Projectwebsite:https://snap.stanford.edu/ne.
OverviewNetworksareabundantinmanyareasofbiology.Thesenetworksoftenentailnon-trivialtopologicalfeaturesandpatternscriticaltounderstandinginteractionswithinthenaturalsystem.However,networksobservedinreal-worldaretypicallynoisy.Thepresenceofhighlevelsofnoisecanhamperdiscoveryofstructuresanddynamicspresentinthenetwork.
WeproposeNetworkEnhancement(NE),anovelmethodforimprovingthesignal-to-noiseratioofasymmetricnetworksandtherebyfacilitatingthedownstreamnetworkanalysis.NEleveragesthetransitiveedgesofanetworkbyexploitinglocalstructurestostrengthenthesignalwithinclustersandweakenthesignalbetweenclusters.AtthesametimeNEalsoalleviatesthecorruptedlinksinthenetworkbyimposinganormalizationthatremovesweakedgesbyenforcingsparsity.NEissupportedbytheoreticaljustificationsforitsconvergenceandperformanceinimprovingcommunitydetectionoutcomes.
Themethodprovidestheoreticalguaranteesaswellasexcellentempiricalperformanceonmanybiologicalproblems.Theapproachcanbeincorporatedintoanyweightednetworkanalysispipelineandcanleadtoimproveddownstreamanalysis.
RunningNEinMatlabAtcurrentstage,weprovideexamplesshowinghowtoapplyNEtotwoproblemsinbiology,whicharediscussedinthemanuscript.Alldatasetsrequiredtoruntheexamplesareincludedinthisrepository.
Thefirstexampleisaboutfine-grainedspeciesidentification.Runtheexampleas:run_butterfly_network.mThisscriptreportsretrievalaccuracyvaluesandgeneratesaretrievalcurveforthetaskofbutterflyspeciesidentification,asreportedinthemanuscript.
ThesecondexampleisaboutdenoisingHi-Cinteractionnetworks.WeprovidesampledataforChrom16withtwomeasurementresolutions,1kband5kbHi-Cdata.Runtheexampleas:run_hiC_network.mInordertousecommunitydetectionwiththeHi-Cinteractionnetworks,youneedtocompiletwoC++files.Instructionsareprovidedinscriptrun_hiC_network.m.
评论