,p=.50.UnlikePT,forFC-thresholded
graphs,wedidnotfindasignificantthresholdxnodetypeinteraction,!2(16)=13.38,p=
.65.
●
●
●
●
●●
●●
●
●
●
●
●
●●
●●
●
●
●
●●
● ●● ● ●
Comparator
Negative
Positive
0.05 0.075 0.1 0.125 0.15 0.175 0.2 0.225 0.25
11
12
13
−7
−6
−5
−0.4
−0.2
Density threshold
Aver
age
grou
p di
ffere
nce t
Graphtheoryasatooltounderstandbraindisorders
22
Functionalconnectivitythresholdingwithdensityascovariate.Bydefinition,FC
thresholdingcannothandletheproblemofgroupdifferencesinmeanFC.Thus,theriskof
FCthresholdingaloneisthatnodalstatisticsmayreflectgroupdifferencesinmeanFCthat
affectinterpretationoftopologicalmetrics(e.g.,smallworldness).Tomitigatethisconcern,
onecouldthresholdatatargetFCvalue,thenincludegraphdensityforeachsubjectasa
covariate(assuggestedbyvandenHeuveletal.,2017).AsdepictedinFigure2c,however,
althoughthisstatisticallycontrolsfordensity,italsoreintroducestheconstraintthatthe
groupsmustbeequalinaveragedegree.Asaresult,thepatternofeffectsmirrorsthePT
graphs(Figures2a,3).Morespecifically,groupdifferencesinPositivenodeswerereliably
detected,averaget=12.39,averagep<.0001.Butpatientsappearedtobesignificantly
morehypoconnectedinNegativenodescomparedtocontrols,averaget=-6.59,averagep
<.001.
Weightedanalysis.Inanalysesofstrengthcentrality,weobservedsignificantly
greaterdegreeinPositivenodes,averaget=5.5,averagep=.0003(Figure2d).Asexpected
givenoursimulationdesign,NegativeandComparatornodeswerenotsignificantly
differentbetweengroups,averageps=.19and.49,respectively.
Discussionofwhack-a-nodesimulation.Inour‘whackanode’simulation,three
nodeswererobustlyhyperconnectedin‘patients’,whilethreenodeswereweakly
hypoconnected.Theprincipalfindingofthissimulationwasthatenforcingequalaverage
degreeusingPTcanspuriouslymagnifychangesingroupcomparisonsofnodalstatistics.
Whenthegroupswereotherwiseequal,hyperconnectivityinselectednodeswas
accuratelydetectedusingPTacrossdifferentgraphdensities.However,nodesthatwere
weaklyhypoconnectedtendedtobeidentifiedasstatisticallysignificant.Theinclusionof
Graphtheoryasatooltounderstandbraindisorders
23
lowmagnitude,spuriousconnectionsintothenetworkinheresfromthewayinwhichPT
handlesthetailsoftheFCdistribution.
ByretainingonlyedgesatthehighendoftheFCdistribution,edgesthatareonthe
cuspofthatcriterionaremostvulnerabletobeingremoved.Forexample,atadensityof
25%,smallvariationinFCstrengthnearthe75thpercentilecouldleadtoinclusionor
omissionofanedge.Asaresult,ifFCforedgesincidenttoagivennodetendstobeweaker
inonegroupthantheother,thenbinarygraphsgeneratedusingPTwillmagnifythe
statisticalsignificanceofdifferencesindegreecentrality.Totheextentthatnodal
differencesinFCstrengthrepresentashiftinthecentraltendencyofthedistribution,this
problemisnotsolvedbyapplyingmultipledensitythresholds(Figure3).Weobservedthe
sameproblemifthedirectionofFCchangeswasflippedinthesimulation:underPT,group
comparisonsweresignificantforweaklyhyperconnectednodesifsomenodeswere
robustlyhypoconnected(seeFigureS1).Moreover,ourfindingswerequalitatively
unchangedwhensimulatedchangesinFCwereappliedproportionatetotheoriginal
connectionstrength,ratherthanshiftingconnectivityincorrelationalunits(seeFigureS2).
ThisadditionalsimulationrespectedbiologicallyplausiblevariationinFC(e.g.,greater
averageconnectivityinfunctionalhubs).
WeexaminedFC-basedthresholding(here,usingPearsonrasthemetric)and
weightedanalysesascomparisonstoPT.Thesemethodsdonotsufferfromthespurious
detectionofnodaldifferencesevidentunderPT.Rather,FCthresholdingaccurately
detectedhyperconnectednodesacrossdifferentthresholdswhilenotmagnifying
significanceoftheweaklyhypoconnectednodes.However,asnotedelsewhere(vanden
Heuveletal.,2017;vanWijketal.,2010),iftwogroupsdifferinmeanFC,thresholdingata
Graphtheoryasatooltounderstandbraindisorders
24
givenlevel(e.g.,r=.3)inbothgroupswillleadtodifferencesingraphdensity.Thiscould
manifestaswidespreaddifferencesinnodalstatisticsduetoglobaldifferencesinthe
numberofedges.Weconsideredwhetherincludingper-subjectgraphdensityasa
covariateingroupdifferenceanalysescouldretainthedesirableaspectsofFCthresholding
whileaccountingforthepossibilityofglobaldifferencesinFC.Wefound,however,that
statisticallycovaryingfordensitywasqualitativelysimilarinitseffectstoPTbecauseit
constrainsthesumofdegreechangesacrossthenetworktobezerobetweengroups(i.e.,
equalaveragedegree).
ThegoalofoursimulationwastoprovideaproofofconceptthatPTmaynegatively
affectnodalstatisticsincase-controlgraphstudiesbyenforcingequalaveragedegree.We
didnot,however,testarangeofparameterstoidentifytheconditionsunderwhichthis
concernholdstrue.Thesimulationfocusedspecificallyondegreecentralityinthebinary
caseandstrengthintheweightedcase.Whileuntested,weanticipatethattheseeffects
likelygeneralizetoothernodalmeasuressuchaseigenvectorcentrality.Importantly,the
problemswithPThighlightedaboveoccurregardlessofedgedensity(Figure3),sotheuse
ofmultipleedgedensitiesdoesnotadequatelyaddressthe“whack-a-node”issue.
CriticalIssue2:Matchingtheorytoscale:Telescopinglevelsofanalysisingraphs
Thesecondmajormethodologicalthemefromourliteraturereviewconcernsthe
alignmentbetweenneurobiologicalhypothesesandgraphanalyses.Werefertothisas
theory-to-scalematching.RSFCgraphsoffertelescopinglevelsofinformationabout
intrinsicconnectivitypatterns,fromglobalinformationsuchasaveragepathlengthto
detailssuchasconnectivitydifferencesinaspecificedge.Forexample,inmajordepression,
resting-statestudieshavefocusedanalysesonspecificsubnetworkscomposedofthe
Graphtheoryasatooltounderstandbraindisorders
25
dorsalmedialprefrontalcortex,anteriorcingulatecortex(ACC),amygdala,andmedial
thalamus(forareview,seeL.Wang,Hermens,Hickie,&Lagopoulos,2012).
Weproposethatgraphanalysesshouldbeconceptualizedandreportedintermsof
telescopinglevelsofanalysisfromglobaltospecific:globaltopology,modularstructure,
nodaleffects,edgeeffects.Ourreviewoftheclinicalnetworkneuroscienceliterature
revealedthatthemajorityofstudies(69%;seeTable5)testedwhethergroupsdifferedon
globalmetricssuchassmall-worldness.Importantly,however,manystudiesprovided
limitedtheoreticaljustificationforwhythepathophysiologyofagivenbraindisorder
shouldaltertheglobaltopologyofthenetwork.Inthefollowing,weprovidemorespecific
commentsonpathlengthandsmall-worldness,andhowinvestigatorsmightideallymatch
hypothesestolevelsofanalysis.
Table5.Levelofgroupdifferencehypothesesingraphanalyses(i.e.,telescoping)
Hypothesislevel n(frequency)Global 73(68.9%)Module(subnetwork) 41(38.7%)Nodal(Region) 52(49.1%)Edge 1(0.9%)Note.Global:examiningwhole-graphnetworkfeatures(e.g.,small-worldness),Module:examiningsubnetworks/modules(e.g.,defaultmodenetwork).Totalfrequencyisgreaterthan100%becausesomestudiestestedhypothesesatmultiplelevels.
Theclinicalmeaningfulnessofsmall-worldness.Inadefiningstudyfornetwork
neuroscience,WattsandStrogatz(1998)demonstratedthattheorganizationofthecentral
nervoussysteminC.elegansreflecteda“small-world”topology(cf.Muldoon,Bridgeford,&
Bassett,2016).Theimpactofthisfindingcontinuestoresonateinthenetwork
neuroscienceliterature20yearslater(Hilgetag&Kaiser,2004;Sporns&Zwi,2004),with
Graphtheoryasatooltounderstandbraindisorders
26
manystudiesfocusingon“disconnection”andthelossofnetworkefficiencyasquantified
bysmall-worldness(31%ofthestudiesreviewedhere).Althoughsmall-worldtopologies
havebeenobservedinmoststudiesofbrainfunction(Bassett&Bullmore,2016),the
relevanceofthisorganizationforfacilitatinghumaninformationprocessingremains
unclear.Otherfeaturesofhumanneuralnetworks,suchasmodularity,mayhavemore
importantimplicationsfornetworkfunctioning(Hilgetag&Goulas,2015).Highernetwork
modularityreflectsagraphinwhichtheconnectionsamongnodestendtoformmore
denselyconnectedcommunities,whichtendtoberobusttorandomnetworkdisruption
(Shekhtman,Shai,&Havlin,2015).
Itremainsuncertainthat,asageneralrule,brainpathologyshouldbereflectedina
measureofsmall-worldness.Forexample,asmall-worldtopologyispreservedevenin
experimentsthatdramaticallyreducesensoryprocessingviaanesthesiainprimates
(Vincentetal.,2007)andindisordersofconsciousnessinhumans(Croneetal.,2014).
Recognizingthatconventionalmeasuresofsmall-worldness(e.g.,Humphries&Gurney,
2008)dependondensityanddonothandlevariationinconnectionstrength,recent
researchhasrecastthisconceptanditsquantificationintermsofthe“smallworld
propensity”ofanetwork(Muldoonetal.,2016).
Althoughwedonotdisputethevalueofglobalgraphmetricssuchassmall-
worldnessasimportantdescriptorsofanetworkorganization,bydefinition,theyprovide
informationatonlythemostmacroscopiclevel.Groupdifferencesinglobalmetricsmay
largelyreflectmorespecificeffectsatfinerlevelsofthegraph.Forexample,removing
connectionsinfunctionalhubregionsselectivelytendstoreduceglobalefficiencyand
clustering(Hwang,Hallquist,&Luna,2013).Likewise,failingtoidentifygroupdifferences
Graphtheoryasatooltounderstandbraindisorders
27
inglobalstructuredoesnotimplyequivalenceatotherlevelsofthegraph(e.g.,nodesor
modules).Todemonstratethepointthatsubstantialgroupdifferencesinfinerlevelsofthe
graphmaynotbeevidentinglobalmetrics,weconductedasimulationinwhichthegroups
differedsubstantiallyinmodularconnectivity.
Simulationtodemonstratetheimportanceofunderstandinggraphsat
multiplelevels:globalinsensitivitytomodulareffects.Extendingthebasicapproachof
ourwhack-a-nodesimulation,weuseda13-moduleparcellationofthe264-node
groundtruthgraphinasimulationof50‘controls’and50‘patients’(modularstructure
fromPoweretal.,2011).Morespecifically,weincreasedFCinthefronto-parietalnetwork
(FPN)anddorsalattentionnetwork(DAN)incontrols,andincreasedFCinthedefault
modenetwork(DMN)inpatients.Thesimulationprimarilyexaminedgroupdifferencesin
small-worldness(globalmetric)andwithin-andbetween-moduledegree(nodalmetrics).
AdditionaldetailsareprovidedintheMethods.
Resultsofglobalinsensitivitysimulation.Consistentwithcommonmethodsinthe
field,weappliedproportionalthresholding(PT)between7.5%and25%toeachgraph.We
computedsmall-worldness,",ateachthreshold(seeMethods),thentestedforgroup
differencesinthiscoefficient.AsdepictedinFigure4,wedidnotobserveanysignificant
groupdifferencesinsmall-worldnessatanyedgedensity,Mt=.36,ps>.3.
Graphtheoryasatooltounderstandbraindisorders
28
Figure4.Groupdifferencesinsmallworldness(")asafunctionofedgedensity.Thecentralbarofeachrectangledenotesthemedian"statistic(patient–control),whereastheupperandlowerboundariesdenotethe90thand10thpercentiles,respectively.
However,consistentwiththestructureofoursimulations,wefoundlargegroup
differencesinwithin-andbetween-moduledegree(Figure5).Inamultilevelregressionof
within-networkdegreeongroup,density,andmodule,wefoundasignificantDMNincrease
inpatientsirrespectiveofdensity,B=1.37,t=33.25,p<.0001.Likewise,controlshad
significantlyhigherwithin-networkdegreeintheFPNandDAN,ts=21.67and13.52,
respectively,ps<.0001.Thesefindingsweremirroredingroupanalysesofbetween-
networkdegreeintheDMN,FPN,andDAN,ps<.0001(seeFigure5).
1.08
1.12
1.16
1.20
1.24
10 15 20 25Density (%)
Smal
l wor
ldne
ss (σ
)
Groupcontrolspatients
Graphtheoryasatooltounderstandbraindisorders
29
Figure5.Groupdifferencesinz-scoreddegreestatisticsat10and20%density.Degreedifferencesforconnectionsbetweenmodulesaredepictedinthetoprow,whereaswithin-moduledifferencesareinthebottomrow.Dotsdenotethemeanzstatisticacrossnodes,whereasthelinesrepresentthe95%confidenceintervalaroundthemean.NonsignificantgroupdifferencesintheVisualnetworkaredepictedforcomparison,whereasconnectivityintheDAN,FPN,andDMNwasfocallymanipulatedinthesimulation.DAN=DorsalAttentionNetwork;FPN=Fronto-ParietalNetwork;DMN=DefaultModeNetwork.
Discussionofglobalinsensitivitysimulation.Intheglobalinsensitivitysimulation,
weinducedlargegroupchangesinFCattheleveloffunctionalmodulesthatrepresent
canonicalresting-statenetworks(e.g.,theDMN).Thesimulationdifferentiallymodulated
FCwithinandbetweenregionsoftheDAN,FPN,andDMN.Ingroupanalysesofwithin-and
between-moduledegreecentrality,wedetectedtheselargeshiftsinFC.However,despite
robustdifferencesinnetworkstructure,thetwogroupswereverysimilarinthesmall-
worldpropertiesoftheirgraphs.
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
10% Density 20% Density
Between−Module
Within−M
odule
−0.5 0.0 0.5 −0.5 0.0 0.5
DMN
FPN
DAN
Visual
DMN
FPN
DAN
Visual
Degree (z)
Mod
ule Group
●
●
controlspatients
Graphtheoryasatooltounderstandbraindisorders
30
Aswiththewhack-a-nodesimulation,wedidnotseektotesttherangeofconditions
underwhichthesefindingswouldhold.Rather,theglobalinsensitivitysimulationprovides
aproofofconceptthatresearchersshouldbeawarethattheabsenceofgroupdifferences
atahigherlevelofthegraph(here,globaltopology)doesnotsuggestthatthenetworksare
otherwisesimilaratlowerlevels(here,modularconnectivity).Aswehavenotedabove,in
graphanalysesofcase-controlresting-statenetworks,weencourageresearcherstostate
theirstudygoalsintermsthatclearlymatchthehypothesestothescaleofthegraph.
Intheglobalinsensitivitysimulation,failingtodetectdifferencesinsmall-worldness
shouldnotbeseenasanomnibustestofmodularornodalstructure.Likewise,ifgroup
differencesaredetectedatthegloballevel,theremaybesubstantialvalueininterrogating
finerdifferencesinthenetworks,evenifthesewerenothypothesizedapriori.
GeneralSummary
Theoverarchinggoalofthisreviewwastopromotesharedstandardsforreporting
findingsinclinicalnetworkneuroscience.Oursurveyoftheresting-statefunctional
connectivityliteraturerevealedthepopularityandpromiseofgraphtheoryapproachesto
networkorganizationinbraindisorders.Thispotentialisevidentinlarge-scaleinitiatives
foracquiringandsharingresting-statedataindifferentpopulations(e.g.,theHuman
ConnectomeProject;Barchetal.,2013).Publiclyavailableresting-statedatacanalso
supportreproducibilityeffortsbyservingasreplicationdatasetstocorroboratespecific
findingsinanindependentsample(e.g.,Jalbrzikowskietal.,2017).Wesharethefield’s
enthusiasmforsuchworkandanticipatethatwithfurthermethodologicalrefinementand
standardization,thecouplingofnetworkscienceandbrainimagingcanprovidenovel
insightsintotheneurobiologicalbasisofbraindisorders.
Graphtheoryasatooltounderstandbraindisorders
31
Ourreview,however,suggestedthattheheterogeneityofmethodsispreventingthe
fieldfromrealizingitspotential.Graphanalysesacrossclinicalstudiesvariedsubstantially
intermsofbrainparcellation,FCquantification,andtheuseofthresholdingmethodsto
defineedgesingraphs.Thesedecisionsarefundamentaltographtheoryandprecede
analysesatspecificlevelssuchasglobaltopology.Inaddition,althoughitwasnotafocusof
ourreview,therewassubstantialvariationinwhatnetworkmetricswerereportedacross
studies.Thelackofstandardizationinmethodsatmultipledecisionpointshas
multiplicativeconsequences:thelikelihoodthatanytwostudiesusedthesameparcellation
scheme,FCdefinition,thresholdingstrategy,andnetworkmetricwasremarkablylow.
Thismakesformalmeta-analysesoftheclinicalnetworkneuroscienceliteraturevirtually
impossibleatthepresenttime,detractingfromeffortstodistinguishdistinct
pathophysiologicalmechanismsortoidentifytransdiagnosticcommonalities.Inaddition,
methodologicalheterogeneityingraphanalysesundercutsthevalueofdatasharingefforts
thathavemademassivedatasetsavailabletothenetworkneurosciencecommunity.For
theimmensepotentialofdatasharingtoberealized,standardizationmustoccurnotonly
indataacquisition,butalsoindataanalysis,withasharedframeworktoguidehypothesis-
to-scalematchingingraphs.
AssummarizedinBox2,webelievethatthefieldshouldworktowardaprincipled,
commonapproachtographanalysesofRSFCdata.Thisisachallengingproposition
becauseoftherapidandexcitingdevelopmentsinfunctionalbrainparcellation(e.g.,
Schaeferetal.,inpress),FCdefinition(e.g.,Cassidy,Rae,&Solo,2015),edgethresholding
(vandenHeuveletal.,2017),andnetworkmetrics(e.g.,Vargas&Wahl,2014).Such
developmentshighlightboththeenthusiasmfor,andrelativeinfancyof,network
Graphtheoryasatooltounderstandbraindisorders
32
neuroscienceasafield.Althoughwearesensitivetotheimportanceofcontinuingtorefine
functionalparcellationsofthebrain,wealsoseegreatvalueindevelopingfield-standard
parcellationstopromotecomparability.Indeed,manyaspectsofnetworkstructure(e.g.,
homogeneityoffunctionalconnectivitypatternswithinaregion)arelargelyconvergent
aboveacertainlevelofdetail(likely200-400nodes)inthefunctionalparcellation
(Craddocketal.,2012;Schaeferetal.,inpress).Likewise,theoptimalapproachfor
quantifyingfunctionalconnectivityisanopenquestion(Smithetal.,2011),yetinthe
absenceofmethodologicalconvergence,graphswereoftennotcomparableacrossstudies.
Arelateddilemmawasthatin57%ofstudies,littleornodetailwasprovidedabout
hownegativeFCvalueswereincorporatedintographanalyses.Thiswasespecially
troublinginsofarasglobalsignalregressiontendstoyieldanFCdistributioninwhich
approximatelyhalfofedgesarenegative(Murphyetal.,2009).Furthermore,FCestimates
atthelowendofthedistributionmayhavedifferenttopologicalpropertiessuchasreduced
modularity(Schwarz&McGonigle,2011).Evenamongthe21%ofstudiesinwhich
negativeedgeswereexplicitlydropped,itremainsunclearwhatconsequencesthis
decisionhasonsubstantiveconclusionsaboutgraphstructure.Inrecentyears,therehave
beenadvancesinquantifyingcommongraphmetricssuchasmodularityinweighted
networksthatincludenegativeedges(Rubinov&Sporns,2011),aswellasincreasingcalls
forweighted,notbinary,graphanalyses(Bassett&Bullmore,2016).Regardless,we
believethatgreaterclarityinreportingofnegativeFCwillpromotecomparisonsamong
clinicalstudies.
Methodologicalheterogeneityalsoresultedinveryfewgraphstatisticsthatwere
reportedincommonacrossstudies,anessentialingredientforexaminingthe
Graphtheoryasatooltounderstandbraindisorders
33
reproducibilityoffindings.Networksthatareresilienttothedeletionofspecificedges
oftenhaveahighlyskeweddegreedistribution(Callaway,Newman,Strogatz,&Watts,
2000)thatmayrelatetosmall-worldnetworkproperties(Achardetal.,2006).Wepropose
thatstudiesshouldroutinelydepictthisdistribution.Likewise,broadmetricssuchasedge
density,meanFC,transitivity,andcharacteristicpathlengthprovideimportant
informationaboutthebasicpropertiesofgraphsthatcontextualizemoredetailed
inferentialanalyses.Thechallengeofdevelopingreportingstandardsinclinicalnetwork
neuroscienceechoesthebroaderconversationinneuroimagingaboutreproducibility,
especiallytheimportanceofdetailandtransparencyintheanalyticapproach(Nicholset
al.,2017).
Inadditiontothegeneralissuesofstandardizinggraphanalysesandreporting
procedures,ourreviewexaminedtwocriticalissuesingreaterdetail.First,weconsidered
thepotentialbenefitsandrisksofusingproportionalthresholding(PT),acommon
procedureforequatingthenumberofedgesbetweengraphs.Second,wearticulatedthe
valueofconsideringthetelescopinglevelsofgraphstructuresinordertomatcha
hypothesistothecorrespondingscaleofthegraph.
RoughlyonethirdofthestudiesincludedinourreviewappliedPT,thresholding
graphsatsingleormultipleedgedensities.Althoughthisapproachisalignedwithprior
workhighlightingconcernsaboutcomparingunequalnetworks(vanWijketal.,2010),its
applicationinclinicalstudiesisoftenconceptuallyproblematic.Manybrainpathologies
appeartoaffecttargetedregionsornetworks,whileleavingtheconnectivityofother
regionslargelyundisturbed.Forexample,althoughfrontolimbiccircuitryisheavily
implicatedinmooddisorders(Price&Drevets,2010),visualnetworkslargelyappear
Graphtheoryasatooltounderstandbraindisorders
34
unaffected.Thereisgrowingevidencethatbraindisordersalterthestrengthoffunctional
couplingandpotentiallythenumberoffunctionalconnections(Hillary&Grafman,2017).
Consequently,ifsomeregionsareaffectedbythepathology,butothersaresimilartoa
matchedcontrolpopulation,PTmayerroneouslyremoveoraddconnectionstographsin
onegroupinordertomaintainequalaveragedegreebetweengroups.Furthermore,ifa
brainpathologyaltersthedensityoffunctionalconnections—forexample,neurological
disruptionisassociatedwithhyperconnectivity(Hillaryetal.,2015)—PTwillpreclude
theinvestigatordetectingdensitydifferencesbetweengroups.If,intruth,thegroupsdiffer
inedgedensity,artificiallyequatingdensityalsodetractsfromtheinterpretabilityofgraph
analyses(cf.vandenHeuveletal.,2017).
Inadditiontotheseconceptualproblems,ourwhack-a-nodesimulation
demonstratedthatPTmayresultinthedetectionofspuriousgroupdifferences(Figure2a).
Altogether,applyingPTinclinicalstudiesmaybeamethodologicaldoublejeopardy,
characterizedbyreducedsensitivitytopathology-relateddifferencesinconnectiondensity
andtheriskofidentifyingnodaldifferencesbetweengroupsthatareastatisticalartifact.
TheserisksmakePTespeciallyunappealingwhenoneconsidersthatFC-based
thresholdingandweightedanalysesaccuratelydetectedgroupdifferences(Figure2b,d)
whilealsoallowingedgedensitytovary.Weacknowledge,however,thataprevious
empiricalstudyfoundthatgroupdifferencesweremoreconsistentacrossmultiple
thresholdsusingproportional,asopposedtoFC-based,thresholding(Garrison,Scheinost,
Finn,Shen,&Constable,2015).
Wethereforehavetworecommendationsforedgethresholdingincase-control
comparisons.First,weightedanalysesorFCthresholdingshouldtypicallybepreferredto
Graphtheoryasatooltounderstandbraindisorders
35
PTifoneisinterestedinnodalstatistics.Second,toruleoutthepossibilitythatnodal
findingsreflectglobaldifferencesinmeanFC,onecouldincludemeanFCasacovariatein
weightedanalysesorper-subjectdensityinanalysesofFC-thresholdedbinarygraphs.
Crucially,weproposethatthesebetreatedassensitivityanalysesconductedonlyafter
establishinganodalgroupdifference.Thatis,ifoneidentifiesgroupdifferencesinFC-
thresholdedgraphs(e.g.,greaterdegreeinanteriorcingulatecortexamongpatients),does
includingedgedensityasacovariateabolishthisfinding?Ifso,itsuggeststhatdifferences
inglobaltopologymayaccountforthenodalfinding.However,oneshouldnotinclude
densityasacovariateinFC-thresholdedgraphsasafirststeptoidentifywhichnodesdiffer
betweengroups,asthiscouldfallpreytothe‘whackanode’problem(i.e.,spuriousnodal
effects).
Thesecondcriticalissuewasthatmanystudiesprovidedlimitedtheoretical
justificationforthealignmentbetweenagivenhypothesisandthecorrespondinggraph
analysis.Amajorityofstudies(69%)testedwhethergroupsdifferedinglobalmetricssuch
assmall-worldness,butmostpathologies(e.g.,braininjury,Alzheimer’sdisease)primarily
affectregionalhubswithinnetworks(Crossleyetal.,2014).Ourglobalinsensitivity
simulationfocusedontheimportanceofmatchinghypothesestographanalyses,or
telescoping.Wedemonstratedthatglobalgraphmetrics,specificallysmall-worldness,may
notbesensitivetogroupdifferencesinmoduleornodecentrality.Metricssuchas
modularityandnodalcentralityoffervitalinformationaboutregionalbrainorganization
thatcanbeinterpretedinthecontextofalterationsinaveragedegree.Thegeneralpointis
thatonecannotgeneralizefindingsfromonelevelofagraphtoanother,norshouldnull
effectsatonelevelbeviewedassuggestingthatthegroupsaresimilaratotherlevels.By
Graphtheoryasatooltounderstandbraindisorders
36
conceptualizingadataanalysisplanintermsofthetelescopingstructureofgraphs,
researcherscanclearlydelineateconfirmatoryfromexploratoryanalyses,whichis
consistentwiththespiritofreproduciblescienceinneuroimaging(Poldracketal.,2017).
Box2:RecommendationsforBestPracticesandConvergenceinClinicalNetwork
Neuroscience
Challenge Recommendation
Brain parcellations vary substantially across studies
In the absence of a field standard, we recommend that brain parcellations should meet the following minimum requirements:
1) Provide comprehensive coverage of functional regions throughout the brain, including cortical and subcortical structures, as well as cerebellum
2) Divide the brain into at least 200 functional regions (Craddock et al., 2012)
3) Delineate regions based on functional connectivity (as opposed to a structural atlas; Gordon et al., 2016), potentially combined with multi-modal imaging (e.g., Glasser et al., 2016)
4) Ensure that regions exhibit high functional connectivity homogeneity (Gordon et al., 2016; Schaefer et al., in press)
5) Provide clear guidance on the modular structure of regions comprising the parcellation (Yeo et al., 2011). In general, we discourage clinical researchers from attempting to identify functional modules in their data (e.g., using community detection algorithms) when a published modular structure exists.
The quality of RSFC data varies as a function of acquisition length and time
1) Longer resting-state acquisitions help to improve the stability of FC estimates and, therefore, their test-retest reliability (Birn et al., 2013). We encourage clinical researchers to acquire resting-state data for at least nine minutes and to use faster temporal sampling (e.g., 1s TR, potentially using multi-band imaging; Demetriou et al., 2016) when possible.
Edge definition Quantifying functional connectivity: The reliability of functional connectomes based on conditional association (e.g., partial correlation) diminishes as the number nodes increases or the number of measurements decreases (Cassidy et al., 2018), requiring substantially longer scans, both in terms of time and measurements. The stability of bivariate correlations, however, is not dependent on the number of nodes and values typically stabilize around 250 measurements (Schönbrodt & Perugini, 2013). For conventional resting-state data (e.g., 180 volumes with TR = 2s) and comprehensive parcellations, we recommend a shrinkage estimator of marginal correlation (Varoquaux & Craddock, 2013), rather than a conditional association measure.
2)
Graphtheoryasatooltounderstandbraindisorders
37
3) Proportional thresholding: We caution against using proportional thresholding to create binary graphs in case-control studies (for more detailed recommendations, see Simulation 1). Negative edge weights: Studies should provide details about how negative FC estimates are handled when constructing graphs, whether weighted or binary. The deletion of negative edges is a largely untested assumption in network neuroscience that deserves greater attention. If researchers have many negative edges (e.g., if global signal regression or partial correlation are used), we encourage analyzing and reporting them, perhaps using a separate graph.
Graph metrics varied across studies
4) To promote formal comparisons across studies, we encourage researchers to report a standard set of graph metrics, even if these are in the form of descriptive tables or supplementary material. As a minimal set, we propose: 1) global clustering coefficient, 2) average path length, 3) modularity, 4) degree, 5) eigenvector centrality, and 6) summary statistics of edge strength.
Need to align neurobiology and the network representation
We encourage researchers to conceptualize and report graph analyses in terms of telescoping levels of analysis, from global to specific. As demonstrated in Simulation 2, global metrics may be insensitive to regional effects of pathology or neurodevelopment. We also advocate articulating the alignment between the level of the graph and the biological account of neuropathology. The relevance of some graph metrics (e.g., small-worldness) to understanding brain disorders remains unclear.
Insummary,thereisaneedforacommonframeworktoinformgraphtheory
analysesofRSFCdataintheclinicalneurosciences.Anyrecommendationsshouldemerge
organicallyfromascientificcommunitywhoseinvestigatorsvoluntarilyadoptprocedures
thatmaximizesensitivitytohypothesizedeffectsandsimultaneouslypermitgraph
comparisonsamongstudies(cf.theCOBIDASeffortinneuroimagingmorebroadly;Nichols
etal.,2017).Theimplementationofbestpracticesintask-basedfMRI(e.g.,handling
autocorrelation;Woolrich,Ripley,Brady,&Smith,2001)hasbeensupportedbypowerful,
usable,andfreesoftware.Althoughsimilarsoftwareisincreasinglyavailableforresting-
statestudies(e.g.,Chao-Gan&ZangYu-Feng,2010;Whitfield-Gabrieli&Nieto-Castanon,
2012),bestpracticesarestillemerging,andwebelievethatsoftwaredeveloperswillbe
crucialtopromotingastandardizedgraphtheoryanalysispipeline.Weanticipatethata
Graphtheoryasatooltounderstandbraindisorders
38
commonmethodologicalframeworkwillpromotehypothesis-drivenresearch,alignment
betweentheoryandgraphanalysis,reproducibility,datasharing,meta-analyses,and
ultimatelymorerapidprogressofclinicalnetworkneuroscience.
Methods
PubmedSearchSyntax
Neurologicaldisordersquery:
(graphORgraphicalORgraph-theor*ORtopology)AND(brainORfMRIORconnectivity
ORintrinsic)AND(resting-stateORrestingORrest)AND(neurologicalORbraininjuryOR
multiplesclerosisORepilepsyORstrokeORCVAORaneurysmORParkinson'sORMCIor
Alzheimer'sORdementiaORHIVORSCIORspinalcordORautismORADHDOR
intellectualdisabilityORDownsyndromeORTourette)AND"humans"[MeSHTerms]
Mentaldisordersquery:
(graphORgraphicalORgraph-theor*ORtopology)AND(brainORfmriORconnectivityOR
intrinsic)AND(resting-stateORrestingORrest)AND(clinicalORpsychopathologyOR
mentaldisorderORpsychiatricORneuropsychiatricORdepressionORmoodORanxiety
ORaddictionORpsychosisORbipolarORborderlineORautism)AND"humans"[MeSH
Terms]
GeneralApproachtoNetworkSimulations
Toapproximatethestructureoffunctionalbrainnetworks,weidentifiedayoung
adultfemalesubjectwhocompleteda5-minuteresting-statescaninaSiemens3TTrio
Graphtheoryasatooltounderstandbraindisorders
39
Scanner(TR=1.5s,TE=29ms,3.1x3.1x4.0mmvoxels)withessentiallynohead
movement(meanframewisedisplacement[FD]=.08mm;maxFD=.17mm).We
preprocessedthedatausingaconventionalpipeline,including:1)motioncorrection(FSL
mcflirt);2)slicetimingcorrection(FSLslicetimer);3)nonlineardeformationtotheMNI
templateusingtheconcatenationoffunctional→structural(FSLflirt)andstructural→
MNI152(FSLflirt+fnirt)transformations;4)spatialsmoothingwitha6mmFWHMfilter
(FSLsusan);5)andvoxelwiseintensitynormalizationtoameanof100.Afterthesesteps,
wealsosimultaneouslyappliednuisanceregressionandbandpassfiltering,wherethe
regressorsweresixmotionparameters,averageCSF,averageWM,andthederivativesof
these(16totalregressors).Thespectralfilterretainedfluctuationsbetween.009and.08
Hz(AFNI3dBandpass).Wethenestimatedfunctionalconnectivity(FC)of264functional
ROIsfromthePowerandcolleaguesparcellation,whereeachregionwasdefinedbya5mm
radiusspherecenteredonaspecificcoordinate.Theactivityofaregionovertimewas
estimatedbythefirstprincipalcomponentofvoxelsineachROI,andthefunctional
connectivitymatrix(264x264)wasestimatedbythepairwisePearsoncorrelationsof
timeseries.
This264x264adjacencymatrix,W,servedasthegroundtruthforallsimulations,
withspecificwithin-person,between-person,andbetween-groupalterationsapplied
accordingtoamultilevelsimulationofvariationacrossindividuals(detailsofsimulation
parametersprovidedinTableS1).Morespecifically,toapproximatepopulation-level
variabilityinacase-controldesign,wesimulatedresting-stateadjacencymatricesfor50
‘patients’and50‘controls’byintroducingsystematicandunsystematicsourcesof
variabilityforeachsimulatedparticipant.Systematicsourceswereintendedtotest
Graphtheoryasatooltounderstandbraindisorders
40
substantivehypothesesaboutproportionalthresholding(whack-a-nodesimulation)and
insensitivitytoglobalversusmodulardifferences(globalinsensitivitysimulation),whereas
unsystematicsourcesreflectedwithin-andbetween-personvariationinedgestrength.
Inthismodel,thesimulatededgestrengthbetweentwonodes,iandj,foragiven
subject,s,is:
$%&' = )%& + +%&' +-%&' + .%&' (1)
where)%& istheedgestrengthfromthegroundtruthadjacencymatrixW.Globalvariation
inmeanFCisrepresentedby+%&' ,whichreflectscontributionsofbothbetween-and
within-personvariation:
+%&' = /' + 0%&' (2)
where/'representsnormallydistributedbetween-personvariationinmeanFC:
1~3(0, "7) (3)
while"7controlsthelevelofbetween-personvariationinthesample.Theterm0%&'
representswithin-personvariationofthisedgerelativetothepersonmeanFC,/'.Within-
personvariationinFCacrossalledgesisassumedtobenormallydistributed:
8..'~3(0, ":) (4)
with": scalingthedegreeofwithin-personFCvariationacrossalledges.
Node-specificshiftsinFCarerepresentedby-%&' ,whichincludesbothbetween-
personandwithin-nodecomponents.Morespecifically,themodulationofFCbetween
nodesiandjisgivenby:
-%&' = ;%' + <%&' (5)
wherebetween-personvariationinFCfornodeiis:
=%.~3(>?@, "?@) (6)
Graphtheoryasatooltounderstandbraindisorders
41
with>?%and"?% capturingthemeanandstandarddeviationinFCshiftsfornodeiacross
subjects,respectively.EdgewiseFCvariationofanodeiacrossitsneighbors,j,isgivenby:
AB.C~3(0, "D%) (7)
where"D@representsthestandarddeviationofFCshiftsacrossneighborsofi.Whennodesi
andjwerebothmanipulated,theshiftswereappliedsequentiallysuchthatFCfortheedge
betweeniandjwasnotallowedtohavecompoundingchanges.Thatis,weset-%&' = 0for
i>j.
Finally,.%&'representstherandomvariationinFCfortheedgebetweeniandjfor
subjects.Thisvariationwasassumedtobenormallydistributedacrossalledgesfora
subject:
E..'~3(0, "F) (8)
where"Fcontrolsthestandarddeviationofedgenoiseacrosssubjects.
Whack-a-NoteSimulationMethods
Asmentionedabove,proportionalthresholding(PT)isoftenappliedincase-control
studiestoruleoutthepossibilitythatnetworkdifferencesbetweengroupsreflect
differencesinthetotalnumberofedges.Importantly,ingraphsofequalorderN(i.e.,the
samenumberofnodes),proportionalthresholdingequatesboththedensity,D,and
averagedegree,⟨H⟩betweensubjects:
J = KLM(MNO)
, ⟨H⟩ = KLM⇒ ⟨H⟩ = J(3 − 1) (9)
Graphtheoryasatooltounderstandbraindisorders
42
whereEdenotesthenumberofuniqueedgesinanundirectedgraphwithnoself-
loops.AsnotedbyvandenHeuvelandcolleagues(2017),whenoneappliesPT,differences
inaverageconnectivitystrengthcanleadtotheinclusionofweakeredgesinmoresparsely
connectedgroups.Furthermore,weakedgesestimatedbycorrelationaremorelikelyto
reflectanunreliablerelationshipbetweennodes.Thus,ifonegrouphaslowermean
functionalconnectivity,PTcouldintroducespuriousconnections,potentiallyundermining
groupcomparisonsofnetworktopology.
Thatis,whengroupsareotherwiseequivalent,thesumofincreasesindegreein
hyperconnectednodesforagroupmustbeoffsetbyequal,butopposite,decreasesin
degreeforothernodesinthatgroup.Thisphenomenonholdsbecauseofthemathematical
relationshipbetweenaveragedegreeandgraphdensity(Eq.9).Forsimplicity,ourfirst
simulationrepresentsthescenariowheretherearemeaningfulFCincreasesinpatientsfor
selectednodesandunreliableFCdecreasesinothernodes.Thisunreliabilityisintendedto
representsamplingvariabilitythatcouldleadtoerroneousfalsepositives.
Structureofwhack-a-nodesimulation.Asdescribedabove,wesimulated50
patientsand50controlsbasedonagroundtruthFCmatrix.Weestimated100replication
sampleswithequallevelsofnoiseinbothgroups(seeTableS1).Ineachreplication
sample,weincreasedthemeanFCinthreenodes,selectedatrandom,byr=0.14.Wealso
appliedsmallerdecreasesofr=-.04tothreeotherrandomlytargetednodes1.Thislevelof
decreasewaschosensuchthatgroupdifferencesinanalysesofnodalstrength(i.e.,
computedonweightedgraphs)werenonsignificantonaverage(Mp=.19,SDp=.02).We
1ResultsarequalitativelysimilarusingothervaluesforgroupshiftsinFC.
Graphtheoryasatooltounderstandbraindisorders
43
appliedPTtobinarizegraphsineachgroup,varyingdensitybetween5%and25%in1%
increments.Likewise,forFCthresholding,webinarizedgraphsatrthresholdbetweenr=
.2and.5in.02increments.Finally,weretainedweightedgraphsforallsimulatedsamples
toestimategroupdifferencesinnodalstrength.Toensurethateffectswerenotattributable
toparticularnodes,weaveragedgroupstatisticsacrossthe100replicationsamples,where
thetargetednodesvariedrandomlyacrosssamples.
Ineachreplicationsample,weestimateddegreecentralityforthePositive
(hyperconnected),Negative(weaklyhypoconnected),andComparatornodes.Comparators
werethreerandomlyselectednodesineachsamplethatwerenotspecificallymodulated
bythesimulation.Theseservedasabenchmarktoensurethatsimulationsdidnotinduce
groupdifferencesincentralityfornodesnotspecificallytargeted.Toquantifytheeffectof
PTversusFCthresholdinginbinarygraphs,weestimatedgroupdifferences(patient–
control)ondegreecentralityusingtwo-samplet-testsforeachPositive,Negative,and
Comparatornode.Forweightedanalyses,weestimatedgroupdifferencesinstrength
centrality.
GlobalInsensitivitySimulationMethods:Hypothesis-to-scalematching
Wesimulatedadatasetof50‘controls’and50‘patients’inwhichthefronto-parietal
network(FPN)anddorsalattentionnetwork(DAN)weremodulatedincontrolscompared
tothegroundtruthmatrix,W.Inthissimulation,thepatientgrouphadincreasedFCinthe
defaultmodenetwork(DMN),acommonfindinginneurologicaldisorders(Hillaryetal.,
2015).Incontrols,weincreasedFCstrengthforedgesbetweenFPN/DANregionsandother
networks,Mr=0.2,SD=0.1.Per-modulevariationinbetween-networkFCchangeswas
Graphtheoryasatooltounderstandbraindisorders
44
assumedtobenormallydistributedwithineachsubject,SD=0.1.Wealsoincreased
controls’FConedgeswithintheFPNandDAN,Mr=0.1,between-subjectsSD=0.05,
within-subjectsSD=.05.Inpatients,between-networkFCforDMNnodeswasincreased,M
r=0.2,between-subjectsSD=0.1,within-subjectsSD=0.1.Likewise,within-networkFCin
theDMNwasincreased,Mr=0.1,between-subjectsSD=.05,within-subjectsSD=.05.That
is,weappliedsimilarlevelsofFCmodulationtotheFPN/DANincontrolsandtheDMNin
patients,althoughthesechangeslargelyaffecteddifferentedgesinthenetworksbetween
groups.
Wecomputedthesmall-worldnesscoefficient,",accordingtotheapproachof
HumphriesandGurney(2008):
" =ST/SV?WXYT/YV?WX
Here,STrepresentsthetransitivityofthegraph,whereasSV?WX denotesthe
transitivityofarandomgraphwithanequivalentdegreedistribution.Likewise,YTand
YV?WXrepresentthecharacteristicpathlengthofthetargetgraphandrandomlyrewired
graph,respectively.Valuesof"muchlargerthan1.0correspondtoanetworkwithsmall-
worldproperties.Togeneratestatisticsforequivalentrandomgraphs,weapplieda
rewiringalgorithmthatretainedthedegreedistributionofthegraphwhilepermuting
347,160edges(10permutationsperedge,onaverage).Thisalgorithmwasappliedtothe
targetgraph100timestogenerateasetofequivalentrandomnetworks.Transitivityand
characteristicpathlengthwerecalculatedforeachofthese,andtheiraverageswereused
incomputingthesmall-worldnesscoefficient,".Wealsoanalyzedwithin-andbetween-
Graphtheoryasatooltounderstandbraindisorders
45
networkdegreecentralityforeachnode,z-scoringvalueswithineachmoduleanddensity
toallowforcomparisons.
Graphtheoryasatooltounderstandbraindisorders
46
References
Abrol,A.,Damaraju,E.,Miller,R.L.,Stephen,J.M.,Claus,E.D.,Mayer,A.R.,&Calhoun,V.D.
(2017).Replicabilityoftime-varyingconnectivitypatternsinlargerestingstate
fMRIsamples.NeuroImage,163,160–176.
https://doi.org/10.1016/j.neuroimage.2017.09.020
Achard,S.,&Bullmore,E.(2007).Efficiencyandcostofeconomicalbrainfunctional
networks.PLoSComputationalBiology,3(2),0174–0183.
https://doi.org/10.1371/journal.pcbi.0030017
Achard,S.,Salvador,R.,Whitcher,B.,Suckling,J.,&Bullmore,E.(2006).Aresilient,low-
frequency,small-worldhumanbrainfunctionalnetworkwithhighlyconnected
associationcorticalhubs.TheJournalofNeuroscience:TheOfficialJournalofthe
SocietyforNeuroscience,26(1),63–72.https://doi.org/10.1523/JNEUROSCI.3874-
05.2006
Anticevic,A.,Cole,M.W.,Murray,J.D.,Corlett,P.R.,Wang,X.-J.,&Krystal,J.H.(2012).The
roleofdefaultnetworkdeactivationincognitionanddisease.TrendsinCognitive
Sciences,16(12),584–592.
Barch,D.M.,Burgess,G.C.,Harms,M.P.,Petersen,S.E.,Schlaggar,B.L.,Corbetta,M.,…WU-
MinnHCPConsortium.(2013).Functioninthehumanconnectome:task-fMRIand
individualdifferencesinbehavior.NeuroImage,80,169–189.
https://doi.org/10.1016/j.neuroimage.2013.05.033
Bassett,D.S.,&Bullmore,E.T.(2016).Small-worldbrainnetworksrevisited.The
Neuroscientist,1073858416667720.
Graphtheoryasatooltounderstandbraindisorders
47
Bassett,D.S.,Bullmore,E.T.,Meyer-Lindenberg,A.,Apud,J.a,Weinberger,D.R.,&Coppola,
R.(2009).Cognitivefitnessofcost-efficientbrainfunctionalnetworks.Proceedings
oftheNationalAcademyofSciencesoftheUnitedStatesofAmerica,106(28),11747–
11752.https://doi.org/10.1073/pnas.0903641106
Betzel,R.F.,Medaglia,J.D.,Papadopoulos,L.,Baum,G.L.,Gur,R.,Gur,R.,…Bassett,D.S.
(2017).Themodularorganizationofhumananatomicalbrainnetworks:Accounting
forthecostofwiring.NetworkNeuroscience,1(1),42–68.
https://doi.org/10.1162/NETN_a_00002
Birn,R.M.,Molloy,E.K.,Patriat,R.,Parker,T.,Meier,T.B.,Kirk,G.R.,…Prabhakaran,V.
(2013).Theeffectofscanlengthonthereliabilityofresting-statefMRIconnectivity
estimates.NeuroImage,83,550–558.
https://doi.org/10.1016/j.neuroimage.2013.05.099
Biswal,B.B.,Mennes,M.,Zuo,X.-N.,Gohel,S.,Kelly,C.,Smith,S.M.,…Colcombe,S.(2010).
Towarddiscoveryscienceofhumanbrainfunction.ProceedingsoftheNational
AcademyofSciences,107(10),4734–4739.
Biswal,B.B.,Yetkin,F.Z.,Haughton,V.M.,&Hyde,J.S.(1995).Functionalconnectivityin
themotorcortexofrestinghumanbrainusingecho-planarMRI.MagneticResonance
inMedicine,34(4),537–541.https://doi.org/10.1002/mrm.1910340409
Bullmore,E.,&Sporns,O.(2009).Complexbrainnetworks:graphtheoreticalanalysisof
structuralandfunctionalsystems.NatureReviewsNeuroscience,10(3),186–198.
https://doi.org/10.1038/nrn2575
Bullmore,E.,&Sporns,O.(2012).Theeconomyofbrainnetworkorganization.Nature
ReviewsNeuroscience,13(5),336–349.https://doi.org/10.1038/nrn3214
Graphtheoryasatooltounderstandbraindisorders
48
Callaway,D.S.,Newman,M.E.,Strogatz,S.H.,&Watts,D.J.(2000).Networkrobustnessand
fragility:percolationonrandomgraphs.PhysicalReviewLetters,85(25),5468–5471.
https://doi.org/10.1103/PhysRevLett.85.5468
Cassidy,B.,Bowman,D.B.,Rae,C.,&Solo,V.(2018).OntheReliabilityofIndividualBrain
ActivityNetworks.IEEETransactionsonMedicalImaging,37(2),649–662.
https://doi.org/10.1109/TMI.2017.2774364
Cassidy,B.,Rae,C.,&Solo,V.(2015).BrainActivity:Connectivity,Sparsity,andMutual
Information.IEEETransactionsonMedicalImaging,34(4),846–860.
https://doi.org/10.1109/TMI.2014.2358681
Chao-Gan,Y.,&ZangYu-Feng.(2010).DPARSF:aMATLABtoolboxfor“pipeline”data
analysisofresting-statefMRI.FrontiersinSystemsNeuroscience,4.
https://doi.org/10.3389/fnsys.2010.00013
Chen,Y.,Wang,S.,Hilgetag,C.C.,&Zhou,C.(2013).Trade-offbetweenMultipleConstraints
EnablesSimultaneousFormationofModulesandHubsinNeuralSystems.PLOS
ComputationalBiology,9(3),e1002937.
https://doi.org/10.1371/journal.pcbi.1002937
Ciric,R.,Wolf,D.H.,Power,J.D.,Roalf,D.R.,Baum,G.L.,Ruparel,K.,…Satterthwaite,T.D.
(2017).Benchmarkingofparticipant-levelconfoundregressionstrategiesforthe
controlofmotionartifactinstudiesoffunctionalconnectivity.NeuroImage,
154(SupplementC),174–187.https://doi.org/10.1016/j.neuroimage.2017.03.020
Craddock,R.C.,James,G.A.,Holtzheimer,P.E.,Hu,X.P.,&Mayberg,H.S.(2012).Awhole
brainfMRIatlasgeneratedviaspatiallyconstrainedspectralclustering.Human
BrainMapping,33(8),1914–1928.https://doi.org/10.1002/hbm.21333
Graphtheoryasatooltounderstandbraindisorders
49
Crone,J.S.,Soddu,A.,Höller,Y.,Vanhaudenhuyse,A.,Schurz,M.,Bergmann,J.,…
Kronbichler,M.(2014).Alterednetworkpropertiesofthefronto-parietalnetwork
andthethalamusinimpairedconsciousness.NeuroImage.Clinical,4,240–248.
https://doi.org/10.1016/j.nicl.2013.12.005
Crossley,N.A.,Mechelli,A.,Scott,J.,Carletti,F.,Fox,P.T.,McGuire,P.,&Bullmore,E.T.
(2014).Thehubsofthehumanconnectomearegenerallyimplicatedintheanatomy
ofbraindisorders.Brain,137(8),2382–2395.
Demetriou,L.,Kowalczyk,O.S.,Tyson,G.,Bello,T.,Newbould,R.D.,&Wall,M.B.(2016).A
comprehensiveevaluationofmultiband-acceleratedsequencesandtheireffectson
statisticaloutcomemeasuresinfMRI.BioRxiv,076307.
https://doi.org/10.1101/076307
Dosenbach,N.U.F.,Koller,J.M.,Earl,E.A.,Miranda-Dominguez,O.,Klein,R.L.,Van,A.N.,…
Nguyen,A.L.(2017).Real-timemotionanalyticsduringbrainMRIimprovedata
qualityandreducecosts.NeuroImage,161,80–93.
Fallani,F.D.V.,Richiardi,J.,Chavez,M.,&Achard,S.(2014).Graphanalysisoffunctional
brainnetworks:practicalissuesintranslationalneuroscience.Philosophical
TransactionsoftheRoyalSocietyB:BiologicalSciences,369(1653),20130521–
20130521.https://doi.org/10.1098/rstb.2013.0521
Fornito,A.,Zalesky,A.,&Bullmore,E.(2016).Fundamentalsofbrainnetworkanalysis.
AcademicPress.
Friston,K.J.(2011).FunctionalandEffectiveConnectivity:AReview.BrainConnectivity,
1(1),13–36.https://doi.org/10.1089/brain.2011.0008
Graphtheoryasatooltounderstandbraindisorders
50
Garrison,K.A.,Scheinost,D.,Finn,E.S.,Shen,X.,&Constable,R.T.(2015).The(in)stability
offunctionalbrainnetworkmeasuresacrossthresholds.NeuroImage,118,651–661.
https://doi.org/10.1016/j.neuroimage.2015.05.046
Glasser,M.F.,Coalson,T.S.,Robinson,E.C.,Hacker,C.D.,Harwell,J.,Yacoub,E.,…Van
Essen,D.C.(2016).Amulti-modalparcellationofhumancerebralcortex.Nature,
536(7615),171–178.https://doi.org/10.1038/nature18933
Goñi,J.,vandenHeuvel,M.P.,Avena-Koenigsberger,A.,deMendizabal,N.V.,Betzel,R.F.,
Griffa,A.,…Sporns,O.(2014).Resting-brainfunctionalconnectivitypredictedby
analyticmeasuresofnetworkcommunication.ProceedingsoftheNationalAcademy
ofSciences,111(2),833–838.
Gonzalez-Castillo,J.,Chen,G.,Nichols,T.E.,&Bandettini,P.A.(2017).Variance
decompositionforsingle-subjecttask-basedfMRIactivityestimatesacrossmany
sessions.NeuroImage,154,206–218.
https://doi.org/10.1016/j.neuroimage.2016.10.024
Goodkind,M.,Eickhoff,S.B.,Oathes,D.J.,Jiang,Y.,Chang,A.,Jones-Hagata,L.B.,…Etkin,A.
(2015).Identificationofacommonneurobiologicalsubstrateformentalillness.
JAMAPsychiatry,72(4),305–315.
https://doi.org/10.1001/jamapsychiatry.2014.2206
Gordon,E.M.,Laumann,T.O.,Adeyemo,B.,Huckins,J.F.,Kelley,W.M.,&Petersen,S.E.
(2016).GenerationandEvaluationofaCorticalAreaParcellationfromResting-State
Correlations.CerebralCortex,26(1),288–303.
https://doi.org/10.1093/cercor/bhu239
Graphtheoryasatooltounderstandbraindisorders
51
Greene,D.J.,Black,K.J.,&Schlaggar,B.L.(2016).ConsiderationsforMRIstudydesignand
implementationinpediatricandclinicalpopulations.DevelopmentalCognitive
Neuroscience,18,101–112.https://doi.org/10.1016/j.dcn.2015.12.005
Hallquist,M.N.,Hwang,K.,&Luna,B.(2013).Thenuisanceofnuisanceregression:spectral
misspecificationinacommonapproachtoresting-statefMRIpreprocessing
reintroducesnoiseandobscuresfunctionalconnectivity.NeuroImage,82,208–225.
https://doi.org/10.1016/j.neuroimage.2013.05.116
Hilgetag,C.C.,&Goulas,A.(2015).Isthebrainreallyasmall-worldnetwork?Brain
StructureandFunction,221(4),1–6.
Hilgetag,C.C.,&Kaiser,M.(2004).Clusteredorganizationofcorticalconnectivity.
Neuroinformatics,2(3),353–360.https://doi.org/10.1385/NI:2:3:353
Hillary,F.G.,&Grafman,J.H.(2017).InjuredBrainsandAdaptiveNetworks:TheBenefits
andCostsofHyperconnectivity.TrendsinCognitiveSciences.
Hillary,F.G.,Rajtmajer,S.M.,Roman,C.A.,Medaglia,J.D.,Slocomb-Dluzen,J.E.,Calhoun,V.
D.,…Wylie,G.R.(2014).TheRichGetRicher:BrainInjuryElicitsHyperconnectivity
inCoreSubnetworks.PLoSONE,9(8),e104021–e104021.
https://doi.org/10.1371/journal.pone.0104021
Hillary,F.G.,Roman,C.A.,Venkatesan,U.,Rajtmajer,S.M.,Bajo,R.,&Castellanos,N.D.
(2015).Hyperconnectivityisafundamentalresponsetoneurologicaldisruption.
Neuropsychology,29(1),59–75.https://doi.org/10.1037/neu0000110
Honnorat,N.,Eavani,H.,Satterthwaite,T.D.,Gur,R.E.,Gur,R.C.,&Davatzikos,C.(2015).
GraSP:geodesicgraph-basedsegmentationwithshapepriorsforthefunctional
parcellationofthecortex.NeuroImage,106,207–221.
Graphtheoryasatooltounderstandbraindisorders
52
Humphries,M.D.,&Gurney,K.(2008).Network‘Small-World-Ness’:AQuantitativeMethod
forDeterminingCanonicalNetworkEquivalence.PLoSONE,3(4),e0002051–
e0002051.https://doi.org/10.1371/journal.pone.0002051
Hwang,K.,Hallquist,M.N.,&Luna,B.(2013).Thedevelopmentofhubarchitectureinthe
humanfunctionalbrainnetwork.CerebralCortex,23(10),2380–2393.
https://doi.org/10.1093/cercor/bhs227
Jalbrzikowski,M.,Larsen,B.,Hallquist,M.N.,Foran,W.,Calabro,F.,&Luna,B.(2017).
DevelopmentofWhiteMatterMicrostructureandIntrinsicFunctionalConnectivity
BetweentheAmygdalaandVentromedialPrefrontalCortex:AssociationsWith
AnxietyandDepression.BiologicalPsychiatry,82(7),511–521.
https://doi.org/10.1016/j.biopsych.2017.01.008
Laird,A.R.,Eickhoff,S.B.,Li,K.,Robin,D.A.,Glahn,D.C.,&Fox,P.T.(2009).Investigating
thefunctionalheterogeneityofthedefaultmodenetworkusingcoordinate-based
meta-analyticmodeling.JournalofNeuroscience,29(46),14496–14505.
Liang,X.,Wang,J.,Yan,C.,Shu,N.,Xu,K.,Gong,G.,&He,Y.(2012).EffectsofDifferent
CorrelationMetricsandPreprocessingFactorsonSmall-WorldBrainFunctional
Networks:AResting-StateFunctionalMRIStudy.PLoSONE,7(3),e32766.
https://doi.org/10.1371/journal.pone.0032766
Muldoon,S.F.,Bridgeford,E.W.,&Bassett,D.S.(2016).Small-worldpropensityand
weightedbrainnetworks.ScientificReports,6,22057–22057.
Müller,V.I.,Cieslik,E.C.,Serbanescu,I.,Laird,A.R.,Fox,P.T.,&Eickhoff,S.B.(2017).
Alteredbrainactivityinunipolardepressionrevisited:Meta-analysesof
neuroimagingstudies.JAMAPsychiatry,74(1),47–55.
Graphtheoryasatooltounderstandbraindisorders
53
Murphy,K.,Birn,R.M.,Handwerker,D.a,Jones,T.B.,&Bandettini,P.a.(2009).Theimpact
ofglobalsignalregressiononrestingstatecorrelations:areanti-correlated
networksintroduced?NeuroImage,44(3),893–905.
https://doi.org/10.1016/j.neuroimage.2008.09.036
Murphy,K.,&Fox,M.D.(2017).Towardsaconsensusregardingglobalsignalregressionfor
restingstatefunctionalconnectivityMRI.NeuroImage,154(SupplementC),169–
173.https://doi.org/10.1016/j.neuroimage.2016.11.052
Newman,M.(2010).Networks:AnIntroduction(1sted.).OxfordUniversityPress,USA.
Nichols,T.E.,Das,S.,Eickhoff,S.B.,Evans,A.C.,Glatard,T.,Hanke,M.,…Yeo,B.T.T.(2017).
BestpracticesindataanalysisandsharinginneuroimagingusingMRI.Nature
Neuroscience,20(3),299–303.https://doi.org/10.1038/nn.4500
Papanicolaou,A.C.,Rezaie,R.,Narayana,S.,Choudhri,A.F.,Wheless,J.W.,Castillo,E.M.,…
Boop,F.A.(2017).ClinicalApplicationsofFunctionalNeuroimaging:Presurgical
FunctionalMapping.TheOxfordHandbookofFunctionalBrainImagingin
NeuropsychologyandCognitiveNeurosciences,371–371.
Parente,F.,Frascarelli,M.,Mirigliani,A.,DiFabio,F.,Biondi,M.,&Colosimo,A.(2017).
Negativefunctionalbrainnetworks.BrainImagingandBehavior.
https://doi.org/10.1007/s11682-017-9715-x
Poldrack,R.A.,Baker,C.I.,Durnez,J.,Gorgolewski,K.J.,Matthews,P.M.,Munafò,M.R.,…
Yarkoni,T.(2017).Scanningthehorizon:towardstransparentandreproducible
neuroimagingresearch.NatureReviewsNeuroscience,18(2),115–126.
https://doi.org/10.1038/nrn.2016.167
Graphtheoryasatooltounderstandbraindisorders
54
Power,J.D.,Cohen,A.L.,Nelson,S.M.,Wig,G.S.,Barnes,K.A.,Church,J.A.,…Petersen,S.E.
(2011).Functionalnetworkorganizationofthehumanbrain.Neuron,72(4),665–
678.https://doi.org/10.1016/j.neuron.2011.09.006
Power,J.D.,Fair,D.A.,Schlaggar,B.L.,&Petersen,S.E.(2010).Thedevelopmentofhuman
functionalbrainnetworks.Neuron,67(5),735–748.
https://doi.org/10.1016/j.neuron.2010.08.017
Power,J.D.,Mitra,A.,Laumann,T.O.,Snyder,A.Z.,Schlaggar,B.L.,&Petersen,S.E.(2014).
Methodstodetect,characterize,andremovemotionartifactinrestingstatefMRI.
NeuroImage,84,320–341.https://doi.org/10.1016/j.neuroimage.2013.08.048
Price,J.L.,&Drevets,W.C.(2010).Neurocircuitryofmooddisorders.
Neuropsychopharmacology,35(1),192–216.
Roy,A.,Bernier,R.A.,Wang,J.,Benson,M.,FrenchJr,J.J.,Good,D.C.,&Hillary,F.G.(2017).
Theevolutionofcost-efficiencyinneuralnetworksduringrecoveryfromtraumatic
braininjury.PloSOne,12(4),e0170541–e0170541.
Rubinov,M.,&Sporns,O.(2011).Weight-conservingcharacterizationofcomplexfunctional
brainnetworks.NeuroImage,56(4),2068–2079.
https://doi.org/10.1016/j.neuroimage.2011.03.069
Saad,Z.S.,Gotts,S.J.,Murphy,K.,Chen,G.,Jo,H.J.,Martin,A.,&Cox,R.W.(2012).Trouble
atRest:HowCorrelationPatternsandGroupDifferencesBecomeDistortedAfter
GlobalSignalRegression.BrainConnectivity,2(1),25–32.
https://doi.org/10.1089/brain.2012.0080
Schaefer,A.,Kong,R.,Gordon,E.M.,Laumann,T.O.,Zuo,X.-N.,Holmes,A.J.,…Yeo,B.T.T.
(inpress).Local-GlobalParcellationoftheHumanCerebralCortexfromIntrinsic
Graphtheoryasatooltounderstandbraindisorders
55
FunctionalConnectivityMRI.CerebralCortex.
https://doi.org/10.1093/cercor/bhx179
Schönbrodt,F.D.,&Perugini,M.(2013).Atwhatsamplesizedocorrelationsstabilize?
JournalofResearchinPersonality,47(5),609–612.
https://doi.org/10.1016/j.jrp.2013.05.009
Schwarz,A.J.,&McGonigle,J.(2011).Negativeedgesandsoftthresholdingincomplex
networkanalysisofrestingstatefunctionalconnectivitydata.NeuroImage,55(3),
1132–1146.https://doi.org/10.1016/j.neuroimage.2010.12.047
Shekhtman,L.M.,Shai,S.,&Havlin,S.(2015).Resilienceofnetworksformedof
interdependentmodularnetworks.NewJournalofPhysics,17(12),123007.
https://doi.org/10.1088/1367-2630/17/12/123007
Sheline,Y.I.,Price,J.L.,Yan,Z.,&Mintun,M.A.(2010).Resting-statefunctionalMRIin
depressionunmasksincreasedconnectivitybetweennetworksviathedorsalnexus.
ProceedingsoftheNationalAcademyofSciences,107(24),11020–11025.
Shirer,W.R.,Jiang,H.,Price,C.M.,Ng,B.,&Greicius,M.D.(2015).Optimizationofrs-fMRI
Pre-processingforEnhancedSignal-NoiseSeparation,Test-RetestReliability,and
GroupDiscrimination.NeuroImage,117,67–79.
https://doi.org/10.1016/j.neuroimage.2015.05.015
Smith,S.M.,Fox,P.T.,Miller,K.L.,Glahn,D.C.,Fox,P.M.,Mackay,C.E.,…Beckmann,C.F.
(2009).Correspondenceofthebrain’sfunctionalarchitectureduringactivationand
rest.ProceedingsoftheNationalAcademyofSciences,106(31),13040–13045.
https://doi.org/10.1073/pnas.0905267106
Graphtheoryasatooltounderstandbraindisorders
56
Smith,S.M.,Miller,K.L.,Salimi-Khorshidi,G.,Webster,M.,Beckmann,C.F.,Nichols,T.E.,…
Woolrich,M.W.(2011).NetworkmodellingmethodsforFMRI.NeuroImage,54(2),
875–891.https://doi.org/10.1016/j.neuroimage.2010.08.063
Sporns,O.,&Zwi,J.D.(2004).Thesmallworldofthecerebralcortex.Neuroinformatics,
2(2),145–162.https://doi.org/10.1385/NI:2:2:145
Tomasi,D.,Wang,G.-J.,&Volkow,N.D.(2013).Energeticcostofbrainfunctional
connectivity.ProceedingsoftheNationalAcademyofSciences,110(33),13642–
13647.https://doi.org/10.1073/pnas.1303346110
Tzourio-Mazoyer,N.,Landeau,B.,Papathanassiou,D.,Crivello,F.,Etard,O.,Delcroix,N.,…
Joliot,M.(2002).AutomatedanatomicallabelingofactivationsinSPMusinga
macroscopicanatomicalparcellationoftheMNIMRIsingle-subjectbrain.
NeuroImage,15(1),273–289.https://doi.org/10.1006/nimg.2001.0978
vandenHeuvel,M.P.,deLange,S.,Zalesky,A.,Seguin,C.,Yeo,T.,&Schmidt,R.(2017).
Proportionalthresholdinginresting-statefMRIfunctionalconnectivitynetworks
andconsequencesforpatient-controlconnectomestudies:Issuesand
recommendations.NeuroImage,152,437–449.
https://doi.org/10.1016/j.neuroimage.2017.02.005
vandenHeuvel,M.P.,&HulshoffPol,H.E.(2010).Exploringthebrainnetwork:areview
onresting-statefMRIfunctionalconnectivity.EuropeanNeuropsychopharmacology:
TheJournaloftheEuropeanCollegeofNeuropsychopharmacology,20(8),519–534.
https://doi.org/10.1016/j.euroneuro.2010.03.008
Graphtheoryasatooltounderstandbraindisorders
57
VanDijk,K.R.A.,Sabuncu,M.R.,&Buckner,R.L.(2012).Theinfluenceofheadmotionon
intrinsicfunctionalconnectivityMRI.NeuroImage,59(1),431–438.
https://doi.org/10.1016/j.neuroimage.2011.07.044
vanWijk,B.C.M.,Stam,C.J.,&Daffertshofer,A.(2010).ComparingBrainNetworksof
DifferentSizeandConnectivityDensityUsingGraphTheory.PLoSONE,5(10),
e13701–e13701.https://doi.org/10.1371/journal.pone.0013701
Vargas,E.R.,&Wahl,L.M.(2014).Thegatewaycoefficient:anovelmetricforidentifying
criticalconnectionsinmodularnetworks.TheEuropeanPhysicalJournalB,87(7),
161.https://doi.org/10.1140/epjb/e2014-40800-7
Varoquaux,G.,&Craddock,R.C.(2013).Learningandcomparingfunctionalconnectomes
acrosssubjects.NeuroImage,80,405–415.
https://doi.org/10.1016/j.neuroimage.2013.04.007
Vincent,J.L.,Patel,G.H.,Fox,M.D.,Snyder,A.Z.,Baker,J.T.,Essen,D.C.V.,…Raichle,M.E.
(2007).Intrinsicfunctionalarchitectureintheanaesthetizedmonkeybrain.Nature,
447(7140),83.https://doi.org/10.1038/nature05758
Wang,L.,Hermens,D.F.,Hickie,I.B.,&Lagopoulos,J.(2012).Asystematicreviewof
resting-statefunctional-MRIstudiesinmajordepression.JournalofAffective
Disorders,142(1),6–12.
Wang,Y.,Kang,J.,Kemmer,P.B.,&Guo,Y.(2016).AnEfficientandReliableStatistical
MethodforEstimatingFunctionalConnectivityinLargeScaleBrainNetworksUsing
PartialCorrelation.FrontiersinNeuroscience,10.
https://doi.org/10.3389/fnins.2016.00123
Graphtheoryasatooltounderstandbraindisorders
58
Watts,D.J.,&Strogatz,S.H.(1998).Collectivedynamicsof“small-world”networks.Nature,
393(6684),440–442.https://doi.org/10.1038/30918
Whitfield-Gabrieli,S.,&Nieto-Castanon,A.(2012).Conn:AFunctionalConnectivity
ToolboxforCorrelatedandAnticorrelatedBrainNetworks.BrainConnectivity,2(3),
125–141.https://doi.org/10.1089/brain.2012.0073
Woolrich,M.W.,Ripley,B.D.,Brady,M.,&Smith,S.M.(2001).Temporalautocorrelationin
univariatelinearmodelingofFMRIdata.NeuroImage,14(6),1370–1386.
https://doi.org/10.1006/nimg.2001.0931
Yang,Z.,Zuo,X.-N.,Wang,P.,Li,Z.,LaConte,S.M.,Bandettini,P.A.,&Hu,X.P.(2012).
GeneralizedRAICAR:Discoverhomogeneoussubject(sub)groupsbyreproducibility
oftheirintrinsicconnectivitynetworks.NeuroImage,63(1),403–414.
https://doi.org/10.1016/j.neuroimage.2012.06.060
Yeo,B.T.T.,Krienen,F.M.,Sepulcre,J.,Sabuncu,M.R.,Lashkari,D.,Hollinshead,M.,…
Buckner,R.L.(2011).Theorganizationofthehumancerebralcortexestimatedby
intrinsicfunctionalconnectivity.JournalofNeurophysiology,106(3),1125–1165.
https://doi.org/10.1152/jn.00338.2011
Supplementarymaterialsfor
Graphtheoryapproachestofunctionalnetworkorganizationinbraindisorders:
Acritiqueforabravenewsmall-world
MichaelN.Hallquist
and
FrankG.Hillary
TableS1.ParametersusedinnetworksimulationsSimulation Parameter Description ValueWhack-a-nodehyper-connectivity
!" Between-personvariationinmeanFC .2!# Within-personvariationinFC .2!$ Edgenoise .2
%&', * ∈ , MeanshiftinPositivenodetargetsinpatientgroup .14 !&', * ∈ , Between-personvariationinFCshiftsforPositivetargets .07 !-', * ∈ , Within-personvariationinFCvariationofPositivenodei
acrossitsneighbors,j.07
%&', * ∈ . MeanshiftinNegativenodetargetsinpatientgroup -.04 !&', * ∈ . Between-personvariationinFCshiftsforNegativetargets .02 !-', * ∈ . Within-personvariationinFCvariationofPositivenodei
acrossitsneighbors,j.02
Whack-a-nodehypo-connectivity
!" Between-personvariationinmeanFC .2!# Within-personvariationinFC .2!$ Edgenoise .2
%&', * ∈ , MeanshiftinPositivenodetargetsinpatientgroup .04 !&', * ∈ , Between-personvariationinFCshiftsforPositivetargets .02 !-', * ∈ , Within-personvariationinFCvariationofPositivenodei
acrossitsneighbors,j.02
%&', * ∈ . MeanshiftinNegativenodetargetsinpatientgroup -.14 !&', * ∈ . Between-personvariationinFCshiftsforNegativetargets .07 !-', * ∈ . Within-personvariationinFCvariationofPositivenodei
acrossitsneighbors,j.07
Globalinsensitivity
!" Between-personvariationinmeanFC .2!# Within-personvariationinFC .2!$ Edgenoise .2
%&', * ∈ FP, DAN Meanshiftinbetween-moduleconnectivityofF-PandDANnodesincontrolgroup
.1
!&', * ∈ FP,DAN Between-personvariationinbetween-moduleconnectivityofFPandDANnodesincontrolgroup
.05
!-', * ∈ FP, DAN Within-personvariationinbetween-moduleconnectivityofFPandDANnodesincontrolgroup
.05
%&', * ∈ FP, DAN Meanshiftinwithin-moduleconnectivityofF-PandDANnodesincontrolgroup
.2
!&', * ∈ FP,DAN Between-personvariationinwithin-moduleconnectivityofFPandDANnodesincontrolgroup
.1
!-', * ∈ FP, DAN Within-personvariationinwithin-moduleconnectivityofFPandDANnodesincontrolgroup
.1
%&', * ∈ DMN Meanshiftinbetween-moduleconnectivityofDMNnodesinpatientgroup
.1
!&', * ∈ DMN Between-personvariationinbetween-moduleconnectivityofDMNnodesinpatientgroup
.05
!-', * ∈ DMN Within-personvariationinbetween-moduleconnectivityofDMNnodesinpatientgroup
.05
%&', * ∈ DMN Meanshiftinwithin-moduleconnectivityofDMNnodesinpatientgroup
.2
!&', * ∈ DMN Between-personvariationinwithin-moduleconnectivityofDMNnodesinpatientgroup
.1
!-', * ∈ DMN Within-personvariationinwithin-moduleconnectivityofDMNnodesinpatientgroup
.1
Note.P=Positivetargets(3persimulation);N=Negativetargets(3persimulation);FP=frontoparietalnetwork;DAN=dorsalattentionnetwork;DMN=defaultmodenetwork.
FigureS1.Theeffectofthresholdingmethodongroupdifferencesindegreecentralitywhenthereisstronghypoconnectivityinthreerandomlyselectednodes(Negative)andweakhyperconnectivityinthreenodes(Positive).Forsimulationdetails,seeTableS1,Whack-a-nodehypoconnectivity.Thecentralbarofeachrectangledenotesthemediantstatistic(patient–control)across100replicationdatasets(patientn=50,controln=50),whereastheupperandlowerboundariesdenotethe90thand10thpercentiles,respectively.Thedarklineatt=0reflectsnomeandifferencebetweengroups,whereasthelightgraylinesatt=-1.99and1.99reflectgroupdifferencesthatwouldbesignificantatp=.05.a)Graphsbinarizedat5%,15%,and25%density.b)Graphsbinarizedatr={.2,.3,.4}.c)Graphsbinarizedatr={.2,.3,.4},withdensityincludedasabetween-subjectscovariate.d)Strengthcentrality(weightedgraphs).
−15−10−5
05
0.05 0.15 0.25Density
Mea
n t
Density thresholda)
−15−10−5
05
0.4 0.3 0.2Pearson r
FC Thresholdb)
−15−10−5
05
0.4 0.3 0.2Pearson r
Mea
n t
FC threshold, covary densityc)
−15−10−5
05 Node Type
PositiveNegativeComparator
Weightedd)
FigureS2.Theeffectofthresholdingmethodongroupdifferencesindegreecentralitywhenthereisstronghyperconnectivityinthreerandomlyselectednodes(Positive)andweakhypoconnectivityinthreenodes(Negative).RelativetoFigure1,thissimulationappliedproportionalchangestoFCsuchthatedgeswereshiftedasafractionoftheiroriginalstrength,ratherthanshiftingedgesbyanabsoluteamount(i.e.,incorrelationalunits).Thus,FCforpositivenodeswasincreasedby14%,onaverage,whereasnegativenodesweredecreasedby4%.AllsimulationparametersareidenticaltoTableS1,Whack-a-nodehyperconnectivity,butwithchangesappliedproportionately.Thecentralbarofeachrectangledenotesthemediantstatistic(patient–control)across100replicationdatasets(patientn=50,controln=50),whereastheupperandlowerboundariesdenotethe90thand10thpercentiles,respectively.Thedarklineatt=0reflectsnomeandifferencebetweengroups,whereasthelightgraylinesatt=-1.99and1.99reflectgroupdifferencesthatwouldbesignificantatp=.05.a)Graphsbinarizedat5%,15%,and25%density.b)Graphsbinarizedatr={.2,.3,.4}.c)Graphsbinarizedatr={.2,.3,.4},withdensityincludedasabetween-subjectscovariate.d)Strengthcentrality(weightedgraphs).
−505
1015
0.05 0.15 0.25Density
Mea
n t
Density thresholda)
−505
1015
0.4 0.3 0.2Pearson r
FC Thresholdb)
−505
1015
0.4 0.3 0.2Pearson r
Mea
n t
FC threshold, covary densityc)
−505
1015
Node TypePositiveNegativeComparator
Weightedd)