SmallMoleculesRotamers:GeneratingandTesting

KristianKaufmannVanderbiltUniversityRosettaCon2008July22nd‐25th2008

Unpublished and Confidential

GeneratingandUsingRotamersforSmallMolecules

•  Rotamersconstructionstrategy•  Predictionsofconformationsonasetof628smallmolecules

•  ModificationstoRGHIJJKLMNKOPtoincorporatesmallmoleculerotamers

•  EvaluationofDockingonasetof7proteinsincomplexwith10smallmolecules

Buildingmolecularmodelsfromfragmentsmoleculescapturedincrystalstructures

hasprovenfruitful

•  Fragmentsfordenovoproteinmodelling

•  Rotamersforproteinsidechainmodelling

•  Corinagenerates3DstructuresfromstatisticsgarneredfromtheCambridgeStructuralDatabase(CSD)ofsmallmolecules.

•  FleXXanincrementalconstructionalgorithmforsmallmoleculedockinganddesignalsomakesuseoftheCSDinitsmallmoleculeconstruction

Rohletal.MethodsEnzymol.2004383p.66‐93DunbrackandKarplusJMolBiol.1993230p.543‐74Gasteigeretal.TetrahedronComp.Method.1990,3,537‐547Rareyetal.JMolBiol.1996261470‐89

PowerfulEnergyfunctionscanbederivedfromcrystalstructures

•  Hydrogenbondingfunctionderivedthroughanalysisofhighresolutionproteincrystalstructures.

•  DunbrackrotamerenergyderivedfromtheprobabilityofobservingtherotamerinthePDB

•  VanderWaalsforcesderivedfromatomicdistancesincrystalstructures

Simonsetal.Proteins.199934p.82‐95Kuhlmanetal.Science.2003302p.1364‐8

Pre‐computeinteractionsforrotamerlibraryoflikely

aminoacidsconformations

ProteinSmallMolecule

Pre‐computeinteractionsforsmallmoleculelibraryoflikelyconformations

Replacementofsidechainconformations

Replacementofsmallmoleculeconformations

1(Setup)

2(coarsediscreteoptimization)

ImplementingSmallMoleculeFlexibilityusingRotamerswhileRetainingRosetta’sFunctionality

SimultaneousoptimizationofProteinbindingsite

Selectrotamersfromsidechainconformationsseenin

thePDB

Constructrotamersfromtorsionprofilesseeninthe

CSD

GeneratingTorsionProfilefromtheCambridgeStructuralDatabase

X

XX

X

DefineAtomTypestocapturechemicalcharacteristicsofdihedralangles

SearchCambridgeStructuralDatabaseforallatompairsexcludingthoseinringsystemsandmeasuredihedral.Bineverytendegrees.

0

1000

2000

3000

4000

5000

‐180

‐160

‐140

‐120

‐100

‐80

‐60

‐40

‐20 0 20

40

60

80

100

120

140

160

180

Freq

uency

DihedralAngle

AromaticCarbonEtherOxygenBond

GeneratingEnergiesfromaTorsionProfile

•  Energiescanbegeneratedbycomputingthenegativelogofthepropensityofastate

•  Thepropensityistheprobabilitynormalizedbytherandomprobabilityofselectingastate.

•  Anyanglewithaenergylessthanzerohasagreaterthanrandomprobability.

•  Minimaintheenergyprofilecanthenbeusedinconstructingsmallmoleculeconformations

010002000300040005000

‐180 ‐120 ‐60 0 60 120 180

Freq

uency

AromaticCarbonEtherOxygenBond

‐2‐10123456

‐180 ‐120 ‐60 0 60 120 180

Dihed

ralEne

rgy

DihedralAngle

CSD

AMBER

GeneratingSmallMoleculeRotamers

Pickarandomsetofoptimaldihedralanglesforallrotatablebonds

Doesitclash?

Storeconformation.10000acceptedconformations?

Acceptconformationwithlowestenergy.Createlistofallconformationswithin2*numberof

rotatablebondsenergyunitsofthelowestenergyconformation.Acceptmaximalspanningensemble

ofconformations.

Yes

Yes

No

No

PredictingBioactiveConformations•  Testsettakenfrom

PDBBindacollectionofcrystalstructuresforsmallmoleculesfromproteincomplexeswithknownbindingconstants

•  PDBBindwasculledtocontainonlymoleculeswith≤6rotatablebonds

•  Upto500rotamersweregeneratedforeachsmallmolecule

Wangetal.JMedChem.200548p.4111‐9.

#of Rotatable Bonds

#of Molecule

Average RMSD of closest conformation

Average RMSD of furthest conformation

1 92 0.14±0.16 1.12±0.47 2 118 0.33±0.26 1.74±0.69 3 118 0.41±0.22 2.13±0.62 4 135 0.47±0.21 2.45±0.69 5 97 0.61±0.30 2.83±0.81 6 118 0.79±0.32 3.07±0.87

Overall Total 628 0.46±0.31 2.23±0.94

AtomicRMSDtocrystallizedconformation

DockingwithSmallMoleculeRotamersinRosetta

Proteincrystalstructureandlowenergysmallmoleculeconformation

Generateensembleofsmallmoleculeconformations

Translatesmallmoleculetorandompointinbindingsite

Pickrandomconformationtranslaterandomlyinside1Aboxrotaterandomlyaddconformation

tolist

X<1000

Selectfirst100nonclashingsmall

conformationsorbestseen

X++

yes

no

Performrotameroptimizationwithproteinsidechainandsmallmoleculerotamers

Performminimizationofproteinsidechainchianglesandrigidbodydegreesof

freedom

Outputstructure

FororiginalprotocolseeMeilerandBakerProteins.200665p.538‐48.

DockingBenchmarks

•  15crystalstructureswithsmallmolecules–  7proteins,10ligands

•  10selfdockingexamples–  Verifiesprotocolsabilityto

sampleandidentifynativestate.

•  11crossdockingexamples–  Testssensitivitytochanges

inbackboneaswouldbeexpectedfromahomologymodel.

DockingBenchmark

!15

!10

!5

0

0 2 4 6 8 10 12 14 16 18 20

Res

idue

Ene

rgy

RMSD in Å

modelsnative

2PRGdockingto1FM9ligand binding domain of the human peroxisome proliferator activated receptor γ

!15

!10

!5

0

0 2 4 6 8 10 12 14 16 18 20

Res

idue

Ene

rgy

RMSD in Å

modelsnative

1P8Ddockingto1P8DOxysterolsreceptorliverXreceptorβ.

DockingBenchmark

!15

!10

!5

0

0 2 4 6 8 10 12 14 16 18 20

Res

idue

Ene

rgy

RMSD in Å

modelsnative

2PRGdockingto1FM9ligand binding domain of the human peroxisome proliferator activated receptor γ

!15

!10

!5

0

0 2 4 6 8 10 12 14 16 18 20

Res

idue

Ene

rgy

RMSD in Å

modelsnative

1P8Ddockingto1P8DOxysterolsreceptorliverXreceptorβ.

PerformanceinaDockingBenchmark

In9outof10casesselfdockingwassuccessfulwhenlookingatthetop1%byenergy

Ligand Protein rank RMSD 1aq1 1aq1 1 0.56 1dm2 1dm2 1 0.31 1dbj 1dbj 1 1.36 2dbl 2dbl 1 1.45 1p8d 1p8d 1 1.63 4tim 4tim 1 1.87 6tim 6tim 1 1.77 2ctc 2ctc 3 0.82 1pph 1pph 6 1.49 2prg 2prg 639 1.94

SelfDockingResults

In8of11casescrossdockingsuccessfulwhenlookingatthetop1%byenergy

CrossDockingResultsLigand Protein rank RMSD 1dm2 1aq1 1 0.56 1dbj 2dbl 1 1.80 1pph 1ppc 2 1.96 4tim 6tim 2 1.90 2ctc 7cpa 3 0.95 6tim 4tim 5 1.77 1p8d 1pqc 10 1.28 2prg 1fm9 16 2.02 1p8d 1pq6 181 1.62 2dbl 1dbj 468 3.49 1aq1 1dm2 4296 1.87

ConclusionsandFutureDirections

•  SmallmoleculerotamerspresentaviablemethodofrepresentingsmallmoleculeflexibilityinRosetta

•  90%ofofselfdockingcasesidentifiednative‐likestructureinthetop1%byenergy.

•  72%ofcrossdockingcasesidentifiedanative‐likestructureinthetop1%byenergy

•  Optimizationofthescoringfunctionfordiscriminationofnativelikemodelsshouldyieldimprovedresults

•  Meldingsmallmoleculerotamerapproachwithincrementalconstructionapproachisunderwayforapplicationtolargersmallmolecules.

Acknowledgements

•  JensMeiler•  RalfMueller

•  KristinGlab•  DARPAandNIHMolecularBiophysicsTrainingGrant