COMPUTATIONAL NEUROSCIENCE A BRIEF OVERVIEW

COMPUTATIONALNEUROSCIENCE:ABRIEFOVERVIEWAliA.MinaiDepartmentofElectricalandComputerEngineering,UniversityofCincinnati,Cincinnati,Ohio,USA.Correspondenceto:Dr.AliA.Minai,AssociateProfessorofElectricalandComputerEngineering,UniversityofCincinnati,Cincinnati,OH45221-0030,USA.Tel.[phone].Email:[e-mail]andmathematicalmodelingisanincreasinglyusefulapproachforinvestigatingthefunctionalityofthenervoussystem.Thoughsuchmodelinghasbeenusedfordecades,recentadvancesincomputationalpowerandnumericaltechniqueshavegreatlyexpandeditsscope,withacorrespondingincreaseinresearchactivity.Thispaperpresentsabrief–andnecessarilyincomplete–reviewofmethodsandapplicationsincomputationalneuroscience.Computationalneurosciencereferstotheuseofmathematicalandcomputationalmodelsinthestudyofneuralsystems.Itispartofthelarger–increasinglyactive–disciplineofcomputationalbiology,whichappliescomputationalmodelingtoallaspectsofbiologicalorganisms.Quantitativemodelinghasbeenakeycomponentofresearchinneuroscienceformanydecades.Indeed,oneofthemostcelebratedachievementsinthefield–theHodgkin-Huxleymodelforthegenerationofactionpotentials1–wasatriumphofthequantitativeapproach.Also,muchofwhatisunderstoodaboutthefunctionalityofthevisual,auditoryandolfactorysystems,aswellastheneuralbasisoflearningandmemory,hasbeeninformedbymathematicalandcomputationalmodeling.Nevertheless,itisfairtosaythat,untilrecently,computationalmodelingrepresentedonlyasmallpartofthetotalresearcheffortinneuroscience,whichhastraditionallybeendominatedbyexperimentalstudies.Thishasbeguntochangeforseveralreasonswhicharediscussedinthenextsection.Therecentmovetowardscomputationalmodelinghasopenedupnewdirectionsofresearch,andallowedinvestigationofissuesbeyondthosethatareaccessibletodirectexperimentalstudy.Moreimportantly,ithasbroughtnewideasfromfieldssuchasstatisticalphysics,informationtheory,nonlinearsystemstheoryandengineeringintoneuroscience,providingaricherconceptualframeworkforansweringthemostdifficultfundamentalquestionsinthefield.Thisoverviewdiscussesthemotivationfortheuseofcomputationalmodelinginneuroscience,brieflydescribessomeoftheapproaches,andlooksatafewareaswheretheseapproacheshavebeenfruitful.Severalexcellenttextsprovidingdetailsofmethodsandapplicationsincomputationalneurosciencearenowavailable.2,3,4,5,6,7,8,9,10,11MOTIVATIONTheprimarymotivationforusingcomputationalmodelingis,ofcourse,tounderstandthebehaviorofthesystemunderstudyusingmathematicalanalysisandcomputersimulation.Thisiscertainlythecaseinneuroscience.However,theapplicationofcomputationalmodelingtolivingsystems–andespeciallytothenervoussystem–issignificantbecause,unlikemanyphysicalsystemswheresuchmodelingisused(e.g.,planetarysystems,fluidflows,mechanicaldevices,structures,etc.),biologicalsystemscanbeseenexplicitlyasprocessorsofinformation.Thus,computationalmodelsinthesesystemsarenotjusttoolsforcalculationorprediction,butoftenelucidateessentialfunctionality.Inthecaseofneuroscience,thiscanbeseenintermsoftworelatedrolesservedbycomputationalmodeling.Theseare:1)Determiningwhatthevariouspartsofthenervoussystemdo;and2)Determininghowtheydoit.Eachoftheseisdiscussednext.ObtainingaFunctionalDescriptionoftheNervousSystemExperimentalstudiesofthenervoussystematalllevels–sub-cellular,cellularandsystemic–arecriticalforunderstandingtheanatomicalstructuresandphysiologicalprocessesofthesystem,buttheseobservationsmustthenbeorganizedintoacoherentmodelofsystemfunctionality.Thisisonlypossibleiftheappropriateconceptualelementsforsuchafunctionaldescriptionareavailable.Psychologistsandneurologistshavetraditionallyusedperformance(oritsdeficits)asthebasisforassigningfunctionalitytocomponentsofthenervoussystem,whichhasproducedusefulqualitativeandphenomenologicalmodels.Theseareoftensufficientforclinicalpurposes,butprovideonlylimitedunderstandingofthesystemperse.Analternative(orcomplementary)approachisprovidedbyviewingthenervoussystemasacquiring,transforming,storingandusinginformationtocontrolanextremelycomplexsystem–thebody–embeddedinacomplexdynamicenvironment.Inthisview,thefunctionalityofthesystememergesfromlowerlevelphenomenasuchasmembranepotentialdynamics,dendriticcurrentflows,channelkinetics,synapticplasticity,etc.,muchasthefunctionalityofacomputeremergesfromthecurrentsandvoltagesinitscomponents.Aswiththecomputer,theemergentfunctionalityofthenervoussystemdependsontheunderlyingphenomenabutcannotbedescribedentirelyintheirterms.Totrulyunderstandthisfunctionality,itisnecessarytorelatetheconcretephenomenameasuredbyexperimentstotheabstractionsofinformationprocessing–andultimatelytothephenomenaofcognitionandbehavior.Computationalmodelingdoesthisbyprovidingawell-developedformalismrelatingsignalsandinformation.Throughsuchmodeling,mathematicalandcomputationalcanbeapplieddirectlytothenervoussystem,leadingtoacoherentquantitativeandtestablefunctionaldescriptionofthebrainratherthanaqualitativemodeloracompendiumofobservations.ElucidatingthePhysicalBasisofNervousSystemFunctionalityThenervoussystemprocessesinformationatmanyscal