ListofAbstractsfortheworkshop“Reactioncoordinatesfrommolecular
trajectories”,LorentzCenter,Leiden,29Aug–2Sept2016
RobertBest(NIH,Bethesda)
Mappingproteinfoldingdynamicsonto1Dreactioncoordinates—howfar
canwepushthisidea?
Oneofthesimplifyingassumptionsoftheenergylandscapetheoryofprotein
foldingwasthatthefoldingdynamicscouldbecapturedasa1Ddiffusiveprocess
onasuitablereactioncoordinate(onemeasuringsimilaritytothenativestate).
Wehaveshownthatindeedthedynamicsofcoarse-grainedfoldingmodelscan
bedescribedinthisway.Hereweconsiderhowwellatomisticsimultions
conformthethepictureof1Ddiffusion,howtheassumedcoordinatemayneed
tobeimprovedincertaincircumstances,andthelimitationsofasimple1D
diffusionpicture.
PeterBolhuis(UAmsterdam)
Networksofreactioncoordinatesorreactioncoordinatesofnetworks?
Rareeventsincomplexsystemsstillprovidechallengesforcomputational
researchers,bothinsamplingandmakingsenseoftheoften-complexmolecular
trajectories.Transitioninterfacesamplinghasenabledefficientsamplingofrare
eventprocesses,includingcomplextransitionsinvolvinglong-lived
intermediates.Theresultingreweightedpathensemble(RPE)mimicsan
unbiasedinfinitetrajectory,butatamuch-reducedcost.TheRPEcanbeusedto
computeprojectedcommittor,freeenergyandtransmissioncoefficients.
Thereactioncoordinateanalysisofcomplexmulti-pathwaymulti-state
transitionsislessdeveloped.ThetraditionalRCanalysisapproachemployingthe
committortest,orcommittorfunctionwithusingneuralnetworksorLikelihood
Maximization,aimstominimizethenumberofdimensions(parameters)needed
todescribethemechanismoftheprocess.Comingfromtheotherside,the
MarkovStateModelingapproachreducesthecomplexprocesstoa(large)
networkoftransientstates,representedbyatransitionmatrixthatdescribesthe
complexkinetics,withoutworryingaboutreactioncoordinates.
Whilecomingfromdifferentsides,theseapproachesshouldmeetinthemiddle
somewhere.Addingmoreandmoredimensionstodescribecomplextransitions
involvingmultiplestatesinordertoreproduceapossiblyverycomplex
committorsurfaceseemscounterproductive,whereasMSMswiththousandsof
statesmightbeconsideredtoooverwhelmingtounderstandwithoutfurther
dimensionreduction.Analyzingtheinformationcontentoftheusedmodels
mightallowawaytodetermininganoptimaldescriptionlevel.
InthispresentationIwillgiveanoverviewofpathsamplingandanalysis
methodsfortransitionsincomplexsystemsanddiscusstherelationbetween
networksofreactioncoordinatesandreactioncoordinatesofnetworks.
CarloCamilloni(TUMunich)
Collectivevariables,ExperimentaldataandPLUMED
Recentlytherehasbeenanincreaseinterestinusingexperimentaldatatodrive
theresultsofmolecularsimulations.InPLUMEDweareintroducinganumberof
newexperimentaldatacollectivevariablesthatcanbeusedtoanalyzeandbias
MDsimulations.Experimentaldata,asforexamplechemicalshifts,canbeused
asreactioncoordinatestospeedupthesamplingaswellasbiasingforcesto
reweightheunderlyingfreeenergy.Iwillintroducethenewversionof
PLUMED2andthemainimprovementsthatallowtoimplementeasilymore
morecomplexcollectivevariables,biasingandanalysistechniquesandthenew
frameworkofMetainferenceMetadynamicstouseexperimentaldatatoimprove
theunderlyingbehaviourofmolecularsimulations.
RodrigoCasasnovas (ForschungszentrumJulich)(contributedtalk) Understandingprotein-ligandunbindingkineticsfrommetadynamics
simulations
Novelliganddesignstrategiesfocusonoptimizingthekineticsofbindingand/or
unbinding.Thisstrategyprovidescontroloftheligand-targetcomplexresidence
timesandallowsimprovingselectivityand,therefore,reducingdrugdoseand
sideeffects.Rationaloptimizationofthebindingandunbindingkineticsrequires
structuralknowledgeoftherelevanttransitionstates,whichisextremely
difficulttoobtainbyexperimentalmeans.Onthecontrary,atomisticmolecular
dynamics(MD)providedetailedstructuralknowledge,beingthemainlimitation
thattypicalunbindingtimescalesaretoolargetobeobservedinstandardMD
simulations.Hereweusedmetadynamics[1]toefficientlysampletheunbinding
eventsandestimatetheescaperates[2-4]ofatypeIIinhibitorofp38MAP
Kinase.Wefoundthattheslowunbinding(i.e.secondstimescale)ofthisligandis
intimatelylinkedwithsolventfluctuationsinthebindingcavity.Wealsoshow
thatsuchsolventdegreeoffreedomshouldbetakenintoaccounttodesign
reactioncoordinatesthatsatisfactorilydescribethepresentcaseofproteinligandunbindingandmaybe,tomoreorlessextent,otherunbindingprocesses.
[1]Barducci,A.;Bussi,G.;Parrinello,M.Phys.Rev.Lett.2008,100,020603.
[2]Tiwary,P.;Parrinello,M.Phys.Rev.Lett.2013,111,230602.
[3]Salvalaglio,M.;Tiwary,P.;Parrinello,M.J.Chem.Theor.Comp2014,10,1420.
[4][Tiwary2015]Tiwary,P.;Limongelli,V.;Salvalaglio,M.;Parrinello,M.Proc.
Natl.Acad.Sci.U.S.A.,2015,1424461112.
MicheleCeriotti(EPFL,Lausanne) Findingpatternsandmappinglandscapes.
Moleculartrajectoriesholdthekeytounderstandingstabilityandreactivityof
materialsandmolecules.Iwilldiscusshowdifferentkindsofmachine-learning
algorithmscanbeusedtoassistonininterpretingandanalyzingatomistic
simulationdata.First,Iwilldiscusshowitispossibletorecognizethemolecular
buildingblocksofamaterial,e.g.hydrogenbonds,coordinationenvironments,
etc.Second,Iwillpresentafewexamplesofhownon-lineardimensionality
reductiontechniquescanbeusedtogeneratecompactandinsightful
representationsoftherelationsbetweendifferent(meta)stablestatesofa
molecule,anddiscusshowsuchamappingcanbeexploitedtoaccelerate
explorationandsamplingofconfigurations.
CeciliaClementi(RiceU,Houston) Lowdimensionalrepresentationofproteinconformationaldynamics:tools
andchallenges
Theunderstandingofemergingcollectivebehaviorsinbiomolecularcomplexes
representsamajorchallengeinmodernbiophysics.Inordertocapturethelong
timescalebehaviorofthesesystemsweuseacombinationoftools,including
multi-resolutionnonlineardimensionalityreductionanddiffusionanalysisto
obtainreliablelow-dimensionalrepresentationsandmodelsoftheirdynamics.
Theresultsshowthattheproposedmethodscanefficientlyfindlowdimensional
representationsofcomplexprocessessuchasproteinfoldingand
conformationalchanges,andsuggeststrategiestosimplifysignificantlythestudy
ofsuchprocesses.Alowdimensionalrepresentationofthedynamicscanalsobe
usedon-the-flytodesignadaptivesamplingstrategies,tospeedupthesampling
ofcomplexmacromolecularsystems.
DaanCrommelin(CWIAmsterdam)
Importancefunctionsformultilevelsplitting
Inthemultilevelsplittingmethodologyforrareeventsimulation,thechoiceof
theimportancefunction(akareactioncoordinateororderparameter)iscrucial
inordertoachievevariancereduction.Iwilldiscusshowresultsfromlarge
deviationstheorycanhelptodesignsuitableimportancefunctionsforsplitting.
Fromanapproximateratefunctiononecanbuildanimportancefunctionwhose
levelsetsareapproximatelevelsetsoftheprobabilitytoreachtherareeventset.
Iwillshownumericalresultsfromasimpleexamplesystemanddiscusshowthe
increasedcomputationalcostofevaluatingthelarge-deviationsbased
importancefunctioncanbereduced.
ChristophDellago(UVienna)
Reactioncoordinateforfreezing:doweunderstandcrystallization?
Onaphenomenologicallevel,thefreezingofasupercooledliquidcanbe
understoodintheframeworkofclassicalnucleationtheory,whichassertsthat
thisprocessproceedsviatheformationofasmallcrystallinenucleusgrowingin
themetastableliquid.However,molecularsimulationsofvarioussystemshave
revealedthatthesimplepictureofclassicalnucleationtheorymissesimportant
aspectsofthefreezingtransition.Inmytalk,Iwilldiscusssomeinsightswehave
obtained,usinglikelihoodmaximization,forthefreezingandmeltingofgold
clustersandthecrystallizationofasquare-wellpolymer.
RubenDemuynck(GenthU,Zwijnaarde) (contributed)
Advancedmoleculardynamicssimulationstoconstructfreeenergy
profilesofcomplextransformationsinnanoporousmaterials
Inthiscontribution,weexploretheapplicabilityofadvancedmolecular
dynamicsmethodssuchasmetadynamics,umbrellasampling,...toconstructfree
energyprofilesforcomplextransformationinnanoporousmaterials.Twocase
studiesareaddressedwhichideallyillustratethecomplexityofchoosingthe
propermethodtoconstructthefreeenergyprofileandtherightcollective
variables.Thefirstcasestudyistakenfromthefieldofmetal-organic
frameworks(MOFs).Thesematerialshaveattractedanenormousattentionthe
last15yearsastheyexhibitfascinatingpropertiesandopenawidevarietyof
possibleapplications.Oneofthemostintriguingpropertiesconcernsthe
inherentflexibilitysomeframeworkmaterialspossess.Otherthanregular
materials,theseMOFsundergolargedeformationsundertheinfluenceof
externalstimulisuchaspressure,temperature,guestad-sorption...This
phenomenaiscalledframeworkbreathing.Akeyelementofvarious
thermodynamicmodelswhichdescribeframeworkbreathingisthefreeenergy
profileoftheemptyframework.[1,2]Hence,thesestructuraltransitionsarerare
events,advancedsamplingsimulationsarerequiredtoconstructsuchprofiles.
Thekeyofadvancedsamplingmethodsconsistsinenhancingthesamplingalong
asetofwell-chosencollectivevariables.Weconsidertwoexamples:thewelldocumentedMIL-53(Al)andthenewlydevelopedUiO-Zr,synthesizedwiththe
flexibletrans-1,4-cyclohexaned-icarboxylicacidaslinker.[3]Intheformerthe
breathingbehavioriseasilycapturedfromafreeenergyprofileintermsofthe
volume,forthelatterweshowthataningeniouscombinationofadvanced
samplingmethodswithproperlychosencollectivevariables,isnecessaryto
describethestructuraltransitionsofthismaterial.
Asecondcasestudyconcernstheconstructionoffreeenergyprofilesforthe
diffusionofsmallguestspeciesinzeolitematerial.Thediffusivebehaviourisof
particularrelevancewithinthecontextofthemethanol-to-olefin(MTO)process.
Hence,knowledgeofthesediffusionbarriersisessentialtoobtainthemolecular
controlovertheMTOprocessconditions.Inparticular,weinvestigatethe
diffusionofpropenethroughan8-ringofSAPO-34.Forthisactivatedprocess,we
employadvancedsamplingmethodsincombinationwithawell-chosen
collectivevariable.Fromthesesimulations,westudyframeworkflexibilityin
relationwiththediffusivebehaviour.Moreover,wediscusshowextraspec-tator
moleculesinfluencestheparticularchoiceofcollectivevariableanditsinfluence
onthediffusionbarrier.
(1)Coudert,F.-X.;Jeffroy,M.;Fuchs,A.H.;Boutin,A.;Mellot-draznieks,
C.;Chimie,D.P.;Uni,V.;Curie,M.;Paris,F.2008,14294–14302.(2)Vanduyfhuys,
L.;Ghysels,A.;Rogge,S.;Demuynck,R.;VanSpeybroeck,V.MolecularSimulation
2015,7022.(3)Bueken,B.;Vermoortele,F.;Cliffe,M.J.;Wharmby,M.T.;
Foucher,D.;Wieme,J.;Vanduyfhuys,L.;Martineau,C.;Stock,N.;Taulelle,F.;Van
Speybroeck,V.;Good-win,A.L.;DeVos,D.ChemistryAEuropeanJournal2016,
22,3264–3267.
RonElber(UTexas,Austin) Globalandlocalapproachesforcalculationsofreactionpathways.
Idescribeanapproachtodetermineglobally-optimizedreactioncoordinates
fromMilestoningnetworks.Anapplicationofanon-MarkovianandMarkovian
approximationtomembranedensityfluctuationsarediscussed.Ialsodiscussan
algorithmtolocallyoptimizeaminimumfreeenergypathway(andareaction
coordinate)bygrowingachainofconfigurationsfromreactanttoproductsand
optimizeafunctionalofthepaths.Thealgorithmallowsforhundredsofcoarse
variablestobeusedintheoptimization.Theadvantageofthelastalgorithm,
comparedtootherfunctionalapproaches,isthatnoinitialguessfortheoverall
pathisrequired.Highenergybarriersalongthepathwaycanmakethealgorithm
lessefficient.
PietroFaccioli(TrentoU,Trento)
Variationalprinciplesforreactioncoordinatesandbiasedreaction
pathways
Acommonstrategytogeneratereactivepathwaysforraremacromolecular
transitionsconsistsinintroducinganunphysicalforceactingalongthedirection
ofapreviouslydeterminedreactioncoordinate(RC).
InthistalkIshalldiscusshowrigorouslyderivedvariationalprinciplescanbe
usedtoreducetoaminimumtheerroronthereactivedynamicswhichis
generatedbyintroducingsuchanexternalbiasingforce.First,wederive
rigorouslyaFeynman-Kac-typeofvariationalprinciplewhichenablestoidentify
theoptimumbiasingreactioncoordinatewithinamodelsub-space,whichyields
probabilitydistributionsinconfigurationspacewhichareclosesttothatofthe
unbiaseddyamics.
NextIdiscusstherecentlyintroducedBiasFunctionalapproach[1]inwhichan
additionalvariationalprincipleisappliedtoselectoutthemostrealisticbiased
pathoutoutofatrialensembleofbiasedreactivetrajectorieswithidentical
boundaryconditions.Fromthismethoditisalsopossibletoreconstructthefree
energyalongthebiasingreactioncoordinate.
Finally,weflashonapplicationsofthisapproachtoall-atomsimulationsof
proteinfolding/misfoldingtransitionswhichoccurovertime-scalesaslongas
severalminutes.Dependingontimeavailability,weshallpresentascheme
whichenablestoachieveadirectcomparisonofthesesimulationsagainstthe
resultsoflinearandnon-linearspectroscopy.
XuhuiHuang (HKUST,HongKong)
UsingtheProjectionOperatorApproachtoIdentifyOptimalKinetic
LumpingandRecoverSlowestConformationalDynamicsofComplex
Systems
MarkovStateModel(MSM)becomesapopularapproachininvestigating
conformationaldynamicsofcomplexsystems.Butinmanycases,oneoften
needsalargenumberofstates(e.g.1000to10000microstates)toconstructa
Markovianmodel,andanyattempttofurtherreducethisnumberwillrenderthe
modelnon-Markovianunlessamuchlongerlagtimeisused.Ononehand,
modelswithtoomanystateswillhindereasycomprehensionofbiological
insights,andkineticlumpingisoftenrequiredtoreducethenumberofstates
(i.e.macrostates).Ontheotherhand,whenthenumberofstatesissmall,amuch
longerlagtimeisneededforthemodeltobeMarkovian,whilethelagtimeis
limitedbythelengthofmoleculardynamicssimulations.
Inthistalk,Iwillintroduceanewmethodtoobtainkineticlumpedmacrostate
MSMsthatcanreproducetheslowestdynamicsofthesystem(frommicrostate
MSM)butwithveryshortlagtime.Toachievethis,weapplytheprojection
operatorframeworktobackwardprojectcoarse-grainedmacrostatetransition
modes(oreigenmodes)toslowesttransitionmodesofadetailedmicrostate
Markovchain,sothatwecancomparethemagainstthosefromtheoriginal
microstateMSM.Onceweidentifyakineticlumpingwithmatchingeigenmodes,
wecanutilizethelumpingandreconstructamacrostateMSMthatcanfaithfully
reproducetheslowestkineticsofthesystem.Thiswillallowustoconstruct
goodmacrostateMSMswithveryshortlagtime.
Finally,Iwillshowourinitialattemptstoapplythisnewmethodtoidentifybest
reactioncoordinatestoinvestigateproteinconformationalchanges.
GerhardHummer(MPIFrankfurt) Reactioncoordinatesintheanalysisofsingle-moleculeexperiments
Single-channelconductancerecordings,single-moleculeFRETtraces,andsinglemoleculeforce,torque,ordistancemeasurementsarerepresentativesofa
rapidlygrowingfamilyofexperimentsreportingonthedynamicsofselect
observableswithinindividualmolecules.InmypresentationIwillexplorethe
questionofidentifying,assessing,andoptimizingreactioncoordinatesonthe
basisofsingle-moleculeexperimentaldata,possiblycombinedwithmolecular
simulations.Iwillalsoexaminetheusefulnessofreactioncoordinatesinkinetic
modeling.Specificallyinthecontextofsingle-moleculeforcespectroscopy,Iwill
lookatproblemsassociatedwithunresolveddynamics
BettinaKeller(FUBerlin) Harnessingchemicalintuitiontofindtheslowdynamicsubspace.
Accordingtochemicalintuitionreactioncoordinatesconnecttheminimainthe
molecularfree-energysurfaceandcanbeexpressedintermsoftheslowinternal
coordinatesofthemolecule.However,itisoftendifficulttotranslatethis
intuitiondirectlyintowell-definedreactioncoordinates.FromMarkovmodel
theory,weknowthatthereactioncoordinatesareembeddedintoalowdimensionalsubspaceofthefullconformationalspace.Apossibleintermediate
stepmightbetousechemicalintuitiontofirstidentifythelowdimensional
subspaceandthenlookforreactioncoordinatesinthisspace.Recent
developmentshavemadeitpossibletoharnesspriorchemicalorstructural
knowledgeaboutthesystemfortheconstructionofhighlyaccurateMarkov
models.Iwilldiscussandcomparetwoofthesemethods.
Thefirstmethodisavariationalapproachwithbasisfunctionscustomizedto
modelpeptidedynamicsinwhichweusetheideathatthedynamicsofpeptides
canbeexpressedintermsofthedynamicswithintheRamachandranplaneof
eachresidue.Inthesecondmethodweidentifytheminimainthefree-energy
landscapebyahierarchicalclusteralgorithmandusethemascoresinthecoresetapproachtomoleculardynamics.
References:
B.Keller,X.Daura,W.F.vanGunsteren,J.Chem.Phys.,132,074110(2010)
C.Schütte,F.Noé,J.Lu,M.Sarich,E.Vanden-Eijnden,134,204105(2011)
F.Nüske,B.G.KellerG.Pérez-Hernández,A.S.J.S.Mey,F.Noé,J.Chem.Theory.
Comput.,10,1739-1752(2014)
F.Vitalini,F.Noé,B.G.Keller,J.Chem.Theory.Comput.,11,3992-4004(2015)
SergeiKrivov (ULeeds)
Nonparametricvariationaloptimizationofreactioncoordinates.
Accuratedeterminationofthecommittorreactioncoordinatefromatomistic
trajectoriesisaverydifficultproblem.Apopularapproachistoselecta
functionalformwithmanyparametersapproximatingthecoordinateandto
trainitonthetrajectoriesusingvariouscriteria.Intheend,onefindsoptimal
valuesoftheparameterssuchthatthefunctionalformprovidesbest
approximationtothecommittorcoordinate.Amajorproblemwithsuch
approachesisthatitisnottrivialtoselectafunctionalformthatcanaccurately
approximatethecommittorandapoorinitialchoicemayleadtosub-optimal
results.Wehavesuggestedanonparametricapproachwhichallowsoneto
optimizethereactioncoordinatewithoutselectingitsfunctionalformandthus
avoidsthissourceoferror.
AlessandroLaio(SISSA,Trieste)
Automatictopographyofcomplexandmultidimensionalprobability
distributions.
Wedescribeanadaptiveandparameter-freedensityestimatorthatprovidesan
accuratemeasureoftheprobabilityandofitsuncertaintyforhighlynon-uniform
samplesembeddedinhighdimensionalspaces.Theestimatorusesonly
distancesbetweenthepointsandnotthecoordinatesanddoesnotrequireany
assumptiononthefunctionalformoftheprobabilitydistribution.We
demonstrateitsaccuracyondatasetsincludingprobabilityvaluesspanning
eightordersofmagnitude,showingthatthepredicteduncertaintyaccurately
accountsforthedeviationofthedensityestimationfromitstruevalue.The
availabilityofanerrorestimateallowsdistinguishinggenuinedensitypeaks
fromstatisticalfluctuationsduetofinitesampling.Thisisanessentialingredient
forrobustlyinferringthetopographyoftheprobabilitydistributions,namelyfor
findingthelocationandheightofthedensitypeaksandofthesaddlesbetween
them.
KrestenLindorff-Larsen(UCopenhagen) Reactioncoordinatesinmolecularsimulations–inputoroutput?
Reactioncoordinatesprovideacompresseddescriptionofkeymolecular
processese.g.inproteinfoldingorconformationalexchange.Theythusenableus
tointerpretthesometimes-overwhelmingdetail-richnessinbiomolecular
simulaions.Atthesametime,reactioncoordinatesareoftenselectedtodescribe
theslowestdegreesofmotioninaprotein,andthusprovidesopportunitiesto
enhancesamplingbybiasingthesimulationsalongthesereactioncoordinates.I
willdiscussexamplesofhowcalculationsofkineticpropertiescanbeusedto
improvereactioncoordinates,andhowimprovedreactioncoordinatescanbe
usedtocalculatekineticsofproteinmotions.
DmitriiEMakarov(UTexas,Austin)
Reactioncoordinatesandpathwaysofmechanochemicaltransformations
Recentsingle-moleculeforcespectroscopystudiesdirectlyandindirectlyprobe
thepathwaysofmechanochemicaltransformations,wheremoleculartransitions
(e.g.proteinorDNAunfolding)areaccelerated(or,insomecases,suppressed)
bymechanicalforces.Inthistalk,Iwilldiscusswhetheritispossibletodevise
low-dimensionaldescriptionsofsuchprocessesandreviewexperimental
evidenceforandagainstessentialmultidimensionalityofmechanochemical
transformations.
FrankNoe(FUBerlin)
"Machinelearningmethodsfordimensionreduction,reactioncoordinate
identificationandextractingkineticsfrommoleculardynamics"
Manyoftherecentlyusedmethodstoperformdimensionreaction,reaction
coordinateidentificationandkineticmodelbuildingfrommoleculardynamics
simulationsareintrinsicallymachinelearningmethods,butontheotherhand
havedeepinterpretationinthestatisticsphysicsofthemolecularsystems
investigated,inthattheyapproximatemeaningfulphysicalquantities.
HereIwilleludetotheseconnections,pointoutpossiblewaystoexploitrecent
developmentsinmachinelearninginourarea,andpresentanewalgorithmfor
theunbiasedidentificationofslowcoordinates/reactioncoordinatesfromshort
off-equilibriumtrajectories.
BaronPeters (UCSB,SantaBarbara)
RareEventsMethods,ReactionCoordinates,andUsefulRateTheories
Amongthearsenalofrareeventsmethodsandreactionratetheories,three
theoriesstandoutfortheirextraordinarycapabilities.Specifically,theseare
harmonictransitionstatetheory,classicalnucleationtheory,andMarcus
theory.Forbondbreaking/makingreactions,nucleation,andelectrontransfer,
thesethreetheorieshavebecomethecentralconceptualframeworksbehind
countlesscomputationalmethods.Theyhaveanunrivaledabilitytopredict
kinetictrendsasafunctionoftemperature,supersaturation,electrochemical
potential,andotherprocessvariables.Theyevenprovidethetheoretical
justificationforfreeenergyrelationshipsthatpredictkinetictrendsacrossseries
ofrelatedreactions.Thecapabilitiesofthesethreeclassictheoriesderivefroma
sharedfeatureattheirfoundations:eachwasbuiltaroundascalarreaction
coordinatewithaclearmechanisticinterpretation.Moreover,theirreaction
coordinatesaregenerallyapplicableacrossmanyreactions,nucleation
processes,andelectrontransferprocesses,respectively.Bycomparison,
computationalframeworksbasedonabstractreactioncoordinateslike
committorsoreigenvectorsofthemasterequationrequireamassivesimulation
effortandyieldkineticsonlyatthesimulatedprocessconditions.Takingthe
threeclassictheoriesasaguide,Iconsidersomeactivatedprocessesthatremain
poorlyunderstoodandpotentialstrategiestodiscoversimilarlygeneralizable
reactioncoordinatesandusefultheories.Iwillhighlightsomerecentprogress
towardthesegoalsusingthelikelihoodmaximizationapproach.
EdinaRosta(KingsCollege,London)
IdentificationandAnalysisofTransitionandMetastableMarkovStates
Wepresentanewmethodthatenablestheidentificationandanalysisofboth
transitionandmetastableconformationalstatesfromatomisticorcoarsegrainedmoleculardynamics(MD)trajectories.Ouralgorithmispresentedand
studiedbyusingbothanalyticalandactualexamplesfromMDsimulationsofthe
helix-formingpeptideAla5,andofalargersystem,theepidermalgrowthfactor
receptor(EGFR)protein.Inallcases,ourmethodidentifiesautomaticallythe
correspondingtransitionstatesandmetastableconformationsinanoptimal
way,withtheinputofasetofrelevantcoordinates,bycapturingaccuratelythe
intrinsicslowestrelaxationrate.Ourapproachprovidesageneralandeasyto
implementanalysismethodthatprovidesuniqueinsightintothemolecular
mechanismandtherarebutcrucialratelimitingconformationalpathways
occurringincomplexdynamicalsystemssuchasmoleculartrajectories.
DavidSwenson(UAmsterdam)(contributed)
GenerationandAnalysisofArbitraryPathEnsemblesusing
OpenPathSampling
Pathsamplingmethods,suchastransitionpathsamplingandtransitioninterface
sampling,areverypowerfultoolsforstudyingthemechanismsandkineticsof
rareeventsincomplexsystems.Thecentralconceptinpathsamplingisthepath
ensemble,whichisarestrictionofthecompletetrajectoryspace.Differentpath
ensemblesusedifferentrestrictions.Inthelastfewyears,thevarietyofpath
samplingalgorithmshasexpandedsignificantly.Historically,eachnewpath
ensemblerequirednewsoftware.
HerewepresentOpenPathSampling,anopen-sourcePythonpackageforgeneric
samplingandanalysisofarbitrarypathensembles.Inaddition,wedemonstrate
someoftheanalysistoolsthatcanbebuiltwiththispackage.Byframing
trajectoryanalysisintermsofpathensembles,manyanalysisquestionscanbe
answeredwithminimalsoftwaredevelopment.Forexample,wecanidentifythe
setsoftrajectoriesthatfollowdifferentreactionchannels,orthatinclude
recrossings.Othercommonanalyses,suchasthelifetimeofastateortheflux
throughadividingsurface,canalsobeeasilywrittenintermsofapath
ensemble-basedapproach.AsaPythonpackage,OpenPathSamplingcanbeused
asalibraryuponwhichotherreactioncoordinateanalysescanbebased,and
futureversionsofthepackagewillincludetoolsforextractingreaction
coordinatesfrompathsamplingsimulations.
PratyushTiwary(ColumbiaU,NewYork)
Identifyingandenhancingimportantfluctuationsforsamplingmolecular
systemswithrareevents
Inmodern-daysimulationsofmany-bodysystems,muchofthecomputational
complexityisshiftedtotheidentificationoflow-dimensionalslowlychanging
molecularorderparametersorcollectivevariables(CVs).Avastarrayof
enhanced-samplingmethodsarebasedontheidentificationandbiasingofthese
CVs,whosefluctuationsareimportantindrivingtherareeventsofinterest.Here,
Iwilldescribeanewalgorithmcalledspectralgapoptimizationoforder
parameters(SGOOP)[1]forfindingsuchoptimallow-dimensionalCVsfor
samplingmolecularsystemswithrareevents,givenamuchlargersetof
candidateCVs.SGOOPisbasedontheideathatthebestCVdisplaysthe
maximumseparationoftimescales,orspectralgap,between
visibleslowandhiddenfastprocesses.Thespectralgapiscalculatedthrough
amaximumpathentropyorCaliberbasedmodelthatuseslimitedpriorstatic
anddynamicinformationaboutthesystem.Finally,Iwillshowsome
applicationsofSGOOPandrelatedmethodswithaspecialfocusonkineticsof
drugunbinding[2]withclassicalMD.
References:
[1]Tiwary&Berne,Proc.Natl.Acad.Sci.1132839(2016)
[2]Tiwary&Berne,submittedtoJ.Chem.Phys.,arxiv:1605.07090(2016)
TitusVanErp(NTNU,Trondheim)
Analyzingcomplexreactionmechanismsusingpathsampling
Weintroduceanapproachtoanalyzecollectivevariablesregardingtheir
predictivepowerforareaction.Themethodisbasedonalreadyavailablepath
samplingdataproducedbyforinstancetransitioninterfacesamplingorforward
fluxsamplingwhicharepathsamplingmethodsusedforefficientcomputationof
reactionrates.Byasearchincollectivevariablespaceameasureofpredictivenesscanbeoptimizedand,inaddition,thenumberofcollectivevariablescanbe
reducedusingprojectionoperations,whichkeepthismeasureinvariant.The
approachallowstestinghypothesesonthereactionmechanism,butcouldin
principlealsobeusedtoconstructthephasespacecommittorsurfaceswithout
theneedofadditionaltrajectorysampling.Theprocedureisillustratedforaonedimensionaldoublewellpotential,atheoreticalmodelforanion-transfer
reactioninwhichthesolventstructurecanlowerthebarrier,andanAbInitio
moleculardynamicsstudyofwaterauto-ionization.Theanalysistechnique
enhancesthequantitativeinterpretationofpathsamplingdatawhichcan
providecluesonhowchemicalreactionscanbesteeredindesireddirections.
JocelyneVreede(UAmsterdam)
Pathsamplingsimulationsofthemechanismsandratesoftransitions
betweenWatson-CrickandHoogsteenbasepairinginDNA
DNAduplexespredominantlycontainWatson-Crick(WC)basepairs.However,
atanytime,anon-negligiblefractionofbasepairsexhibitadifferentinteraction
pattern,calledHoogsteen(HG)wherethepurinebaseisrotatedapproximately
180°relativetoitsorientationintheWCpairing.TheconversionfromWCtoHG
alterstheconformationofDNA,andmayplayaroleinseveralprocesses,
includingrecognitionandreplication.Thetransientnatureofthesetransitions
hamperthoroughexperimentalinvestigation.Moleculardynamicssimulations
complementexperimentalworkbyprovidinginsightsatveryhighspatialand
temporalresolution.Pathsamplingmethodsfocusthemoleculardynamicseffort
onthedynamicsduringatransition,thusavoidingthelongwaitingtimesin
stablestates.Weapplytwodifferentpathsamplingtechniquestostudythe
transitionsbetweentheWCandHGbasepairingmotifs.Byusingtransitionpath
samplingweobservedthattheWCtoHGconversioncanproceedalongseveral
mechanisticrouteswithvaryingdegreesofexposureofthepurine.Wecomputed
therateforeachprocessbyemployingtransitioninterfacesampling.
MaxWelling(UAmsterdam)
ModernMachineLearningToolsRelevantforMolecularDynamics
Machinelearninghasrecentlydeliveredanumberoftoolsthathaveproven
usefulforbothindustrialaswellasscientificapplications.
InthistalkIwillreview2ofthesetoolsthatIbelievemayberelevantforMD.
1)DeepLearning.DeepLearningisamethodforclassifyinginputdatainto
categories.Itisparticularlysuitedtoworkwithraw,unstructuredsignals.
WithintheMDdomainitcouldforinstancebeusedtodeterminesuitable
reactioncoordinates.
2)BayesianOptimization.BayesianOptimizationisusedtooptimizeparameters
ofacomplicated(non-differentiable)function.WithintheMDdomainitcouldfor
instancebeusedtooptimizetheparametersthatdeterminethebehaviorofan
MCMCorMDsamplingrun.
IwillclosewithabriefdiscussionofsomeotherMLmethodsthatmay(ormay
not)beusefulforchemistrymoregenerally.
MichaelWoodside(U.Alberta,Edmonton)
Measuringtransitionpathsinthefoldingofsinglemolecules
Transitionpathsarethetrajectoriesfollowedduringthefleetingmomentswhen
molecularstructurechangesduringfoldingreactions.Becausetheyprovidea
directlookattheintermediatetransitionstatesthatdominatethedynamicsof
folding,transitionpathsencapsulatethecriticalinformationabouthowstructure
forms.Owingtotheirbrevity,however,ithasnotpreviouslybeenpossibleto
measuretransitionpathsdirectly,andonlypropertiessuchastheaveragetime
tocrossthetransitionpathhavebeenaccessible.Usinghigh-resolutionoptical
tweezerstounfoldandrefoldsinglemoleculesundermechanicalload,wehave
nowmeasuredthousandsoftransitionpathsdirectlyforbothnucleicacidand
proteinfolding,observingagreatdiversityofbehaviorasthemolecules
traversedthebarrierregion.Iwilldiscussourstudiesofthedistributionof
transitiontimes,whichweusedtolearnaboutthediffusioncoefficientgoverning
thetimescaleofthekinetics,aswellastheshapeofthetransitionspaths,which
weusedtolearnaboutthedynamicsofthemotionsthroughthetransition
states.Iwillalsoshowhowweusedthestatisticsoftransitionpathoccupancyto
testthequalityoftheend-to-endextensionasareactioncoordinateinpulling
experiments.Finally,Iwilldiscussinitialworkapplyingthismethodtocompare
thequalityofextensionasareactioncoordinatewhentheforceisappliedto
differentpartsofthesameprotein.