Relative Binding Free Energy Calculations in Drug Discovery：Recent Advances and Practical Considerations.docx

ThisisanopenaccessarticlepublishedunderanACSAuthorChoiceLicense,whichpermitscopyingandredistributionofthearticleoranyadaptationsfornon-commercialpurposes.ACSEftxsJournaldfCHEMICALINFORMATIONandMODELING®CiteThis:J.Chern.Inf.Model2017,57,2911-2937RelativeBindingFreeEnergyCalculationsinDrugDiscovery：AdvancesandPracticalConsiderationsPerspectivepubs.acs.org/jcimRecentZoeCournia,*,+BryceAllen/andWoodySherman*tBiomedicalResearchFoundation,AcademyofAthens,4SoranouEphessiou,11527Athens,GreecetSiliconTherapeutics,30()AStreet,Boston,Massachusetts()221(),UnitedStatesfreeenergy (RBFE) calculations, which rely on physics-based niolecjpromise in reliably generating accurate predictions in the context oaccumulating developments in the underlying scientific mathods ialgorithms) coupled with vast increases in computational resdyi图Mounting evidence from retrospective validation studies, 碰新suggests that RBFE simulations can now predict the affinity diMM)mrulatiolfcand statistical mecABSTRACT:Accurateinsilicopredictionofprotein-ligandbindingaffinisha$heen;歌“螃Uobjectiveofstructurcbaseddrugdesignfordecadesduetotheputativevalueitwouldbringtothedr6sSiscovcryprocess.However,computationalmethodshavehistoricallyfailedtodelivervalueinial-wdtlddr耐discoveixapplicationsduetoavarietyofscientific,technical,andpracticalchallenges.Recently,afapiilyo碗叫tU)|srelativebindinghies,hvTshownugiscoveryprojeciades of research onisadvancearisesifromrcefieldsandjthroughputtodeliverconsiderablevalueinhit-to-leadandleadoptimizationefforts.Here,wepresentanoverviewofcurrentRBFEimplementations,highlightingrecentadvancesandremainingchallenges,alongwithexamplesthatemphasizepracticalconsiderationsforobtainingreliableRBFEresults.Wefocusspecificallyonrelativebindingfreeenergiesbecausethecalculationsarelesscomputationallyintensivethanabsolutebindingfreeenergy(ABFE)calculationsandmapdirectlyontothehit-to-leadandleadoptimizationprocesses,wherethepredictionofrelativebindingenergiesbetweenareferencemoleculeandnewideas(virtualmolecules)canbeusedtoprioritizemoleculesforsynthesis.WedescribethecriticalaspectsofrunningRBFEcalculations,fromboththeoreticalandappliedperspectives,usingacombinationofretrospectiveliteratureexamplesandprospectivestudiesfromdrugdiscoveryprojects.Thisworkisintendedtoprovideacontemporaryoverviewofthescientific,technical,andpracticalissuesassociatedwithrunningrelativebindingfreeenergysimulations,withafocusonreal-worlddrugdiscoveryapplications.WeofferguidelinesforimprovingtheaccuracyofRBFEsimulations,especiallyforchallengingcases,andemphasizeunresolvedissuesthatcouldbeimprovedbyfurtherresearchinthefield.Journal of Chemical Information and Modelingsu-aEBpuqs = qnd UJEqs Aollnu三3。一 oi MOL- uo suopdo sQ.sQp-5MMU-zBqs/el)JOso«sqnd、/一 sdlzQQS05 8寸V-ZI急 zzoz zroJE 工 § SON-G-soz B>P3PBOCMOCI氐ofofin回INTRODUCTIONGeneralOverview.Optimizationofbindingaffinity,selectivity,andotheroff-targetinteractionsisacriticalpartofhit-to-leadandleadoptimizationeffortsindrugdiscovery.Relativebindingfreeenergy(RBFE)calculationsofferanattractiveapproachtopredictprotein-ligandbindingaffinitiesinsilicousingmolecularsimulationsandstatisticalmechanicsasawaytocomputefreeenergydifferencesbetweencongenericmolecules.Fromacomputationalperspective,RBFEsimulationsareofparticularinterestduetotheirrigorousstatisticalmechanicalframework,accuratemodelingofbiologicalsystems(eg,proteinflexibility,explicitsolvent,cofactors,ions,concertedmotions,andentropy,tonameafew),anddirecttranslationtoreal-worldproblems(e.g,，hit-to-leadandleadoptimization).However,historicalchallengeshavelimitedthesuccessoffreeenergysimulationsindrugdiscovery?hamely,thehighcomputationalcosts,limitedforcefieldaccuracy,andtechnicalchallengestosetup/run/analyzefreeenergysimulations.ThehighcomputationaldemandshavebeenovercomeACSPublications5r©2017AmericanChemicalSocietytoalargeextentbytherecentadvancesingraphicsprocessingunits(GPUs)1-5andmassivecomputeresourcesavailableonthecloud.Inaddition,decadesofworkfromacademicandindustryresearchlaboratoriesaroundtheworldhaveproducedforcefieldsandsamplingalgorithmsthatarecapableofpredictingrelativebindingfreeenergiesatalevelofaccuracynecessarytobeusefulindrugdiscovery.6-9Finally,automationtoolshaveaddressedmanyofthetime-consuminganderroipronestepsassociatedwithrunningRBFEsimulations.Thesefundamentaladvances,coupledwithworkflowautomation,l0J1haveenabledfreeenergysimulationstobeperformedinarigorous,high-throughputmodethatcanbereadilydeployedwithinstructurallyenableddrugdiscoveryprojects.12Furthermore,withthebroadavailabilityofopen-source(eg,OpenMM13andGromacs143"),academic(e.g.,AMBER,16CHARMM,1andNAMDIS),andcommercialcodes(AceMD2Received:September19,2017Published:December15,2017andDesmond19),itispossibletomaximallyleveragethelargenumberofgraphicalprocessingunits(GPUs)t