Tutorial:RosettatoolsforstructuredeterminationincryoEMdensityBrandonFrenz,RayY.-R.Wang,FrankDiMaioLastupdated:August2018

ThistutorialisintendedtointroduceuserstoseveraldifferentwaysRosettamaybeusedtosolvevariousstructuredeterminationtasksgiven3-5ÅcryoEMdensitydata.Itisnotintendedtoreplacetheuser’sguide,availableathttps://www.rosettacommons.org/manuals/latest/main/.

Thetutorialissplitupintofourparts.

1. AnintroductiontoRosettaingeneral,showinghowonemayscorestructuresandminimizestructuresguidedbyexperimentaldensitydata

2. Ourfragment-basedrefinementprotocolforrefinementagainst3-5ÅEMdensity3. Ourmodelrebuildingprotocol(RosettaCM),whereonewishestorecombinehomologous

structures,andrebuildsmallmissingregions(<12residues)4. Ourdenovomodel-buildingtools,whereonewishestorebuildmissingregionsofastructure(for

example,fromahomologymodel5. Ourmodelcompletiontools,whereonewishestocompleteapartialmodelbuiltbythedenovo

toolorwishestorebuildlargemissingregions(12ormoreresidues)

Ineachscenario,wepresentthemostbasicusageofRosettaforthetask,andthendescribeadditionaloptionsthatmaybeuseful.Command-lineflagsandinputscriptsareprovidedinshadedboxes,withboldfacedtextindicatingparametersofnote.Theseparametersaredescribedinthetextfollowingthecommandline.

Note:inallsections,youwillneedtoupdatethecommandscriptstopointatyourinstallationofRosettaandtheRosettadatabase.

Whichsectionshouldoneread?

• IfyouwantastraightforwardintroductiontoscoringandbasicrefinementofstructuresinRosetta,readSection1

• Ifyouhaveamodelthatyouwanttorefineintoa3-4.5Ådensitymap,readSection2.• Ifyouhaveadensitymapat3-4.5Å(orworse!),oneormorepartialmodels,andyouwantto

combinethemodelsandrebuildshortinsertionsanddeletions,readSection3.• Ifyouhavea3-4.5Ådensitymapwithnohomologyinformationavailable,andwanttobuilda

model,readSection4.• Ifyouhavea3-4.5Ådensitymapandaveryincompletemodelyouwanttocomplete,readSection5.

Othernotes.

Foralltheapplicationsinthistutorial,itisrecommendedthatyoudownloadthelatestweeklyreleaseofRosetta.

Also,forbrevity,someofthecommandlinesandXMLscriptshavebeentrimmedinthisdocument.ThetutorialfilescontainthefullcommandlinesandXMLscripts;ifsomethingisomittedinthisdocument,itshouldnotbechangedfromthevalueinthetutorialfile.

1)Rosettaandelectrondensitybasics

ThissectionprovidesabriefintroductiontousingRosetta,andanoverviewofusingdensitydatawithinRosetta.

OverviewofRosetta

TheRosettadocumentationisagoodsourceofadditionalinformationonseveralofthetoolsdescribedinthisdocument.Thisisavailableathttps://www.rosettacommons.org/docs/latest/Home.

ThetoolsdescribedinthisdocumentusetheRosettaScriptsframework,describedathttps://www.rosettacommons.org/docs/latest/scripting_documentation/RosettaScripts/RosettaScripts.Briefly,thisallowsprotocolstobedefinedasaseriesofatomic"Movers"whichmanipulateastructure.Theformatisasfollows:

<ROSETTASCRIPTS> <SCOREFXNS> </SCOREFXNS> <MOVERS> </MOVERS> <PROTOCOLS> </PROTOCOLS> <OUTPUT /> </ROSETTASCRIPTS>

Eachblockcontainsinformationinrunningtheprotocol:<SCOREFXNS>and<MOVERS>areusedtodeclarescorefunctionsandmovers;while<PROTOCOLS>iswherethestepsoftheprotocolareenumerated.

Rosettatoolsarerunviathecommandline,withflagscontrollinggeneralprogrambehavior.Manyoftheflagsspecificallyfordensityrefinementareoutlinedinthesectionsfollowing.

DensityscoringinRosetta

AgreementtodensityisimplementedinRosettaasanadditionalenergyterm.Rosettaassessesagreementtodensitybycomputingthedensitythatonewouldexpecttosee,givenamodel,andmeasuringtheagreementoftheexpectedandexperimentaldensity.

elec_dens_fastThisscoretermisrecommendedfornearlyallusesofdensityrefinementinRosetta.Itusesinterpolationonaprecomputedgridofper-atomscorestoapproximatethedensitycorrelations.Thisversionissignificantlyfaster(~10x)thentheexactscoringtermbelow,andisveryhighlycorrelated.

TheseenergytermsmaybeprovidedtoRosettaintwoways.First,itmaybeprovidedinaRosettaScriptXMLfileasinput:

<Reweight scoretype="elec_dens_fast" weight="35.0"/>

Fornon-Rosettascriptapplications,thefollowingflagcontrolsthedensityscoringfunctionweight:

-edensity:fast_dens_wt 35.0 Therecommendedweightsforeachofthesetermsvarydependingonthedensitymapresolution,starting

modelquality,andprotocol.Section2describeshowtheweightsmaybetuned.However,thefollowingaregoodrulesofthumbforsettingthedensityweightwithinRosetta:Atresolutionsbetterthan2.5Å:anelec_dens_fastweightof65.0isgenerallyreasonable.Atresolutionsbetween2.5Åand3.5Å:anelec_dens_fastweightof50.0isgenerallyreasonable.Atresolutionsworsethan3.5Å:anelec_dens_fastweightof35.0isgenerallyreasonable.Incentroidmode(describedinpart4):anelec_dens_fastweightof10.0isgenerallyreasonable

Atverylowresolutions(worsethan6Å),theweightmayneedtobefurtherreduced.Ingeneral,iftheRosettaenergiesarepositive(orsignificantoutliersareflaggedbyMolprobityorothervalidationprograms)thentheweightsneedtobereduced.

Inadditiontothescoretermsabove,therearealsoseveralflagsthatcontrolmapscoringbehavior.MapsarereadintoRosettausingeithertheflag:

-edensity::mapfile mapfile.mrc

OrfromXML:

<LoadDensityMap name="loaddens" mapfile="mapfile.mrc"/>

MapsmaybeineitherCCP4orMRCformat(themaptypeisautomaticallydetectedfromtheheaderinfo).

Theresolutionofthemap,usedwhencomparingcalculatedtoexperimentaldensity,isspecifiedwiththeflag:

-edensity::mapreso 5.0

Mapsmayalsoberesampledtoreducememoryusageandruntime.Thisisdonethroughtheflag:

-edensity::grid_spacing 2.0

Noticethatthisflagshouldneverbemorethanhalfthegivenresolution,andifusingthefastscoringfunctionnevermorethanathirdoftheresolution.Forbothparameters,thedefaultisgenerallyfine(don’tresample,andassumetheresolutionis~3xthegridsampling).

Finally,onemaychoosetocalculatedensityusingeithercryoEMorX-rayscatteringfactors.Atlowresolution,thisprobablymakeslittledifference,butmightatresolutionsbetterthanabout3.5Å.ThedefaultistouseX-rayscatteringfactors;toturnoncryoEMscatteringfactorsinstead,usethefollowingflag:

-edensity::cryoem_scatterers

Example1A:ScoringaPDBinRosettawithdensity

Mostsimply,onemaywishtosimplyscoreamodelusingRosetta’senergyfunctionincludingthedensityterms.Thisiseasilyaccomplishedusingthescore_jd2application.Asamplecommandlinetorescorethestructureindensityisgivenin1_rosetta_basics/A_run_rescore.sh.ItillustratestheuseofvariousdensityflagstoprovideRosettawithexperimentaldensityinformation.

$ROSETTA3/source/bin/score_jd2.macosclangrelease \ -database $ROSETTA3/database/ \ -in::file::s 1isrA.pdb 1issA.pdb \ -ignore_unrecognized_res \ -edensity::mapfile 1issA_6A.mrc \ -edensity::mapreso 5.0 \ -edensity::grid_spacing 2.0 \ -edensity::fastdens_wt 35.0 \ -edensity::cryoem_scatterers \ -crystal_refine

Someflagsofnoteareboldfacedabove.First,theinputstructureisprovidedwiththecommand-in::file::s.ThisiscommontomanyRosettaapplications,andmorethanoneinputmaybeprovided;eachwillbeprocessedindependently.Theflagsbeginningwith–edensity::tellRosettaaboutthedensitymapintowhichitisbeingfit.Thenameofthemapfile(inCCP4orMRCformat),theresolutionofthemap,thegridsamplingofthemap(whichshouldneverbemorethanhalftheresolution),andtheweightsonthevariousfit-to-densityscoringfunctions.Thesesameflagsarereusedformanydifferentprotocolsinadditiontorelax.Finally,theflag-crystal_refinetheflagturnsonseveraldensity-relatedoptionsrelatedtoPDBreadingandwriting,andshouldalwaysbeusedwhenrefiningagainstdensitydata.

Note:TheinputPDBmustbealignedtothedensitymapusingsomeexternaltool.Rosettawilloptionallyrigid-bodyminimizethestructureintodensitybeforerescoringbyprovidingtheflag–edensity::realignmintotheapplication.Ifthisisdone,theflag–out::pdbwillwritetheminimizedPDBfiletoaPDBfile.

Thiscommandlineoutputsascorefile,score.sc,thatgives,foreachstructurespecifiedwith-in::file::s,thescorewithrespecttoeachterminRosetta’senergyfunction.ThemeaningofindividualscoretermsaswellasanoverviewoftheRosettaenergyfunctioncanbefoundinthepaper:

AlfordRF,Leaver-FayA,JeliazkovJR,O'MearaMJ,DiMaioFP,ParkH,ShapovalovMV,RenfrewPD,MulliganVK,KappelK,LabonteJW,PacellaMS,BonneauR,BradleyP,DunbrackRLJr,DasR,BakerD,KuhlmanB,KortemmeT,GrayJJ.TheRosettaAll-AtomEnergyFunctionforMacromolecularModelingandDesign.JChemTheoryComput.2017Jun13;13(6):3031-3048.

Example1B:SimplerefinementintodensityusingRosettaScriptsandrelax

InthissectionweintroduceRosettaScriptsbywayofaverysimplerefinement-into-densityexample.RosettaScriptsprovidesanXMLscriptinginterfacetoRosettathatallowsfine-grainedcontrolofprotocols.ThesyntaxisfullydescribedintheRosettadocumentation;however,averybriefintroductionisprovidedhere.ThebasicsyntaxfortheXMLisillustratedhere(1_rosetta_basics/B_relax_density.xml)

<ROSETTASCRIPTS> <SCOREFXNS> <ScoreFunction name="dens" weights="beta_cart"> <Reweight scoretype="elec_dens_fast" weight="35.0"/> <Set scale_sc_dens_byres="R:0.76,K:0.76,E:0.76,D:0.76,M:0.76, C:0.81,Q:0.81,H:0.81,N:0.81,T:0.81,S:0.81,Y:0.88,W:0.88, A:0.88,F:0.88,P:0.88,I:0.88,L:0.88,V:0.88"/> </ScoreFunction> </SCOREFXNS> <MOVERS> <SetupForDensityScoring name="setupdens"/> <LoadDensityMap name="loaddens" mapfile="1issA_6A.mrc"/> <FastRelax name="relaxcart" scorefxn="dens" repeats="2" cartesian="1"/> </MOVERS> <PROTOCOLS>

<Add mover="setupdens"/> <Add mover="loaddens"/> <Add mover="relaxcart"/> </PROTOCOLS> <OUTPUT scorefxn="dens"/> </ROSETTASCRIPTS>

Therearethree"blocks"ofdeclarationsinthisscript.Inthefirst,<SCOREFXNS>…</SCOREFXNS>,thescorefunctionstobeusedthroughouttheprotocolaredeclared;thesecond,<MOVERS>…</MOVERS>,movers–oratomicoperationsthatmodifyastructure–aredeclared;finally,thethird,<PROTOCOLS>…</PROTOCOLS>,afullprotocolisdeclaredasasequenceofmovers.

Inthisparticularexample,wedeclareasinglescorefunction,dens,whichusesthescorefunctionbeta_cart(adefaultscorefunction,don’tneedtoworryaboutit),andturnsonelec_dens_fast,thefit-to-densityscore,withaweightof35.Wethendeclarethreemovers,SetupForDensityScoring,LoadDensityMap,andFastRelax,whichsetsuptheloadedstructurefordensityscoring,loadsamapintomemory,andthenrefinesthestructureusingtheFastRelaxprotocol.Thedeclaredscorefunction,dens,isusedasaninputtotheFastRelaxmover.

Torunthisscript,weusethefollowingcommandline(1_rosetta_basics/B_relax_density.sh):

$ROSETTA3/source/bin/rosetta_scripts.macosclangrelease \ -database $ROSETTA3/database/ \ -in::file::s 1isrA.pdb \ -parser::protocol ex_B1_run_RS_relax_density.xml \ -ignore_unrecognized_res \ -edensity::mapreso 5.0 \ -edensity::cryoem_scatterers \ -crystal_refine \ -out::suffix _relax \ -beta

Note:Wedonothavetospecifythedensityweightorthemapfileonthecommandline,sincetheyarehandledwithintheXMLfile.However,otherdensityoptionsmustbespecifiedonthecommandline.WhenusingRosettaScripts,thedensityweightsmustbespecifiedintheXML,theinputmapmaybespecifiedeitherway.

Finally,inthepreviousXMLfile,thetagcartesian=1appears,whichrefinesthestructureinCartesianspace.Rosettaalsoallowsrefinementintorsionalspace,whichmaybebetterforcapturingdomainmotion,andforfurtherreductioninmodelparametersagainstlow-resolutiondata.Toenabletorsionalrefinement(1_rosetta_basics/C_relax_tors_density.xml),wemakethreesmallchangestotheXML:

<ROSETTASCRIPTS> <SCOREFXNS> <ScoreFunction name="dens" weights="beta"> <Reweight scoretype="elec_dens_fast" weight="35.0"/> <Set scale_sc_dens_byres="R:0.76,K:0.76,E:0.76,D:0.76,M:0.76, C:0.81,Q:0.81,H:0.81,N:0.81,T:0.81,S:0.81,Y:0.88,W:0.88, A:0.88,F:0.88,P:0.88,I:0.88,L:0.88,V:0.88"/> </ScoreFunction> </SCOREFXNS> <MOVERS> <SetupForDensityScoring name="setupdens"/> <LoadDensityMap name="loaddens" mapfile="1issA_6A.mrc"/> <FastRelax name="relaxcart" scorefxn="dens" repeats="5" cartesian="0"/> </MOVERS> <PROTOCOLS> <Add mover="setupdens"/> <Add mover="loaddens"/>

<Add mover="relaxcart"/> </PROTOCOLS> <OUTPUT scorefxn="dens"/> </ROSETTASCRIPTS>

2)Modelrefinementviaiterativelocalrebuilding

Inthissection,weintroduceourcryoEMrefinementprotocol,whichusesaniterativelocalrebuildingproceduretoescapelocalminimaduringrefinement.Thesectionisdividedintotwoparts,inthefirst,weintroducethemethodfornon-symmetricsystems;inthesecond,wedescribehowtousethismethodforsymmetricsystems.

Asarunningexample,werefinemodelsofthetransmembraneregionoftheTRPV1ionchannel,usinga3.4ÅcryoEMsingleparticlereconstruction(M.Liao,E.Cao,D.Julius,Y.Cheng,Nature,2013),andthedepositedmodel(id:3j5p)asastartingmodel.Wewillfirstrefinethisasymmetrically,andthenintroducesymmetricrefinement.

Example2A:AsymmetricrefinementintocryoEMdensity

AsummaryoftheXMLusedforrefinement(2_cryoem_refinement/A_asymm_refine.xml)isshownbelow.Following,abriefdescriptionofthemoversandoptionsavailableisprovided.

<ROSETTASCRIPTS> <SCOREFXNS> <ScoreFunction name="cen" weights="score4_smooth_cart"> <Reweight scoretype="elec_dens_fast" weight="20"/> </ScoreFunction> <ScoreFunction name="dens_soft" weights="beta_soft"> <Reweight scoretype="cart_bonded" weight="0.5"/> <Reweight scoretype="pro_close" weight="0.0"/> <Reweight scoretype="elec_dens_fast" weight="%%denswt%%"/> </ScoreFunction> <ScoreFunction name="dens" weights="beta_cart"> <Reweight scoretype="elec_dens_fast" weight="%%denswt%%"/> <Set scale_sc_dens_byres="R:0.76,K:0.76,E:0.76,D:0.76,M:0.76, C:0.81,Q:0.81,H:0.81,N:0.81,T:0.81,S:0.81,Y:0.88,W:0.88, A:0.88,F:0.88,P:0.88,I:0.88,L:0.88,V:0.88"/> </ScoreFunction> </SCOREFXNS> <MOVERS> <SetupForDensityScoring name="setupdens"/> <LoadDensityMap name="loaddens" mapfile="%%map%%"/> <SwitchResidueTypeSetMover name="tocen" set="centroid"/> <MinMover name="cenmin" scorefxn="cen" type="lbfgs_armijo_nonmonotone" max_iter="200" tolerance="0.00001" bb="1" chi="1" jump="ALL"/> <CartesianSampler name="cen5_50" automode_scorecut="-0.5" scorefxn="cen" mcscorefxn="cen" fascorefxn="dens_soft" strategy="auto" fragbias="density" rms="%%rms%%" ncycles="200" fullatom="0" bbmove="1" nminsteps="25" temp="4" fraglens="7" nfrags="25"/> <CartesianSampler name="cen5_60" automode_scorecut="-0.3" scorefxn="cen" mcscorefxn="cen" fascorefxn="dens_soft" strategy="auto" fragbias="density" rms="%%rms%%" ncycles="200" fullatom="0" bbmove="1" nminsteps="25" temp="4" fraglens="7" nfrags="25"/> <CartesianSampler name="cen5_70" automode_scorecut="-0.1" scorefxn="cen" mcscorefxn="cen" fascorefxn="dens_soft" strategy="auto" fragbias="density" rms="%%rms%%" ncycles="200" fullatom="0" bbmove="1" nminsteps="25" temp="4" fraglens="7" nfrags="25"/> <CartesianSampler name="cen5_80" automode_scorecut="0.0" scorefxn="cen" mcscorefxn="cen" fascorefxn="dens_soft" strategy="auto" fragbias="density" rms="%%rms%%" ncycles="200" fullatom="0" bbmove="1" nminsteps="25" temp="4" fraglens="7" nfrags="25"/> <ReportFSC name="report" testmap="%%testmap%%" res_low="10.0" res_high="%%reso%%"/>

<BfactorFitting name="fit_bs" max_iter="50" wt_adp="0.0005" init="1" exact="1"/> <FastRelax name="relaxcart" scorefxn="dens" repeats="1" cartesian="1"/> </MOVERS> <PROTOCOLS> <Add mover="setupdens"/> <Add mover="loaddens"/> <Add mover="tocen"/> <Add mover="cenmin"/> <Add mover="relaxcart"/> <Add mover="cen5_50"/> <Add mover="relaxcart"/> <Add mover="cen5_60"/> <Add mover="relaxcart"/> <Add mover="cen5_70"/> <Add mover="relaxcart"/> <Add mover="cen5_80"/> <Add mover="relaxcart"/> <Add mover="relaxcart"/> <Add mover="report"/> </PROTOCOLS> <OUTPUT scorefxn="dens"/> </ROSETTASCRIPTS>

Theprotocolisabitinvolved,butisdescribedinthefollowing.Thefirstthingtonoteistheuseofmacroslike"%%denswt%%".Thesearecommandargumentsthatmaybespecifiedfromthecommandlinethroughtheflag–parser::script_varsdenswt=25.0.Theprotocolaboveusesthesemacrosinplaceofparametersthatuserswouldmostliketochange;otherparametersshouldbeleftasisexceptforadvancedusers.

ThemainadditionistheCartesianSamplermover.Thismoveriterativelylocallyrebuildsthestructureinuser-specifiedorautomaticallydeterminedregions.Abriefdescriptionoftheargumentstothismover:

name="cen5_70":thenameofthemover,referredtointhe<PROTOCOLS>block(thiscanbeanything)strategy="auto":thestrategytouseinselectingwhattorebuild.Oneof:

• auto:selectregionsautomaticallybasedondensityfit&localstrain(usingthecutoffinautomode_scorecut,e.g.,automode_scorecut="-0.1")

• user:manuallyspecifyresidues(usingtheflagresidues,e.g.,residues="32A-48A")• rama:selectregionsautomaticallybasedonramascoreonly

rms="1.5":thecutoffonsimilaritywhenlocallyrebuilding.Increasingthisvaluewillincreasemodel

diversity(allowingworsestartingmodelstoberefinedbutrequiringadditionalsampling)ncycles="200":thenumberofrebuildingcyclestoconsider.Increasingthiswillincreaseruntimeand

slightlyincreasemodeldiversity.fraglens="7":thesegmentlengthtoreplace.Thismustbeanoddnumberfrom5-13,andincreasingthis

valuewillincreasemodeldiversitysignificantly.

Theremainingoptionsshouldneverbechanged.

Anotheroptionispassedtothedensityscoringviathe<Setscale_sc_dens_byres=.../>tag.Intherefinementprotocol,thissetsaper-amino-acidsidechainreweighing.Theweightsshowninthisexampleweredeterminedbyfittingtheseparametersintorefinedstructuresintoseveral3-5ÅcryoEMdensitymaps;theendresultisaslightdownweighingofsidechaindensity,particularlyforchargedsidechains.Thisshouldnotbechangedexceptbyadvancedusers.

TheMinMoverfirstminimizesthestructureusingalow-resolutionenergyfunction(cen).Wehavefoundthisstepismostusefulforimprovingproteinbackbonegeometry,particularlywithhand-tracedmodels.Thislow-resolutionscorefunctionusesthecentroidrepresentation,whichisenabledbytheSwitchResidueTypeSetmover.

TheFitBFactorsmoverfitsreal-spaceatomicBfactorstomaximizemodel-mapcorrelation.AconstraintenforcingnearbyatomstotakethesameBfactorsisalsoemployed,andtheweightonthistermiscontrolledwiththewt_adpterm(0.0005isgenerallywell-behaved).Finally,init=1meanstodoaquickscanofoverallBfactorsbeforebeginningrefinement;ifthereismorethanonecalltothismoverinasingletrajectory,thenonlythefirstneedstohaveinit=1.Exact=1shouldalwaysbeused.

Finally,theReportFSCmoverassessesmodelagreementtothemapusedforfittingaswellasanindependentmapusingtheintegratedFSCoverhigh-resolutionshells.Wehavefoundintegratingfrom10Åtotheresolutionofthedataisbestformodeldiscrimination.

Finally,thiscommandisexecutedusingthefollowing(scenario2_cryoem_refinement/A_asymm_refine.sh):

$ROSETTA3/source/bin/rosetta_scripts.macosclangrelease \ -database $ROSETTA3/database/ \ -in::file::s 3j5p_transmem_A.pdb \ -parser::protocol A_asymm_refine.xml \ -parser::script_vars denswt=35 rms=1.5 reso=3.4 map=half1_34A.mrc testmap=half2_34A.mrc \ -ignore_unrecognized_res \ -edensity::mapreso 3.4 \ -default_max_cycles 200 \ -edensity::cryoem_scatterers \ -beta \ -out::suffix _asymm \ -crystal_refine

Inboldaretheparametersthatshouldbechangedinadaptingthisrunforothersystems.Thefirstistheinputstructure,whichshouldbespecifiedwiththeargument–in::file::s.Thesecondaretheparameterstobepassedthroughtothescript(usingthemacroreplacementmachineryofRosettaScripts).Threeofthesedescribetheinputmaps:

map=half1_34A.mrc–themaptorefineagainsttestmap=half2_34A.mrc–anindependentmapforvalidation

(ifnotusingsplitmaps,justprovidethesamemapasthepreviousargument)reso=3.4–theresolutionofthedata

(notethatthisneedstobeprovidedtwiceinthecommandline,onceforscoringandonceforreporting)

Theothertwoareparameterstothealgorithm:denswt=35–theweightontheexperimentaldensitydatarms=1.5–theamountofdeviationtoallowinfragmentinsertionmoves

(largervalueswillleadtomoremodeldeviation)

Thedensityweightof25worksreasonablywellasastartingpoint,butonemightwanttoexploreseveraldifferentvaluesusinganindependentreconstruction.Manualinspectionofoutputmodelsformolprobityscore,freeFSC,and(freeFSC–workFSC)shouldprovidecluesastowhichweightworksbest.

Adescriptionofmuchoftheworkinthissectionisdescribedinthereference:

WangRY,SongY,BaradBA,ChengY,FraserJS,DiMaioF.Automatedstructurerefinementofmacromolecularassembliesfromcryo-EMmapsusingRosetta.Elife.2016Sep26;5.pii:e17219.

JobdistributionItisgenerallyusefultosample~100modelsfromeachstartingpoint.Forthispurpose,itmaybeusefultorunmultiplejobsinparallel.Topreventoutputstructuresfromclobberingoneanother,theflag–out::suffixmaybeuseful,whereeachseparatejobisgivenadifferentsuffix.

Forexample,ona16-coremachine,wemayspecify-out::suffix_$1,then(usingGNUparallel)runthefollowing:

parallel –j16 ./A_asymm_refine.sh {} ::: {1..16}

Finally, GNU parallel allows launching of jobs remotely if SSH keys have been set up for passwordless login. To run:

parallel –S 16/node1,16/node2,16/node3,16/node4 –-workdir . ./B1_rosettaCM_singletarget.sh {} ::: {1..48}

Thiswilllaunchinstead48jobsspreadacrossfourmachines.SeetheGNUparalleldocumentation(https://www.gnu.org/software/parallel/)formoreinformation.

AnalyzingresultsandmodelselectionWhilethisisanactivetopicofresearch,generally–onceadensityweighthasbeenchosen–toselectthebestmodelsfromamongthefullset,wewanttoselectmodelsoptimizingbothmodelgeometryandfreeFSCvalues.Modelgeometrymaybeevaluatedusingeither:a)Molprobity,orb)Rosettaenergiesaftersubtractingdensityenergies.Thelattermaybedonebyinspectingthescore*.scfilesproducedasoutput.

Usingtheabovescript,thefreeFSCvaluemaybedeterminedfromtheoutputPDBfiles.Theheadercontainsalinelike:

REMARK 1 FSC[mask=4.45657](10:3) = 0.590966 / 0.591017

Thetwonumbersattheendofthelineindicatethe"work"and"free"integratedFSCvalues.

Generally,weselectthebest20%ofmodelsbygeometry,andselectingthebestoverallbyfreeFSC.Thetop5modelsshouldbeinspectedformodelconvergenceaswellasvisuallyinspectedfordensitymapagreement.

Example2B:SymmetricrefinementintocryoEMdensity

Asthisisasymmetricsystem,tocorrectlyevaluatetheenergeticsofthesystem,weneedtomodelwithsymmetry-relatedcopiespresent.Thismaybedonethroughatwo-stepprocess:first,werunthemake_symmdef_file.plscripttoprepareadescriptionofthesymmetryofthesysteminaRosetta-readableformat.Next,weenablesymmetricscoringandoptimizationwithintheXMLfile.

TheinformationthatRosettaneedstoknowaboutasymmetricsystemisencodedinthesymmetrydefinitionfile.IttellsRosetta:(a)howtoscoreastructuresymmetricallyfromonlyasymmetricunitinteractions,and(b)howtherigid-bodydegreesoffreedomareallowedtomovetomaintainthesymmetryofthesystem.

Toaidincreatingasymmetrydefinitionfilefromasymmetric(ornear-symmetric)PDB,anapplication,

make_symmdef_file.pl,hasbeenincludedinsrc/apps/public/symmetry.TogeneratethesymmetrydefinitionfileforTRPV1,werunthecommandin2_cryoem_refinement/B1_make_symmdef.sh.

$ROSETTA3/source/src/apps/public/symmetry/make_symmdef_file.pl \ -m NCS -a A -i B \ -p 3j5p_transmem.pdb –r 1000 > TRPV1.symm

Thisscriptneedsafewpiecesofinformation:with–m,thetypeofsymmetrytogenerate(hereNCS),with–a,theprimarychain(hereA),andwith–i,anadjacentchainineachsymmetrygroup,separatedbyspaces(herejustB).ForCnsymmetries,onlyoneadjacentchainisgiven;forDn,twoaregiven.Finally,with–r,wegivethecontactdistancebetweenaneighborchainandtheprimarychainnecessarytoincludethatsubunitexplicitly(here,1000,toensureeverysymmetricallyrelatedcopyisincluded).Iftheinputsystemisasymmetric,thescriptwillmakeasymmetricalversionofit(sometimessignificantlyperturbingitintheprocess).Therearealotofotheroptions,includingforcingsymmetricalorderandhelicalandhigher-ordersymmetries,seethedocumentation!

InadditiontothedefinitionfilewrittentoSTDOUT,thescriptwillalsowriteafile3j5p_transmem_symm.pdb,containingthesymmetrizedversionoftheinputfile,andafile3j5p_transmem_INPUT.pdb,thatcontainsonlythemainchain,tobeusedasinput(inadditiontothesymmetrydefinitionfile).

Thesymmetrydefinitionfilelookssomethinglikethis:

symmetry_name TRPV1__4 E = 4*VRT0_base + 4*(VRT0_base:VRT3_base) + 2*(VRT0_base:VRT2_base) anchor_residue CoM ... set_dof JUMP0_to_com x(11.7023996817515) set_dof JUMP0_to_subunit angle_x angle_y angle_z ...

Theomittedsectionsdescribeasystemofvirtualresiduesthatmaintainthesymmetryofthesystem,andtheygenerallyshouldremainunedited.

Thetwoset_doflinesshouldbeedited.Therearetwopossibilitieswhenrefiningsymmetricstructuresintodensity:

A)wedon’twanttorefinetherigidbodyorientationoftheentiresystem;weknowthesymmetryaxesandwedon’twantthemtomove

B)wedowanttorefinetheorientationoftheentiresystem,includingsymmetryaxes

Generally,incryoEM,wherethemapsaresymmetricallyaveraged,andthesymmetryisknown,wewanttousestrategyA.However,insomecases(forexample,ifourstartingmodelwasnotperfectlysymmetric)wewanttousestrategyB.Inbothcases,aminoredittotheset_doflinesinthesymmdeffileisnecessary.

ForstrategyA,becauseweareusingdensity,weneedtochangethefirstset_doflinetothefollowing:

set_dof JUMP0_to_com x y z

ForstrategyB,weleavethetwolinesunchangedandinsteadaddathirdline:

set_dof JUMP0 x y z angle_x angle_y angle_z

ForDnsymmetries,thechangesaresimilar,exceptinAthejumpnameisJUMP0_0_to_com.TherestofthissectionusesstrategyA;theeditedsymmetrydefinitionfileisinscenario1_cryoem_refinement/C4_edit.symm

Onceasymmetrydefinitionfilehasbeengenerated,thenrefiningstructuresinRosettasymmetricallyisstraightforward.ThechangestothepreviousXMLfileareindicatedbelow(see2_cryoem_refinement/B2_symm_refine.xml):

... <ScoreFunction name="cen" weights="score4_smooth_cart" symmetric="1"> <ScoreFunction name="dens_soft" weights="eta_soft" symmetric="1"> <ScoreFunction name="dens" weights="talaris2013_cart" symmetric="1"> ... <SetupForSymmetry name="setupsymm" definition="%%symmdef%%"/> ... <SymMinMover name="cenmin" scorefxn="cen" type="lbfgs_armijo_nonmonotone" max_iter="200" tolerance="0.00001" bb="1" chi="1" jump="ALL"/>

Inallthreedeclaredscorefunctions,symmetric=1mustbegiven.Additionally,theSetupForDensityScoringmovermustbereplacedwiththeSetupForSymmetrymover.Finally,theMinMovermustbereplacedwithitssymmetriccounterpart,SymMinMover.

Thecommandforrunningthisscriptislargelythesameasbefore,withafewadditions:

$ROSETTA3/source/bin/rosetta_scripts.macosclangrelease \ -database $ROSETTA3/database/ \ -in::file::s 3j5p_transmem_A.pdb \ -parser::protocol A_asymm_refine.xml \ -parser::script_vars \ denswt=25 rms=1.5 reso=3.4 map=half1_34A.mrc \ testmap=half2_34A.mrc symmdef=TRPV1_edit.symm \ -ignore_unrecognized_res \ -score_symm_complex false \ -edensity::mapreso 3.4 \ -default_max_cycles 200 \ -edensity::cryoem_scatterers \ -beta \ -out::suffix _symm \ -crystal_refine Thecommandsymmdef=TRPV1_edit.symmpassesthesymmetrydefinitionfiletoRosetta.Theflag-score_symm_complexfalsedependsonwhetherstrategy(a)or(b)wasemployedabove.If(a),thenfalseshouldbeused;if(b),thentrueshouldbeused.Note:TheinputPDB(-in::file::s)isofthemonomer(thatis,theasymmetricunit).

3)Modelrebuildingguidedbyexperimentaldensitydata

Inthisscenario,weintroduceatool,RosettaCM,forbuildingmissingportionsofamodelguidedbydensitydata.Whileprimarilygearedtowardscomparativemodeling,itmayalsobeusefulforbuildingportionsofaproteinthataredisorderedwhencrystallizedordifficultregionsinhand-builtmodels.Inthisscenario,weintroducethebasicrebuildingprotocol,thenshowhowthetoolmayalsobeusedto:

• Combinepiecesfrommultipletemplatemodelsguidedbydensity• Rebuildwithuser-definedrestraints• Iterativelyrebuildmodelsindifficultcasesdifficultcases

Asarunningexample,weusethe20Sproteasome(XuemingLietal.,NatureMethods,2013),whereonlyasubsetofparticleswereused,resultingina4.1Åreconstruction.Wearebuildingmodelsstartingfromahomologousstructure(pdbid:1iru)asthestartingmodel(25%/32%sequenceidentitytochainsA/B).

Example3A:PreparingtemplatesforuseinRosettaCM

Inmanycases,muchofthesetupworkishandledbyascript,setup_RosettaCM.pyinRosettaTools(aseparaterepositoryavailablefromrosettacommons.org).Thisscripttakesaninputalignmentinavarietyofformats,andpreparestheinputsautomatically.Itisexecutedbyrunningthecommand:

setup_RosettaCM.py \ -–fasta t20s.fasta \ --alignment tmpl.fasta \ --alignment_format fasta \ --templates tmpl.pdb \ --rosetta_bin ~/Rosetta/main/source/bin \ --verbose

Inputsincludethefull-lengthfasta,analignmentfile–ineitherfasta,ClustalW,orHHSearchformat–andthecorrespondingtemplatePDBfiles.ThisscriptwillprepareallthenecessaryinputsinordertorunRosettaCM.

Alternately,thesetupmaybeperformedmanually.Inthiscase,sinceweareusingsomenonstandardfeatures(symmetryanddensity),andwehavetwochainsintheasymmetricunitwewilldothis;alternately,theinputsfromthepreviousstepmaybeusedasastartingpointandsubsequentlymodified.Inthiscase,wefirstconvertouralignmenttoRosettaformat(scenario2_model_rebuilding/20S_1iru.ali):

## 1XXX_ 1iruAH_thread # hhsearch scores_from_program: 0 1.00 0 TVFSPDGRLFQVEYAREAVKK-GSTALGMKFANGVLLISDKKVRSRLIEQNSIEKIQLIDDYVAAVTSGLVADAR... 0 TIFSPEGRLYQVEYAFKAINQGGLTSVAVRGKDCAVIVTQKKVPDKLLDSSTVTHLFKITENIGCVMTGMTADSR... --

Inthisformat,thefirstlineis'##'followedbyacodeforthetargetandoneforthetemplate.Thesecondlineidentifiesthesourceofthealignment;thethirdjustkeepasitis.Thefourthlineisthetargetsequenceandthefifthisthetemplate;thenumberisan'offset',identifyingwherethesequencestarts.However,thenumberdoesn'tusethePDBresidbutjustcountsresiduesstartingat0.Thesixthlineis'--'.Multiplealignmentsmaybeconcatenatedinasinglefile,withthetemplatecodeidentifyingthetemplatecorrespondingtoeachalignment.

RosettaCMtakesasinputspartiallythreadedmodels,thatismodelswherealignedpositionshavetheir

residueidentitiesremapped,andunalignedresiduesarenotpresent.Togeneratethesemodelsfromanalignmentfileandtemplate,wecanruntheRosettacommand(3_model_rebuilding/A_partialthread.sh):

$ROSETTA3/source/bin/ partial_thread.macosclangrelease \ -database $ROSETTA3/database/ \ -in::file::fasta t20s.fasta \ -in::file::alignment 20S_1iru.ali \ -in::file::template_pdb 1iruAH_aln.pdb

Thiswilloutputapartiallythreadedmodelin1iruA_thread.pdbthatiscorrectlynumberedforinputintoRosettaCM.

Next,weneedtosetupsymmetricmodelingwithRosettaCM.AsinScenario1,weusethemake_symmdef_file.plscriptinordertogenerateasymmetrydefinitionfileforuseinRosetta.AstraightforwardwaytodosoistouseChimeratodockthenecessarychainsintodensity.Weneedasingle"primarychain"andacoupleofanadjacentchainineachpointgroup;sincetheproteasomefeaturesD7symmetry,thatmeansweneedanadjacentchaininthe7-foldcomplex,aswellasachainintheoppositering.Anexamplehasbeencreatedinscenario2_model_rebuilding/setup_symm.pdbwherethreecopiesofthethreadedmodelhavebeendockedintodensitywithChimera.TogenerateourD7symmetryfilefromthisinput,wethensimplyhavetorunthecommand(3_model_rebuilding/B_make_symmdef.sh):

~/rosetta_source/src/apps/public/symmetry/make_symmdef_file.pl \ -m NCS -a A -i B C \ -p setup_symm.pdb –r 1000 > D7.symm

Sincewehavealreadycreatedtheinputtemplatesusingthepartial_threadapplication,wecanignorethesetup_symm_INPUT.pdbfileandusetheoutputofpartialthreadastheinput.However,westillneedtoalignallthethreadedmodelstothisinputstructure.ThiscaneitherbedonewiththeprogramTMalign(externaltoRosetta)orbyusingChimeratodocktheindividualthreadedmodelsintodensity.Inthiscase,wherewehavejustonetemplate,ithasalreadybeenalignedtothetemplateinscenario2_model_rebuilding/tmpl_thread_aln.pdb.

AsinScenario1,weneedtomakeasmalledittothesymmetrydefinitionfilefordensityrefinement.Changethefollowinglines:

set_dof JUMP0_0_to_com x(35.3434689631743) set_dof JUMP0_0_to_subunit angle_x angle_y angle_z set_dof JUMP0_0 x(39.2905097135684) angle_x

To(3_model_rebuilding/D7_edit.symm):

set_dof JUMP0_0_to_com x y z set_dof JUMP0_0_to_subunit angle_x angle_y angle_z

Note:The20Sproteasomecaseweareusingcontainstwochainsintheasymmetricunit.TospecifythisasinputstoRosettaCM,weneedtolistthefastafile,separatingthechainsbytheslashcharacter‘/’.ThisisreallyonlynecessaryinthefastaprovidedasinputtoRosettaCM(nextstep)however,thereisnoharmisdoingthisineverystep.

Example3B:RunningRosettaCMusingasingletemplatemodelasinput.

LikethemethodsintroducedinScenario1,RosettaCMiscontrolledthroughanXMLscriptusingRosettaScripts.TheXMLisasfollows(3_model_rebuilding/C_rosettaCM_singletarget.xml):

<ROSETTASCRIPTS> <TASKOPERATIONS> </TASKOPERATIONS> <SCOREFXNS> <ScoreFunction name="stage1" weights="score3" symmetric="1"> <Reweight scoretype="atom_pair_constraint" weight="0.1"/> <Reweight scoretype="elec_dens_fast" weight="10"/> </ScoreFunction> <ScoreFunction name="stage2" weights="score4_smooth_cart" symmetric="1"> <Reweight scoretype="atom_pair_constraint" weight="0.1"/> <Reweight scoretype="elec_dens_fast" weight="10"/> </ScoreFunction> <ScoreFunction name="fullatom" weights="beta_cart" symmetric="1"> <Reweight scoretype="atom_pair_constraint" weight="0.1"/> <Reweight scoretype="elec_dens_fast" weight="35"/> <Set scale_sc_dens_byres="R:0.76,K:0.76,E:0.76,D:0.76,M:0.76, C:0.81,Q:0.81,H:0.81,N:0.81,T:0.81,S:0.81,Y:0.88,W:0.88, A:0.88,F:0.88,P:0.88,I:0.88,L:0.88,V:0.88"/> </ScoreFunction> </SCOREFXNS> <FILTERS> </FILTERS> <MOVERS> <Hybridize name="hybridize" stage1_scorefxn="stage1" stage2_scorefxn="stage2" fa_scorefxn="fullatom" batch="1" stage1_increase_cycles="1.0" stage2_increase_cycles="1.0" linmin_only="0" realign_domains="0"> <Template pdb="1iruA_thread.pdb" weight="1.0" cst_file="AUTO" symmdef="D7_edit.symm"/> </Hybridize> </MOVERS> <PROTOCOLS> <Add mover="hybridize"/> </PROTOCOLS> </ROSETTASCRIPTS>

Themainworkisdonethroughasinglemover,Hybridizewhichhandlesallstagesofmodel-building.InputstructuresarespecifiedviaTemplatelines(inthiscasethereisonlyone).Foreachtemplateline,wespecifythepdbinput,aswellasacoupleofotherparameters:aweight(therelativefrequencywesampleeachtemplatewith);aconstraintfile(settingthisto"auto"setsupautomaticconstraintstothetemplate,whilesettingthisto"none"turnsoffallconstraints,user-definedconstraintsaredescribedlater);andan(optional)symmetrydefinitionfile.

Note:Besurethatyourtemplatesarealignedtothedensity!

GiventhisXML,RosettaCMisthenrunwiththefollowingcommandline(3_model_rebuilding/B1_rosettaCM_singletarget.sh):

$ROSETTA3/source/bin/rosetta_scripts.macosclangrelease \ -database $ROSETTA3/database/ \ -in:file:fasta t20s.fasta \ -parser:protocol C_rosettaCM_singletarget.xml \ -nstruct 50 \ -relax:jump_move true \ -relax:dualspace \ -out::suffix _singletgt \ -edensity::mapfile t20S_41A_half1.mrc \ -edensity::mapreso 5.0 \ -edensity::cryoem_scatterers \ -beta \ -default_max_cycles 200

Theinputcommandissimilartothoseseenbefore,butwithafewkeydifferences.First,theinputtoRosetta

isspecifiedwith-in:file:fastaratherthan-in:file:s.Alsonotethattheinputargument–nstruct50isgiven,tellingRosettatogenerate50modelsforeachprocess.Generally,hundredstothousandsofmodelsarenecessarytosufficientlysampleconformationalspace;moreandlongerregionstorebuildrequiremoremodels.

Runningwithoutsymmetry

RunningwithoutsymmetryrequiresonlytwosmallchangestotheXMLfile:• Removethetag:symmdef="D7_edit.symm"• Removethethreetagssymmetric=1

JobdistributionAswithsection2,acombinationof-out::suffixandGNUparallelisusefulfordistributingjobs.Forexample,onemayreplacethe–out::suffixlineabovewith–out::suffix_$1,thenlaunchjobswith:

parallel –j16 ./B1_rosettaCM_singletarget.sh {} ::: {1..16}

Thetotalnumberofstructuresgeneratedisthenumberofstructuresspecifiedwith–nstructtimesthenumberofjobslaunched(inthisinstance,50structuretimes16jobs=800structures).Dependingontheruntimeperstructure(variabledependingonstructuresize)andthenumberofCPUsavailable,bothofthesenumbersmaybeadjusted.

Finally,GNUparallelallowslaunchingofjobsremotelyifSSHkeyshavebeensetupforpasswordlesslogin.Torun:

parallel –S 16/node1,16/node2,16/node3,16/node4 -–workdir . ./B1_rosettaCM_singletarget.sh {} ::: {1..48}

Thiswilllaunchinstead48jobsspreadacrossfourmachines.SeetheGNUparalleldocumentationformoreinformation.r

Example3C:RunningRosettaCMusingmultipletemplatemodelsasinput.

OneofthestrengthsofRosettaCMisitsabilitytomakeuseofmultipletemplatestructures,andtorecombineportionsofthesemodelsduringconformationalsampling.Thisisparticularlyusefulwhenmultiplehomologousstructuresareavailable,somewithclosersequenceidentity,andsomewithmorecompletecoverage.Theprotocolallowsthecombinationoffeaturesofbothmodels.

Tomakeuseofthisfeature,wesimplyaddadditionaltemplatelinesintheinputXML.Inthiscase,weaddthetemplate1ryp(scenario3_model_rebuilding/C1_rosettaCM_multitarget.xml):

... <Hybridize name="hybridize" stage1_scorefxn="stage1" stage2_scorefxn="stage2" fa_scorefxn="fullatom" batch="1" stage1_increase_cycles="1.0" stage2_increase_cycles="1.0" linmin_only="0" realign_domains="0"> <Template pdb="1iruA_thread.pdb" weight="1.0" cst_file="auto" symmdef="D7_edit.symm"/> <Template pdb="1rypA_thread.pdb" weight="1.0" cst_file="auto" symmdef="D7_edit.symm"/> </Hybridize> ...

Therestishandledautomaticallybytheprotocol.However,thereareafewcaveatswhenusingmultiple

inputstructures:

• Withdensity,weneedtoensurethatallinputmodelsarealignedtothedensity.ThiscanbedoneusingeitherTMalignorChimera’salignmenttools.

• Ineachtrajectory,astartingmodelischosenatrandom;theconstraintsandsymmetryfromthisselectedmodelarechosenatthestartofeachrun.Ifwewishtouseaportionofamodel,butdonotwanttouseitssymmetryorconstraints,wecanassignitaweightof0:backboneconformationsfromthismodelwillbeusedinconformationalsampling,butthesymmetryandconstraintswillneverbeused.

• Similarly,gapsintheselectedstartingmodelarerebuiltbeforerecombinationoccurs.Ifoneofthetemplateshaspoorcoverage,butprovidesvaluablestructuralfeatures,itshouldbeused,butwithweight0.

Example3D:RunningRosettaCMwithuserspecifiedconstraints.

AnotherstrengthofRosettaCMistheabilitytomakeuseofadditionalexperimentalinformationthatprovidesrestraintsoverconformationalspace.Whilepreviously,wehaveusedcst_file=autotoautomaticallygenerateconstraintsfromtemplatestructures,ifexperimentsprovidedistanceconstraints(orsomeotherpositionalrestraint,wemaymakeuseoftheminmodelrebuildingaswell.

TheRosettadocumentationprovidesagoodoverviewofthetypesofconstraintsthatmaybeused,withanumberofdifferentconstrainttypesandfunctionalformspossible.Forthisdemo,wewillassumewehaveknowledgeonthedistancebetweenresidues107and143thatwewanttouseduringrebuilding.

Thisinformationcanbeencodedinaconstraintfile(scenario3_model_rebuilding/D1_constraints.cst):

AtomPair CA 107 CA 143 HARMONIC 5.0 1.0

Note:Thenumberingofresiduesisbasedupontheorderintheinputfastafile(anddoesnotresetbetweenchains!).

Wethenreplacethecst_file=autolinesintheXMLwithourownconstraintfile(scenario3_model_rebuilding/D1_constraints.xml):

... <Template pdb="tmpl_thread_aln.pdb" weight="1.0" cst_file="D1_constraints.cst" symmdef="D7_edit.symm"/> ...

Wecanthenrebuildandrefineasbefore.

Example3E:ModelselectionandrunningRosettaCMiteratively

Withpossiblyhundredsofgeneratedmodels,thereareafewstrategiestoidentifythebest-sampledmodels.Generally,modelsshouldbefilteredontwodifferentcriteria–thetotalscoreandthedensityscore–insomeway.Weoftenselectthebest10-20%ofmodelsbasedontotalscore,andthesortthesemodelsbydensityscore,butvisualinspectionofthebestbybothcriteriacanoftenbebeneficialindifficultcases.

Finally,onestrategyforsolvingdifficultstructuresistoapplyRosettaCMiteratively.Usingtheabovecriteria,wecanselectthebest5-10modelsfromthefirstroundofrefinement,andfeedthemasinputsintothenextround.Modelscanbeselectedbyenergybylookingatthescorecolumnintheoutput.scfiles.

ThisisverybrieflyillustratedinthefollowingXML(scenario3_model_rebuilding/E1_rosettaCM_iter.xml):

... <Hybridize name="hybridize" stage1_scorefxn="stage1" stage2_scorefxn="stage2" fa_scorefxn="fullatom" batch="1"> <Template pdb="expected_outputs/S_multitgt_0001_A.pdb" weight="1.0" cst_file="NONE" symmdef="D7_edit.symm"/> <Template pdb="expected_outputs/S_multitgt_0002_A.pdb" weight="1.0" cst_file="NONE" symmdef="D7_edit.symm"/> <Template pdb="expected_outputs/S_multitgt_0003_A.pdb" weight="1.0" cst_file="NONE" symmdef="D7_edit.symm"/> </Hybridize> ...

Thereisalsosomemanipulationofinputmodelsthatcanprovebeneficial.Ifonewantstoforcerebuildingsomesegmentofbackbone,theycansimplydeletethoseresiduesinallinputmodels.Similarlyifonewantstoforcesomeregiontoadoptaconformationtakinginoneinputmodel,theycandeleteallotherconformationsfromallmodels.

Example3F:RunningwithLigandsand/orNucleicAcids

RosettaCMcanberunwithligandsaswellasnucleicacids.WhilenucleicacidsarereadbyRosettanatively,ligandsrequireanadditionalinputtoRosetta,aparamsfile.

Forligands,amol2fileoftheligandwhichcontainshydrogenatomsisrequired.TheprogramOpenBabel(http://openbabel.org/wiki/Main_Page)iscapableofbothconvertingfromPDBtomol2aswellasaddinghydrogenstoamoleculegivenaPDBfileoftheligand.

Givenamol2file,molfile_to_params.py,includedinRosetta,at$ROSETTA3/source/scripts/python/public/willgenerateaparamsfile.Touse:

python $ROSETTA3/source/scripts/python/apps/public/molfile_to_params.py \ --keep-names --clobber –centroid XXX.mol2 -p XXX -n XXX

Replace"XXX"withtheligandsthreelettercode.ThisscriptwillcreateanXXX.paramsfileaswellasseveralotherfiles.ThisfilecanbepassedintoRosettausingtheflag:-extra_res_fa XXX.fa.params -extra_res_cen XXX.cen.params

Ifthereismorethanoneuniqueligand,runthescriptoneachuniqueligand,andpassalistofparamsfilesusingeachoftheflags.WhenmodellingwithligandsornucleicacidsinRosettaCM,twoadditionalthingsareneeded:1.TheligandornucleicacidsmustbeaddedtotheENDofeachtemplatefilewithaweight>02.AnadditionalflagneedstogetaddedtoHybridize:<Hybridize name="hybridize" stage1_scorefxn="stage1" stage2_scorefxn="stage2" fa_scorefxn="fullatom" batch="1" add_hetatm="1">

4)Denovomodel-buildingguidedbyexperimentaldensitydata

Inthisscenario,weintroduceatool,denovo_density,aimedatautomaticallybuildingbackboneandplacingsequencein3-4.5ÅcryoEMdensitymaps.Thistoolisprimarilyintendedforcaseswhereamodelistobebuiltwithnoknownstructuralhomologues.Itisrelativelyexpensivecomputationally,andconsistsoffourbasicsteps:

• Searchforlocalbackbone"fragments"inthedensitymap• Scorethe"compatability"ofsetsofplacedfragments• MonteCarlosamplingforthe"maximallycompatable"fragmentset• Consensusassignmentfromthebest-scoringMonteCarlotrajectories

Theinputofthemethodisasegmenteddensitymap,andthesequencecontainedinthisregion.Theoutputofthemethodisa"partialmodel"that–ideally–places70-80%ofthesequenceintothedensitymap.ThispartialmodelcanthenbeusedasinputforRosettaCM(seesection3ofthetutorial).

Thismethodisunderconstantdevelopment.Currentlimitationsoftheapproach–hopefullytobeaddressedinfuturerevisions–include:

• Thecodeassumessegmenteddensitymaps.Itcannothandlesymmetry,andpoorlyhandlesunsegmenteddensitymaps.Resultsarebestwhenthemapissegmentstoonlycontainonecopyoftheresiduesgettingbuilt.

• Ithasonlybeentestedbuildingproteins~600residuesorless.Whileitshouldconceptuallyscaletolargerproteins,thisisuntested.Furthermore,withlargerproteins,thememoryusageofsteps2and3increasessignificantly,socaremustbetaken.

• Itcurrentlydoesnotidentifyandbuildligandsornucleicacids

Thissectionofthetutorialwalksthroughthesefourstepsoftheprotocol.Asarunningexample,weagainusethe20Sproteasome(XuemingLietal.,NatureMethods,2013),inthiscaseusinga4.8Åreconstructiondeterminedwithoutusingmotioncorrection.Wewillpretendthatknownhomologousstructuresarenotavailable,andinsteadwillbuildmodelsintodensitydenovo.

Inputfilepreparation:downloadfragmentfilesfromRobetta

Beforerunningthemethod,ausermustfirstcreatea"fragmentfile"thatpredictslocalbackboneconformationsgiventheamino-acidsequence.Theeasiestwaytodosoistosubmityoursequenceathttp://robetta.bakerlab.org/.

Alternately,theRosettausersguidedescribeshowfragmentfilesmaybebuiltlocally:(https://www.rosettacommons.org/manuals/archive/rosetta3.5_user_guide/dc/d10/app_fragment_picker.html)

Step4A.Localfragmentsearch

Inthefirstpartoftheprocedure,wesearchthedensitymapforeachsequence-predictedbackbonefragment.Thispart,likeallthestepsinthissection,usesaRosettaapplicationdenovo_density.

Thecommandtorunfragmentsearchingforasingleresidue(4_denovo_demo/A_search.sh):

$ROSETTA3/source/bin/denovo_density.macosclangrelease \ -in::file::fasta t20sA.fasta \ -fragfile ./t001_.25.9mers \ -mapfile ./T20S_48A_alpha_chainA.mrc \ -n_to_search 500 -n_filtered 2500 -n_output 100 \ -bw 16 \ -atom_mask_min 2 \ -atom_mask 3 \ -clust_radius 3 \ -clust_oversample 4 \

-point_radius 3 \ -movestep 1 \ -delR 2 \ -frag_dens 0.8 \ -ncyc 3 \ -min_bb false \ -pos $1 \ -out:file:silent round1/t20s.$1.silent

Mostoftheargumentsshownhereshouldbeleftas-is.However,thereareafew–highlightedaboveinboldface–thatyoumightwanttochange:-n_to_search 500 -n_filtered 2500

Theseflagscontrolthenumberoftranslationstosearch,andthenumberofintermediatesolutionstokeep.Asaruleofthumb,theseshouldbeabout2and10timesthenumberofresiduesinthemap,respectively.

Makenoteofthefollowingflag:-pos $1

Thisflagtellsthecodetoonlysearchforfragmentsattheassignedpositions.Thisallowsforparallelizationofthescript,byrunningseparatejobsforeachpositionintheprotein.($1meansthescripttakesthepositionasaninputargument).

Forthisstep,youneedtorunthescriptonceforeachpositionintheprotein.Thiscanbedoneverysimplywiththebashcommand(forthiscase,the221indicatesthereare221residuesintheprotein):for i in `seq 1 221`; do ./A_search.sh $i; done

Theoutputofthisscriptisasinglefileforeachpositionintheprotein,thatidentifiestheplacementandconfigurationofeachdockedfragment.Thesefilesareusedasinputforthenextstepoftheprocess.

However,aseachpositionrunsindependently,andeachpositionmighttake30minutestoanhourforthesearch,youwillprobablywanttoparallelizethisovermanyprocessors.

Jobdistribution

Asthisisthemostcomputationallyintensivestep,itmakessensetoparallelizethisstep,byrunthisprocedureseparatelyforeachpositionintheprotein.UsingGNUparallel.parallel –j16 ./A_search.sh {} ::: {1..221}

GNUparallelallowslaunchingofjobsremotelyifSSHkeyshavebeensetupforpasswordlesslogin.Torun:

parallel –S 16/node1,16/node2,16/node3,16/node4 -–workdir . ./A_search.sh {} ::: {1..221}

Thiswilllaunchinstead48jobsspreadacrossfourmachines.SeetheGNUparalleldocumentationformoreinformation.

Otherusefuloptions

Therearetwooptionscontrollingfragmentplacementthatmayalsobeusefulincaseswherethereissomepreviousknowledgeaboutthestructureinquestion.Thefirstofthesedealswiththecasewherethebackbonestructureisknown(oratleastsomewhatknown)butregistrationofthesequencewiththebackbonemodelisambiguous.Inthiscase,aknownbackbonemodelcanbeprovidedwiththeflag:

-ca_positions backbone.pdb

Inthiscase,thecodewillonlyconsiderfragmentscenteredontheCalphapositionsfromtheinputmodel.Thisoffersasignificantspeedupaswellasreducedsearchspace.

Alternately,ifpartofthestructureisknowninadvance,itmaybeprovidedwiththeflag:

-startmodel start.pdb

Inthiscase,thematchingroutinewillmatchonlythenativefragmentscoveredbystartmodel,ensuringthatthesepositionswillbemaintainedthroughouttherefinement.Whenusingthisoptionstherearetwoimportantthingstokeepinmind:

• ThenumberinginthePDBfilemustmatchthenumberingoftheinputfasta• Anycontinuoussegmentsshorterthan9aminoacidsintheinputfilewillgetignored

Step4B.Placedfragmentscoring

Inthisstep,wewanttotaketheplacementsfromthepreviousstepandscorethemforcompatability.TheoutputsfromstepAareusedasinputsinthisstep(4_denovo_demo/B_score.sh):

$ROSETTA3/source/bin/denovo_density.macosclangrelease \ -mode score \ -in::file::silent round1/t20s*silent \ -scorefile round1/scores1 \ -n_matches 50

Highlightedinboldaretheinputfiles(-in::file::silent)–theoutputsfromthepreviousstep–andthescorefiletobewritten(-scorefile),theoutputofthisstep.Iftheoutputiswrittentoaseparatefolder,youwillneedtopointthecommandlinetothisalternatelocation.Thisstepisrelativelyfast(lessthan5minutes)andcanberunonasingleprocessor.

Step4C.MonteCarlofragmentassembly

Inthethirdstep,weusetheoutputsfromtheprevioustwosteps,andtrytogeneratea"maximally

consistent"fragmentassignment.ItusesMonteCarlosamplingandascorefunctionassessingfragmentcompatibilitytoidentifythisfragmentset.Thecommandline(4_denovo_demo/C_assemble.sh):

$ROSETTA3/source/bin/denovo_density.macosclangrelease \ -mode assemble \ -nstruct 5 \ -in::file::silent round1/t20s*silent \ -scorefile round1/scores1 \ -assembly_weights 4 20 6 \ -null_weight -150 \ -out:file:silent round1/assembled.$1 \ -scale_cycles 1 \ -mute core

AswithstepB,theoutputsoftheprevioustwostepsneedtobeprovidedasinputs:-in::file::silent round1/t20s*silent -scorefile round1/scores1

Thescriptthenwritesasinglefileforeachindependenttrajectory:-out:file:silent round1/assembled.$1

Finally,eachjobwillgenerateseveral(inthiscase5)independenttrajectories:-nstruct 5

Itisrecommendedtogenerateatotalof1000independenttrajectories.AswithstepA,thiscanbesomewhattime-consuming(thoughnotastime-consumingasstepA).Thereforeitisrecommendedtoparallelizethisasbefore:parallel –j16 ./C_assemble.sh {} ::: {1..200}

Oracrossmultiplemachines:

parallel –S 16/node1,16/node2,16/node3,16/node4 -–workdir . ./C_assemble.sh {} ::: {1..200}

Step4D.Consensusassignment

Thefinalstepoftheprotocolistoidentifytheconsensusassignmentfromthelowest-scoringMonteCarlotrajectories.Thisisdoneusingthefollowingcommand(4_denovo_demo/D_consensus.sh):

$ROSETTA3/source/bin/denovo_density.macosclangrelease \ -mode consensus \ -in::file::silent round1/assembled.*silent \ -consensus_frac 1.0 -energy_cut 0.05 \ -mute core

Theoutputtrajectoriesofthepreviousstepareprovidedasinputtothisscriptwith–in::file::silent.Thisscriptlooksforaconsensusassignmentinthebest-scoringtrajectories,andwilloutputaPDBfile,S_0001.pdb.

ThisPDBfilecontainssequenceplacedintodensity.Ideally,themodelatthispointismorethan70%complete,andthisfilecanthenbeusedasinputtoRosettaCM(seesection3).Ifinsteadthisstructurecontainsareasonablepartialmodel,butwithlessthan70%coverage,theiterativeapproachofthenextsectioncanfurtherimprovethecoverageofthepartialmodel.

Step4E.Iterativeassemblytoincreasemodelcoverage

Insomecases,itmaybenecessarytoiteraterefinement,assubsequentroundsofdenovobuildingmaytraceportionsofthemodelunabletobeplacedinpreviousrounds.Thefollowingcommandlineillustrateshowassemblymaybeiterated(4_denovo_demo/E_search_iter.sh):

$ROSETTA3/source/bin/denovo_density.macosclangrelease \ -in::file::fasta t20sA.fasta \ -fragfile ./t001_.25.9mers \ -startmodel round1_model.pdb \ -mapfile ./T20S_48A_alpha_chainA.mrc \ -n_to_search 250 -n_filtered 1250 -n_output 50 \ -bw 16 \ -atom_mask_min 2 \ -atom_mask 3 \ -clust_radius 3 \ -clust_oversample 4 \

-point_radius 3 \ -movestep 1 \ -delR 2 \ -frag_dens 0.8 \ -ncyc 3 \ -min_bb false \ -pos $1 \ -out:file:silent round2/t20s.$1.silent

ThisstepisnearlyidenticaltostepA,withtwokeychageshighlightedinbold.Thefirstchangeindicatesthattheoutputofthepreviousstepistobeusedasaninitialmodel:-startmodel round1_model.pdb

Followinginspection,thismodelmayalsobemanuallyeditedaswell.Thesecondchangemakessuretheoutputsdon’tclobbertheoutputsfromthepreviousstep:-out:file:silent t20s.rd2.$1.silent

AswithstepA,thisshouldbeparallelizedovermanyprocessors,e.g,usingGNUparallel:parallel –S 16/node1,16/node2,16/node3,16/node4 -–workdir . ./E_search_iter.sh {} ::: {1..221} Oncecomplete,stepsB,C,andDmaybethenfollowedinorder,makingsurethattheinputandoutputfilesareupdatedtoindicatetheround2models.NOTE:Ateachstep,becertainthatyouroutputsarenotoverwritingoneanother.Oncethepartialmodelhasbeencalculated,itissafetodeletetheintermediatefilescreatedaspartoftheconstructionprocess.Inthiscase,thesameoutputfilenamesmaybereused(thus,thesamescriptscanbeusedforeachiterationasidefromthefirst).

5)Completingpartialmodelsguidedbyexperimentaldensitydata

Whilethepreviousworkflowshaveaddressmodelbuildingandmodelrefinement,noneoftheaforementionedtoolsdealwithcompletionoflargesegmentsofprotein.Thesemayariseinseveralcases:

1. Homologymodels(particularlydistantones)mayhavelargeinsertions,orevenentiredomainsthatarelacking.

2. Themodelsproducedfromdenovo_densitymaybemissingsignificantfractionsofthebackbone3. Itmaybedifficulttomanuallytracelongstretchesoflowlocalresolutionintodensity

Toaddresstheseissues,wehavedevelopedatoolcalledRosettaEnumerativeSampling,whichusesaensemblesearchalgorithmtodeterminealargenumberofconformationsthatarebothconsistentwiththedensityandtheRosettaenergyfunction.Thistoolcanbeusedonapartialmodelsfromthedenovo_denistyapplication,anincompletehomologymodel,oranyotherstartingstructure. RosettaESrunsbestwhenworkingwithdataatresolutions5Åorbetterwithsegmentstorebuildshorterthan50residues.However,withverylargeamountsofsampling(e.g.,ensemblesizes>250),reliablemodelsmaybeproducedwithsegmentslongerthan100residues.Atresolutionsworsethan5Å,thistoolmaybeunreliable.Themethodcanbeusedonbothsegmentedandunsegmenteddensitymaps,however,removalofdensitybelongingtopartsofthestructurenotbeingmodeledmayimproveresults. RosettaESmodelbuildingconsistsofthreesteps.Initially,apreparationstepbuildsthefragmentsthataretobeusedinconformationalsampling.Thenarebuildingstepwillidentifyeachunassignedsegmentintheinitialmodelandbuildanensembleofpossiblesolutionsforeach.Finally,acombinationstepfindsalltheconsistentsubsetsofinteractions,andrefinesallsuchmodels(ifthereisonlyonesegment,thescriptsimplyrefinesallstructuresintheensemble).Inthiscombinationstep,ifassemblyfailstofindaconsistentsetofsolutions,anadditionalroundofsamplingwillbecarriedout,forcingdifferentsolutionsthanthepreviousmodel. Comparedtotheothersections,theworkflowisabitmorecomplicatedwhenextendedtomultiplecomputecores.TohandlejobdistributionwehaveincludedapythonscriptRunRosettaES.pythatmanagesthisjobdistributionamongavailableCPUsonasinglemachine.(ThescriptisincludedaspartofRosetta,in/main/source/scripts/python/public/EnumerativeSampling,aswellasinthistutorial).Fordealingwithjobschedulersorclustersincompatiblewiththisscript,section5EgivesanoverviewofjobdistributionwithRosettaES.

Step5A.FragmentPicking

Thefirststep–muchlikeScenario4–involvesselectionof"fragmentfiles,"whichpredictbackboneconformationfromlocalsequence.UnlikeScenario4,wehaveacustomalgorithmforfragmentpicking.ThesefragmentswillneedtobegeneratedbeforerunningRosettaES;thefollowingcommandwillgeneratethesefiles(5_rosettaES/A_PickFragments.sh): $ROSETTA3/source/bin/grower_prep.default.macosclangrelease \ -pdb input.pdb \ -in::file::fasta t20sA.fasta \ -fragsizes 3 9 \ -fragamounts 100 20

Thiswillgenerate1003residuefragmentsand209residuefragments,named100.3mersand20.9mers,thatarethenusedinsubsequentstepsoftherebuildingprocess.

Step5B.GeneratePossibleConformationsForEachSegment Thegrowerconsidersassigningpositionsforeachunassignedsegmentofdensity(thatis,eachstretchofaminoacidspresentinthefastafilebutmissingfromtheinputstructure).Eachsegmentisreferredtousingasegmentid,inwhicheachsegmentisnumberedfromN-toC-terminus(withmultiplechainsgiveninorderintheinputfastafile).Thescriptisrunintwoparts:first,thescriptisrunonceforeachsegmenttorebuild;then,thescriptisrunin“assemblymode”giventheoutputsproducedbyrebuildingeachsegmentindividually.Thus,forrebuildingthetwosegmentsinthetestcase,thescriptiscalledthreetimes:oncetobuildeachsegment,andoncetoassembletheresults. Inthefirststep,weperformconformationalsamplingofeachofthetwosegments,generatinganensembleofputativesolutionsforeach.Thiscanbedonecallingthecommand(5_rosettaES/B1_SampleSegment1.sh): python RunRosettaES.py \ -rs runES.sh \ -x RosettaES.xml \ -f t20sA.fasta \ -p input.pdb \ -d T20S_48A_alpha_chainA.mrc \ -l 1 \ -c 16 \ -n loop_1

Theargumentstothisprogramareasfollows:

• -rsrunES.sh-thescriptthatislaunchedoneachcoreandcontainsRosettaflagsandinputs • -xRosettaES.xml-theXMLscriptdescribingparametersforconformationalsampling(seebelow) • -ft20sA.fasta-theinputfastafile(withchainbreaksspecifiedby‘/’) • -pinput.pdb-theinputpdbfile.Thisneedstomatchtheinputsequence,andallresiduespresentin

thefastabutabsentinthePDBwillgetbuilt. • -dT20S_48A_alpha_chainA.mrc-theinputdensitymap • -l1-thesegmentidofthesegmenttorebuild.Thiscommandshouldbecalledonceforeachsegment

torebuild,varyingthisargumentfrom1toN • -c16-thenumberofcomputecorestouse • -nloop_1-theoutputtagforthisjob(resultswillbeplacedinafolderwiththisname).Tagsshould

beuniqueforeachsegment. TheinputXMLfileexposeskeyparametersforconformationalsampling.Inthetutorial,thisfile,5_rosettaES/RosettaES.xml,containsablock: ... <FragmentExtension name="ext" fasta="full.fasta" scorefxn="dens" censcorefxn="cendens" beamwidth="32" dumpbeam="0" samplesheets="1" read_from_file="0" continuous_weight="0.3" looporder="1" comparatorrounds=”100” windowdensweight=”30” readbeams="%%readbeams%%" storedbeams="%%beams%%" steps="%%steps%%" pcount="%%pcount%%" filterprevious="%%filterprevious%%" filterbeams="%%filterbeams%%"> <Fragments fragfile="100.3mers"/> <Fragments fragfile="20.9mers"/> </FragmentExtension> ...

ThesamplingbehaviorofRosettaESiscontrolledbytheblockabove.Manyofthetagsinthisblock–fasta,dumpbeam,read_from_file,storedbeams,steps,pcount,filterprevious,comparitorrounds,andfilterbeams–areusedbythejobdistributionscripttopassresultsfromonesteptothenext,andtheyshouldbeleftas-is.

Othersareuser-specified,andcanbemodifiedbasedonthesizeoftheloopandresolutionofthedata: • beamwidth:controlsthemaximumnumberofsolutionstobeheldateachstep.

Settingthevaluehigherwillincreaseruntimebutmayimproveaccuracy. • windowdensweight:therelativecontributionofdensityinmodelselection

Formanycases,thedefaultparametersaresufficient.However,ifthesegmenttogrowislong(50+residues),youmayneedtoincreasebeamwidth;ifthedensityislowresolution,youmightneedtodecreasewindowdensweightto15or20. Severaloptionsshouldrarelybemodified,butmayneedtobeinspecificcases:

• samplesheets:Controlswhetherornotbetasheetsamplingshouldbeperformed.Itisrecommendedtousethisexceptwhenworkingwithsymmetricsystems.

• continuous_weight:Controlsthepenaltyondiscontinuousdensity.Settingthisvalueto1willcompletelyremoveanypenaltyondiscontinuousdensity;settingitcloserto0willincreasethepenalty.Youmaywishtoraisethisvalueto0.7(ormore)ifyouanticipatethesegmentyouaretryingtomodeldoesnotfollowacontinuouspathofdensity.

Finally,theoptioncomparitorroundsisusedinmulti-segmentassembly(seesection5C)AfterrunningthescriptwiththisXML,therearetwoimportantintermediateoutputfiles,placedinthefolderloop_1(theargumentto-n):

• .lps(forlooppartialsolution)files,whicharethencombinedinstep5C,incaseswheretherearemultiplesegmentstomodel

• taboo/beamX.txtfiles,whereXcorrespondstothenumberofresiduesaddedtothesegment.Thesearegeneratedasthesearchaddsresidues,andareusedtopassinformationfromonesteptothenext(asadditionalresiduesareaddedinasinglesegment).

Note:ThisprocessshouldthenberepeatedforallremainingsegmentstorebuildInthetutorial,thecommand5_rosettaES/B2_SampleSegment2.shbuildsconformationsforthesecondsegmentinthisfile.Allsegmentscanbesampledindependentlyofoneanother,soifmanycomputenodesareavailable,eachsegmentcanbesampledsimultaneouslyonseparatenodes.Finally,whileinmostcases,userswillwanttotakethesemodelsintotheassemblystep(part5C),ifthereisonlyonesegmenttorebuild,orifthesamplingresultswanttobeinspected,thefinaloutputensemblecanbesavedasPDBfileswiththecommand(5_rosettaES/B1.2_InspectIntermediates.sh): python RunRosettaES.py \ -rs runES.sh \ -x RosettaES.xml \ -f t20sA.fasta \ -p input.pdb \ -d T20S_48A_alpha_chainA.mrc \ -l 1 \ -db loop_1/taboo/beam17.txt

Note,thenumberofthebeamfile(17)correspondstothetotalnumberofresiduesbuilt.Intermediateresults(aftergrowingNresidues)canbeinspectedbychangingthistoalowernumber(e.g.,beam14.txtshowssolutionsafter14ofthe17residueshavebeenrebuilt).

Step5C.Findasetofconsistentconformations.

Incaseswheretherearemultipleinteractingsegments,wewanttofindallnonclashingcombinations.Thisstepwilltakethelooppartialsolution(lps)filesgeneratedinstepBanduseaMonteCarloAssembly(MCA)algorithminordertoidentifysetsofsolutionsthatareself-consistent.ThissectionassumesthatallmissingsegmentshavebeenbuiltinstepB.Torunthisassembly,weperformconformationalsamplingusingthescript,passingthe.lpsfilesgeneratedinstepB(5_rosettaES/C_AssembleResults.sh): python RunRosettaES.py \ -rs runES.sh \ -x RosettaES.xml \ -f t20sA.fasta \ -p input.pdb \ -d T20S_48A_alpha_chainA.mrc \ -lps loop_*/lps*.txt

Theflag-lpspointstotheoutputsoftheindividualsegmentjobs’output.Asbefore,thescriptwilluse(andmodify)aninputXMLfile.Forassembly,onlyasingleparameterisimportanttomodify:comparatorrounds="100”.ThisparametercontrolsthenumberofMonteCarlotrajectories(itisunlikelyyouwillneedtochangethisparameter). TheoutputofthisstepisPDBfiles,thatwillbeplacedintheworkingdirectorywiththeprefixaftercomparator_RRR_XXX.pdb,whereRRRisthetrajectoryid,andXXXistheenergy.Atextfilerecommendation.txtwillbewrittenthatreportstheclashscoreofthebestmodel. Ifthenumberinrecommendation.txtisabove+100itisrecommendyouperformadditionalroundsofsampling.Todoso,additionalpotentialsolutionsshouldbesampledbyrepeatingstepBandprovidingthesamedirectorynameasinput,withoutdeletingtheintermediatefiles.Thescriptwillthenenterthatdirectoryandusethealready-computedsolutions(storedinthefolder"taboo")toguidesamplingtowardpreviouslyunexploredregions.SeesectionDformoredetails. Ifthenumberinrecommendation.txtisbelow+100,thenmodelsshouldberunthroughRosettarefinementtoaccuratelyrankthem.Thatcanbedoneusingthiscommand(5_rosettaES/D_RefineOutput): python RunRosettaES.py \ -rlxs runrelax.sh \ -x RosettaES.xml \ -p input.pdb \ -d T20S_48A_alpha_chainA.mrc \ -c 16 \ -rp aftercomparator_*.pdb

Whererunrelax.shcontainsacommandforrelaxingstructuresinRosetta.

Step5D.Interpretingresults RosettaESwillproducea100modelsasoutput(orwhateverisspecifiedincomparatorrounds),withscoresinthefilescore.sc.Whenfirstlookingattheoutputfromarun,itisgoodtovisuallyinspectthelowest-energy5-10structures.Ideally,theselowest-energymodelswillbeverytightlyconverged,butthelackofconvergencedoesnotindicatefailureinsampling,andthepresenceofconvergencedoesnotnecessaryindicatethesolutioniscorrect.Instead,thelowest-scoringmodelsshouldbeexaminedwithattentiontothefollowing:

Insufficientsampling.ModelsshouldinitiallybeinspectedforclearlyincorrectfeaturesthatmightnothavebeenproperlypenalizedbyRosetta.Theseinclude:

• unexplaineddensity,particularlysmallsidechainsplacedintoalargedensityprotrusions • regionswithpoorfittodensity • unresolvedclashes

Ifthesefeaturesarepresentinthelowest-energymodels,itsuggeststhattheconformationalsamplingperformedbyRosettaESmaynothavebeensufficient.Thereareseveralwaystoaddressthisissue.Thefirstistoincreaseconformationalsampling.Thiscanbedonebyeither:a)increasedthebeamwidthparameterandrerunninganew,b)and/orperformingadditionalroundsoftaboosearch(taboosamplingoccursautomaticallyifthe“taboo/”folderintherunningdirectoryispopulatedwithbeamfiles”),orc)reducingthesearchspacebyeliminatingregionsofdensity. Thelattercanalsobeperformedbymanuallyremovingpartsofthemapyoudonotwishtosampleorbyincludingadditionalportionsofthemodel,forexampleifyouarebuildingamodelthathas4missingregionsandyoubelieveyouhaveaccuratelysampled3ofthem(theydonotpossesanyofthepathologieslistedaboveandfitthedensitywell),youcancombinethembytreatingthemastemplatesforRosettaCM(useafastafilewiththemissingsegmentremoved)andusetheresultasinputforRosettaEStobuildthelastregion. Unresolvedresidues.RosettaESwillalwaysattempttobuildallresiduespresentinthefastafile,however,inmanycasesnotallresidueswillberesolvedinthemap(particularlyattermini).BecauseRosettaESwillheavilypenalizemodelsthatdonotfitthedensity,youwilloftenfindmodelsthatare"overlycompacted"atterminiorinternalloops,totrytosqueezetheseresiduesintodensity. Ifthishappensatterminiitissuggestedthatyouexamineintermediatestructuresthathaveyettoattempttoassigntheseunresolvedresiduesinordertofindagoodmodel.Ifthishappensinternally(orifyouhaveagoodideaaprioriwhatresiduesshouldbemodeled),theseregionscanberemovedfromtheinputfasta.Ifinternalsegmentsaredeleted,besuretotreatthedeletioncorrectly,byputtinga'/'inthefastafile. Step5E.Customizingjobdistribution. JobdistributioninRosettaESiscomplicated,sinceeach"growingstep"canbeparallelized,butsubsequentstepsneedalltheinformationfromthepreviousstep.Consequently,thescriptprovided(RosettaES.py)managesjobs,bycallingRosettajobsateachround,collectingandcombiningtheresultsofthepreviousround,thensplittinginputfilesforthenextround.Onsystemswhereitisnotpossibletorunthisscript,thissectiondescribesinsomedetailwhatthescriptisdoing. TomanagethisthepythonscriptusestheprovidedXMLfileasinput,rewritingseveralparametersdependingontheprotocolstep.Thesevaryingparametersinclude:

• readbeams,abooleanoptionthattellsRosettawhetheritshouldloadintermediatesolutions • beamfile,thenameofthebeamfilethatstorestheintermediatesolutions • steps,howmanyresiduestoadd(0meansonlydofiltering,otherwisesetto1,settingto“-1”will

buildallmissingresidues) • pcount,usedtouniquelytagoutputfilesfromeachcore • filterprevious,abooleanoptionthatcontrolswhetherintermediatesolutionsshouldbereadfortaboo

search • filterbeams,thefilenameoftheintermediatesolutionsforuseintaboo.

Inordertobuildamissingsegmentthescriptwillperformthefollowingsteps:

1. Launchajobwithreadbeams="0",beamfile="na",steps="1",pcount="1",filterprevious="0",filterbeams="na".Thiswillproduceafilenamedbeam1.1.txtfromRosetta.

2. Parsethebeam1.1.txtfileandsplitintoNfiles(whereNisthenumberofparalleljobstorun).Newfilesarelabeledbeam_$r.$i.txtwhere$risthenumberofresiduesaddedsofar,and$irangesfrom1toN.

3. SubmitNrosettajobswithpcountsetasthejobnumber,readbeamssetto"true",andbeamfilesettothecorrespondingoutputofstep2.

4. Outputfilesareparsedbythescriptandcompiledintoasinglefilewiththenamebeam$r.txt,with$rthenumberofresiduesgrown.

5. Rosettaisrunonasinglecoretofiltertheaggregatesolutionset.Here,readbeams=”1”,beamfile=”beam$r.txt”,andsteps="0".Afilenamedbeam_0.txtwillhavethefilteredresults.

6. Repeatsteps2-6usingthebeam_0.txtastheinputforstep2.ThisisdoneuntilthesegmentiscompleteandRosettaproducesanemptyfilecalledfinished.txt.Thepresenceofthisfiletriggerstheprogramtoperformonefinalroundoffilteringandexit.

ThelaststepofRosettaESwilladditionallyproduceafilewiththename"lpsfile_$s.0.txt,"usedforassembly.Torunassemblywiththesefilesfirstcombinethemintoasinglefile,describedbelow,andrunwithreadfromfile=”filename”.Thisnewfileshouldstartwiththeanumbercorrespondingtothetotalnumberofmissingsegments,thenforeachmissingsegmentprovideanumberforthetotalsolutionsinthatsegmentedfollowedbythesolutioninformationcontainedinthelpsfile_$s.0.txtdescribedabove.Segmentsshouldbearrangedtomatchtheorderinwhichtheyoccurinthefasta.