OpenBabel:Anopenchemicaltoolbox - PDF document

SOFTWARE OpenAccess OpenBabel:Anopenchemicaltoolbox NoelMO ’ Boyle 1 ,MichaelBanck 2 ,CraigAJames 3 ,ChrisMorley 4 ,TimVandermeersch 4 andGeoffreyRHutchison 5* Abstract Background: Afrequentproblemincomputationalmodelingistheinterconversionofchemicalstructures betweendifferentformats.Whilestandardinterchangeformatsexist(forexample,ChemicalMarkupLanguage)and defacto problemduetothemultitudeofdifferentapplicationareasforchemistrydata,differencesinthedatastoredby differentformats(0Dversus3D,forexample),andcompetitionbetweensoftwarealongwithalackofvendor- neutralformats. Results: Wediscuss,forthefirsttime,OpenBabel,anopen-sourcechemicaltoolboxthatspeaksthemany languagesofchemicaldata.OpenBabelversion2.3interconvertsover110formats.Theneedtorepresentsucha widevarietyofchemicalandmoleculardatarequiresalibrarythatimplementsawiderangeofcheminformatics algorithms,frompartialchargeassignmentandaromaticitydetection,tobondorderperceptionand canonicalization.WedetailtheimplementationofOpenBabel,describekeyadvancesinthe2.3release,and outlineavarietyofusesbothintermsofsoftwareproductsandscientificresearch,includingapplicationsfar beyondsimpleformatinterconversion. Conclusions: OpenBabelpresentsasolutiontotheproliferationofmultiplechemicalfileformats.Inaddition,it providesavarietyofusefulutilitiesfromconformersearchingand2Ddepiction,tofiltering,batchconversion,and substructureandsimilaritysearching.Fordevelopers,itcanbeusedasaprogramminglibrarytohandlechemical datainareassuchasorganicchemistry,drugdesign,materialsscience,andcomputationalchemistry.Itisfreely availableunderanopen-sourcelicensefromhttp://openbabel.org. Introduction Thehistoryofchemicalinformaticshasincludedahuge varietyoftextualandcomputerrepresentationsofmole- culardata.Suchrepresentationsfocusonspecificatomic ormolecularinformationandmaynotattempttostore allpossiblechemicaldata.Forexample,linenotations likeDaylightSMILES[1]donotoffercoordinateinfor- mation,whilecrystallographicorquantummechanical formatsfrequentlydonotstorechemicalbondingdata. Hydrogenatomsarefrequentlyomittedfromx-raycrys- nates,andareoftenignoredbysomefileformatsasthe “ implicitvalence ” ofheavyatomsthatindicatestheir presence.Othertypesofrepresentationsrequirespecifi- cationofatomtypesonthebasisofaspecificvalence bondmodel,inclusionofcomputedpartialcharges, indicationofbiomolecul arresidues,ormultiple conformations. Whileattemptshavebeenmadetoprovideastandard formatforstoringchemicaldata,includingmostnotably thedevelopmentofChemicalMarkupLanguage(CML) [2-6],anXMLdialect,suchformatshavenotyet achievedwidespreaduse.Consequently,afrequentpro- blemincomputationalmodelingistheinterconversion ofmolecularstructuresbetweendifferentformats,apro- cessthatinvolvesextraction andinterpretationoftheir chemicaldataandsemantics. Weoutlineforthefirsttime,thedevelopmentanduse oftheOpenBabelproject,afull-featuredopenchemical toolbox,designedto “ speak ” themanydifferentrepre- sentationsofchemicaldata.Itallowsanyonetosearch, convert,analyze,orstoredatafrommolecularmodeling, chemistry,solid-statematerials,biochemistry,orrelated areas.Itprovidesbothready-to-useprogramsaswellas acomplete,extensibleprogrammer ’ stoolkitfordevelop- ingcheminformaticssoftware.Itcanhandlereading, *Correspondence:geoffh@pitt.edu 5 UniversityofPittsburgh,DepartmentofChemistry,219ParkmanAvenue, Pittsburgh,PA15217,USA Fulllistofauthorinformationisavailableattheendofthearticle O ’ Boyle etal . JournalofCheminformatics 2011, 3 :33 http://www.jcheminf.com/content/3/1/33 ©2011O ’ Boyleetal;licenseeChemistryCentralLtd.ThisisanOpenAccessarticledistributedunderthetermsoftheCreative CommonsAttributionLicense(http://creativecommons.org/licenses/by/2.0),whichpermitsunrestricteduse,distribution,and reproductioninanymedium,providedtheoriginalworkisproperlycited. writing,andinterconvertingover110chemicalfilefor-mats,supportsfilteringandsearchingmoleculefilesusingDaylightSMARTSpatternmatching[7]andothermethods,andprovidesextensiblefingerprintingandmolecularmechanicsframeworks.Wewilldiscusstheframeworksforfileformatinterconversion,fingerprint-ing,fastmolecularsearching,bondperceptionandatomtyping,canonicalnumberingofmolecularstructuresandfragments,molecularmechanicsforcefields,andtheextensibleinterfacesprovidedbythesoftwarelibrarytoenablefurtherchemistrysoftwaredevelopment.OpenBabelhasitsorigininaversionofOELibreleasedasopen-sourcesoftwarebyOpenEyeScientificundertheGPL(GNUPublicLicense).In2001,OpenEyedecidedtorewriteOELibin-houseastheproprietaryOEChemlibrary,sotheexistingcodefromOELibwasspunoutintothenewOpenBabelproject.Since2001,OpenBabelhasbeendevelopedandsubstantiallyextendedasaninternationalcollaborativeprojectusinganopen-sourcedevelopmentmodel[8].Ithasover160,000downloads,over400citations[9],isusedbyover40softwareprojects[10],andisfreelyavailablefromtheOpenBabelwebsite[11].FeaturesFileFormatSupportWiththereleaseofOpenBabel2.3,OpenBabelsup-ports111chemicalfileformatsintotal.Itcanread82formatsandwrite85formats.Theseencompasscom-monformatsusedincheminformatics(SMILES,InChI,MOL,MOL2),inputandoutputfilesfromavarietyofcomputationalchemistrypackages(GAMESS,Gaussian,MOPAC),crystallographicfileformats(CIF,ShelX),reactionformats(MDLRXN),fileformatsusedbymoleculardynamicsanddockingpackages(AutoDock,Amber),formatsusedby2Ddrawingpackages(Chem-Draw),3Dviewers(Chem3D,Molden)andchemicalkineticsandthermodynamics(ChemKin,Thermo).For-matsareimplementedasinOpenBabel,whichmakesiteasyforuserstocontributenewfilefor-mats(seeExtensibleInterfacebelow).Dependingontheformat,otherdataisextractedbyOpenBabelinaddi-tiontothemolecularstructure;forexample,vibrationalfrequenciesareextractedfromcomputationalchemistrylogfiles,unitcellinformationisextractedfromCIFfiles,andpropertyfieldsarereadfromSDFfiles.Anumberofutilityfileformatsarealsodefined;thesearenotstrictlyspeakingawayofstoringthemolecularstructure,butratherpresentcertainfunction-alitythroughthesameinterfaceastheregularfilefor-mats.Forexample,thereportformatisawrite-onlyutilityformat[12]thatpresentsasummaryofthemole-cularstructureofamolecule;thefingerprintformattandfastsearchformat[14]areusedforsimilarityandsubstructuresearching(seebelow);theMolPrint2DandMultilevelNeighborhoodsofAtomsformatscalculatecir-cularfingerprintsdefinedbyBenderetal.[15,16]andetal.[17,18]respectively.Eachformatcanhavemultipleoptionstocontroleitherreadingorwritingaparticularformat.Forexam-ple,theInChIformathas12optionsincludinganoptiontogenerateanInChIKey,Tparam㚐truncatetheInChIdependingonasuppliedparameterandtoignorecertainInChIwarnings.Theavailableoptionsarelistedinthedocumentation,areshownintheGraphicalUserInterface(GUI)ascheckboxesortextboxes,andcanbelistedatthecommand-line.Infact,allthreearegeneratedfromthesamesource;adocumentationstringintheC++code.FingerprintsandFastSearchingDatabasesarewidelyusedtostorechemicalinformationespeciallyinthepharmaceuticalindustry.Akeyrequire-mentofsuchadatabaseistheabilitytoindexchemicalstructuressothattheycanbequicklyretrievedgivenaquerysubstructure.OpenBabelprovidesthisfunctional-ityusingapath-basedfingerprint.Thisfingerprint,referredtoasFP2inOpenBabel,identifiesalllinearandringsubstructuresinthemoleculeoflengths1to7(excludingthe1-atomsubstructuresCandN)andmapsthemontoabit-stringoflength1024usingahashfunc-tion.Ifaquerymoleculeisasubstructureofatargetmolecule,thenallofthebitssetinthequerymoleculewillalsobesetinthetargetmolecule.Thefingerprintsfortwomoleculescanalsobeusedtocalculatestruc-turalsimilarityusingtheTanimotocoefficient,thenum-berofbitsincommondividedbytheunionofthebitsClearly,repeatedsearchingofthesamesetofmole-culeswillinvolverepeateduseofthesamesetoffinger-prints.Toavoidtheneedtorecalculatethefingerprintsforaparticularmulti-moleculefile(suchasanSDFfile),OpenBabelprovidesaformatthatsolelystoresafingerprintalongwithanindexintotheoriginalfile.Thisindexleadstoarapidincreaseinthespeedofsearchingformatchestoaquery-datasetswithseveralmillionmoleculesareeasilysearchedinteractively.Inthisway,amulti-moleculefilemaybeusedasalight-weightalternativetoachemicaldatabasesystem.BondPerceptionandAtomTypingAsmentionedabove,manychemicalfileformatsofferrepresentationsofmoleculardatasolelyaslistsofatoms.Forexample,mostquantumchemicalsoftwarepackagesandmostcrystallographicfileformatsdonotofferdefinitionsofbonding.AsimilarsituationoccursinthecaseoftheProteinDataBank(PDB)format;whilestandardized[19]filescontainconnectivityetalJournalofCheminformaticshttp://www.jcheminf.com/content/3/1/33Page2of14 information,non-standardfilesexistthatoftendonotprovidefullconnectivityinformation.Consequently,OpenBabelfeaturesmethodstodeterminebondcon-nectivity,bondorderperception,aromaticitydetermina-tion,andatomtyping.Bondconnectivityisdeterminedbythefrequentlyusedalgorithmofdetectingatomscloserthanthesumoftheircovalentradii,withaslighttolerance(0.45Å)toallowforlongerthantypicalbonds.Tohandledisorderincrystallographicdata(e.g.,PDBorCIFfiles),atomscloserthan0.63Åarenotbonded.Afurtherfilteringpassismadetoensurestandardbondvalencyismain-tained;eachelementhasamaximumnumberofbonds,ifthisisexceededthenthelongestbondstoanatomaresuccessivelyremoveduntilthevalenceruleisAfterbondconnectivityisdetermined,ifneededorrequestedbytheuser,bondorderperceptionisper-formedonthebasisofbondanglesandgeometries.ThemethodissimilartothatproposedbyRogerSayle[20]andusestheaveragebondanglearoundanun-typedatomtodeterminespandsphybridizedcenters.5-memberedand6-memberedringsarecheckedforpla-naritytoestimatearomaticity.Finally,atomsmarkedasunsaturatedarecheckedforanunsaturatedneighbortogiveadoubleortriplebond.Afterthisinitialatomtyp-ing,knownfunctionalgroupsarematched,followedbyaromaticrings,followedbyremainingunsatisfiedbondsbasedonasetofheuristicsforshortbonds,atomicelec-tronegativity,andringmembership.Atomtypingisperformedbylazyevaluation,match-ingatomsagainstSMARTSpatternstodeterminehybri-dization,implicitvalence,andexternalatomtypes.Atomtypeperceptionmaybetriggeredbyaddinghydrogens(whichrequiresdeterminationofimplicitandexplicitvalence),exportingtoafileformatthatrequiresatomtypes,orasrequestedbytheuser.Tominimizetheamountoftypingrequired,whenimportingfromaformatwithatomtypesspecified,alookuptableisusedtotranslatebetweenequivalenttypes.Animportantpartofatomtypingisaromaticitydetec-tionandassignmentofKekulébondorders(kekuliza-tion).InOpenBabel,acentralaromaticitymodelisused,largelymatchingthecommonlyusedDaylightSMILESrepresentation[1],butwithaddedsupportforaromaticphosphorousandselenium.Potentialaromaticatomsandbondsareflaggedonthebasisofmember-shipinaringsystempossiblycontaining4n+2elec-trons.Aromaticityisestablishedonlyifawell-definedvalencebondKekulépatterncanbedetermined.Todothis,atomsareaddedtoaringsystemandcheckedagainstthe4n+2electronconfiguration,graduallyincreasingthesizetoestablishthelargestpossiblecon-nectedaromaticringsystem.Oncethisringsystemisdetermined,anexhaustivesearchisperformedtoassignsingleanddoublebondstosatisfyallvalencesinaKekuléform.Sincethisprocessisexponentialincom-plexity,thealgorithmwillterminateifmorethan30levelsofrecursionor15secondsareexceeded(whichmayoccurinthecaseoflargefusedringsystemssuchascarbonnanotubes).CanonicalRepresentationofMoleculesIngeneral,foranyparticularmolecularstructureandfileformat,therearealargenumberofpossiblewaysthestructurecouldbestored;forexample,thereareN!waysoforderingtheatomsinanMOLfile.Whileeachoftheorderingsencodesexactlythesameinformation,itcanbeusefultodefineacanonicalnumberingoftheatomsofamoleculeandusethistoderiveacanonicalrepresentationofamoleculeforaparticularfileformat.Forazero-dimensionalfileformatwithoutcoordinates,suchasSMILES,thecanonicalrepresentationcouldbeusedtoindexadatabase,removeduplicatesorsearchformatches.OpenBabelimplementsasophisticatedcanonicaliza-tionalgorithmthatcanhandlemoleculesormolecularfragments.Theatomsymmetryclassesaretheinitialgraphinvariantsandencodetopologicalandchemicalproperties.Acooperativelabelingprocedureisusedtoinvestigatetheautomorphicpermutationstofindthecanonicalcode.AlthoughthealgorithmissimilartotheoriginalMorgancanonicalcode[21],variousimprove-mentsareimplementedtoimproveperformance.Mostnotably,thealgorithmimplementsheuristicsfromthepopularnautypackage[22,23].Anotheraspecthandledbythecanonicalcodeisstereochemistryasdifferentlabelingscanleadtodifferentparities.Thisisfurthercomplicatedbythepossibilityofsymmetry-equivalentstereocentersandstereocenterswhoseconfigurationisinterdependent.Thefulldetailswillbethesubjectofaseparatepublication.CoordinateGenerationin2Dand3DOpenBabel,version2.3,hassupportfor2Dcoordinategeneration(Figure1)throughthedonationofcodebySergeiTrepalin,basedonthecodeusedintheMCDLchemicalstructureeditor[24-26].TheMCDLalgorithmaimstolayoutthemolecularstructurein2Dsuchthatallbondlengthsareequalandallbondanglesarecloseto120°.Thelayoutalgorithmincludesasmalldatabaseofaround150templatestohelplayoutcagesandlargefragmentcycles.Todealwiththeproblemofoverlap-pingfragments,thealgorithmincludesanexhaustivesearchprocedurethatrotatesaroundacyclicbondsbyCoordinategenerationin3DwasintroducedinOpenBabelversion2.2,andimprovedinversion2.3,toenableetalJournalofCheminformaticshttp://www.jcheminf.com/content/3/1/33Page3of14 conversionfrom0DformatssuchasSMILESto3Dfor- matssuchasSDF(Figure1).The3Dstructuregenera- torbuildslinearcomponentsfromscratchfollowing geometricalrulesbasedonthehybridizationofthe atoms.Single-conformerringtemplatesareusedfor ringsystems.Thetemplatematchingalgorithmiterates throughthetemplatesfromlargesttosmallestsearching formatches.Ifamatchisfound,thealgorithmcon- tinuesbutwillnotmatchanyringatomspreviously templatedexceptinthecaseofasingleoverlap(thetwo ringsystemsofaspirogroup)oranoverlapinvolving exactlytwoadjacentatoms(twofusedringsystems). Afteraninitialstructureisgenerated,thestereochemis- try(cis/transandtetrahedral)iscorrectedtomatchthe inputstructure.Finally,theenergyofthestructureis minimizedusingtheMMFF94forcefield[27-31]anda lowenergyconformerfoundusingaweightedrotor search. Whilethe3Dstructurebuilderproducesreasonable conformationsformoleculeswithoutringsorwithring systemsforwhichatemplateexists,theresultsmaybe poorformoleculeswithmorecomplexringsystemsor organometallicspecies.Futureworkwillbeperformed tocomparetheresultsofOpenBabelwithotherpro- gramswithrespecttobothspeedandthequalityofthe generatedstructures[32]. Stereochemistry ArecentfocusofOpenBabeldevelopmenthasbeento ensurerobusttranslationofstereochemicalinformation betweenfileformats.Thisisparticularlyimportant whendealingwith0Dformatsastheseexplicitlyencode theperceivedstereochemistry.OpenBabel2.3includes classestohandlecis/transdoublebondstereochemistry, tetrahedralstereochemistryandsquare-planarstereo- chemistry(thislastisstillunderdevelopment),aswell asperceptionroutinesfor2Dand3Dgeometries,and routinestoqueryandalterthestereochemistry. Thedetectionofstereogenicunitsstartswithanana- lysisofthegraphsymmetryofthemoleculetoidentify thesymmetryclassofeachatom.However,giventhata completesymmetryanalysisalsoneedstotakestereo- chemistryintoaccount,thismeansthattheoverall stereochemistrycanonlybefounditeratively.Ateach iteration,thecurrentatomsymmetryclassesareusedto identifystereogenicunits.Forexample,atetrahedral centerisidentifiedaschiralifithasfourneighborswith differentsymmetryclasses(orthree,inthecasewherea lonepairgivesrisetothetetrahedralshape). Forcefields Molecularmechanicsfunctionsareprovidedforuse withsmallmolecules.Typicalapplicationsinclude energyevaluationorminimization,aloneoraspartofa largerworkflow.Theselectionofimplementedforce fieldsallowsmostmolecularstructurestobeusedand parameterstobeassignedautomatically.TheMMFF94 (s)forcefieldcanbeusedfororganicordrug-likemole- cules[27-31].Formoleculescontaininganyelementof theperiodictableorcomplexgeometry(i.e.notsup- portedbyMMFF94),theUFFforcefieldcanbeused instead[33].Recently,codeimplementingtheGAFF forcefield[34,35]wasalsocontributedandreleasedas partofversion2.3.Alloftheforcefieldsallowtheappli- cationofconstraintsonparticularatompositions,or particulardistances. Severalconformersearchingmethodshavebeen implementedusingtheforcefields,allbasedonthe “ tor- sion-driving ” approach.Thisapproachinvolvessetting torsionanglesfromasetofpredefinedallowedvalues foraparticularrotatablebond.Themostthorough searchmethodimplementedisasystematicsearch method,whichiteratesoveralloftheallowedtorsion anglesforeachrotatablebondinthemoleculeand retainstheconformerwiththelowestenergy.Sincea systematicsearchmaynotbefeasibleforamolecule withmultiplerotatablebonds,anumberofstochastic searchmethodsarealsoavailable:therandomsearch method,whichtriesrandomsettingsforthetorsion angles(fromthepredefinedallowedvalues),anda weightedrotorsearch,astoc hasticsearchmethodthat convergesonalowenergyconformerbyweightingpar- ticulartorsionanglesbasedontherelativeenergyofthe generatedconformer.WithOpenBabel2.3,conformer searchbasedonageneticalgorithmisalsoavailable whichallowstheapplicationoffilters(e.g.adiversityfil- ter)anddifferentscoringfunctions.Thislattermethod canbeusedtogeneratealibraryofdiverseconformers, Figure1 Interconversionof0D,2Dand3Dstructures .The structuresshownareofsertraline,aselectiveserotoninreuptake inhibitor(SSRI)usedinthetreatmentofdepression.ASMILESstring forsertralineisshownatthetop;thiscanbeconsidereda0D structure(onlyconnectivityandstereochemicalinformation).From this,OpenBabelcangeneratea2Dstructure(bottomleft,depicted byOpenBabel)ora3Dstructure(bottomright,depictedby Avogadro),andallofthesecanbeinterconverted. O ’ Boyle etal . JournalofCheminformatics 2011, 3 :33 http://www.jcheminf.com/content/3/1/33 Page4of14 orliketheothermethodstoseekalowenergyconfor- mer[36]. Implementation TechnicalDetails OpenBabelisimplementedinstandards-compliantC+ +.ThisensuressupportforawidevarietyofC++com- pilers(MSVC,GCC,IntelCompiler,MinGW,Clang), operatingsystems(Windows,MacOSX,Linux,BSD, Windows/Cygwin)andplatforms(32-bit,64-bit).Since version2.3,itiscompiledusingtheCMakebuildsystem [37,38].Thisisanopen-sourcecross-platformbuildsys- temwithadvancedfeaturesfordependencyanalysis. Thebuildsystemhasanassociatedunittestframework CTest,whichallowsnightlybuildstobecompiledand testedautomaticallywiththeresultscollatedanddis- playedonacentralizeddashboard[39]. TosimplifyinstallationOpenBabelhasasfewexter- naldependenciesaspossible.Wheresuchdependencies exist,theyareoptional.Forexample,iftheXMLdevel- opmentlibrariesarenotavailable,OpenBabelwillstill compilesuccessfullybutnoneoftheXMLformats (suchasChemicalMarkupLanguage,CML)willbe available.Similarly,ifthe Eigenmatrixandlinearalge- bralibraryisnotfound,any classesthatrequirefast matrixmanipulation(suchasOBAlign,whichperforms leastsquaresalignment)willnotbecompiled. WhilethemajorityoftheOpenBabellibraryiswrit- teninC++,bindingshavebeendevelopedforarangeof otherprogramminglanguages,includingJavaandthe. NETplatform,aswellastheso-called “ dynamic ” script- inglanguagesPerl,Python,andRuby.Theseareauto- maticallygeneratedfromtheC++headerfilesusingthe SWIGtool.Asdescribedpreviously[40],inthecaseof PythonanadditionalmoduleisprovidednamedPybel thatsimplifiesaccesstotheC++bindings.Theseinter- facesfacilitatedevelopmentofweb-enabledchemistry applications,aswellasrapiddevelopmentand prototyping. CodeArchitecture TheOpenBabelcodebasehasamodulardesignas showninFigure2.Thegoalofthisdesignisthreefold: 1.Toseparatethechemistry,theconversionprocess andtheuserinterfacesreducing,asfaraspossible, thedependencyofoneuponanother. 2.Toputallofthecodeforeachchemicalformatin oneplace(usuallyasinglefile)andmaketheaddi- tionofnewformatssimple. 3.Toallowtheformatconversionofnotjustmole- cules,butalsoanyotherc hemicalobjects,suchas reactions. Thecodebasecanbeconsideredasconsistingofthe followingmodules(Figure2):  TheChemicalCore,whichcontainsOBMoletc. andhasallofthechemicalstructuredescriptionand manipulation.Thisistheheartoftheapplication anditsAPIcanbeusedasachemicaltoolbox.It hasnoinput/outputcapabilities.  TheFormats,whichreadandwritetofilesofdif- ferenttypes.Theseclassesarederivedfromacom- monbaseclass,OBFormat,whichisinthe ConversionControlmodule.Theyalsomakeuseof thechemicalroutinesintheChemicalCoremodule. Eachformatfilecontainsaglobalobjectofthefor- matclass.Whentheformatisloadedtheclasscon- structorregistersthepresenceoftheclasswith OBConversion.Thismeansthattheformatsareplu- gins-newformatscanbeaddedwithoutchanging anyframeworkcode.  CommonFormatsincludeOBMoleculeFormatand XMLBaseFormatfromwhichmostotherformats (likeFormatAandFormatBinthediagram)are derived.IndependentformatslikeFormatCarealso possible.  TheConversionControl,whichalsokeepstrackof theavailableformats,theconversionoptionsandthe inputandoutputstreams.Itcanbecompiledwith- outreferencetoanyotherpartsoftheprogram.In particular,itknowsnothingoftheChemicalCore: mol.hisnotincluded.  TheUserInterface,whichmaybeacommandline application,aGraphicalUserInterface(GUI),or maybepartofanotherprogramthatusesOpen Babel ’ sinputandoutputfacilities.Thisdependsonly Figure2 ArchitectureoftheOpenBabelcodebase . O ’ Boyle etal . JournalofCheminformatics 2011, 3 :33 http://www.jcheminf.com/content/3/1/33 Page5of14 ontheConversionControlmodule(obconversion.h isincluded),butnotontheChemicalCoreoron anyoftheFormats.  TheFingerprintAPI,aswellasbeingusablein externalprograms,isemployedbythefastsearchand fingerprintformats.  TheFingerprints,whicharebitarraysthatdescribe anobjectandwhichfacilitatefastsearching.They arealsobuiltasplugins,registeringthemselveswith theirbaseclassOBFingerprintwhichisintheFin- gerprintAPI.  OtherfeaturessuchasForcefields,PartialCharge ModelsandChemicalDescriptors,althoughnot showninthediagram,arehandledsimilarlyto Fingerprints.  TheErrorHandlingcanbeusedthroughoutthe programtologanddisplayerrorsandwarnings. ExtensibleInterface TheutilityofsoftwarelibrariessuchasOpenBabel dependsontheabilityofthedesigntobeextendedover timetosupportnewfunction ality.Tofacilitatethis, OpenBabelimplementsa plugininterface forfilefor- mats,fingerprints,char gemodels,descriptors, “ opera- tors ” andmolecularmechanicsforcefields.Thisensures acleanseparationoftheimplementationofaparticular pluginfromthecoreOpenBabellibrarycode,and makesiteasyforanewplugin(e.g.anewfileformat)to becontributed;allthatisneededisasingleC++fileand atrivialchangetooneofthebuildfiles.Theoperator pluginsprovideaverygeneralmechanismforoperating onamolecule(e.g.energyminimizationor3Dcoordi- nategeneration)oronalistofmolecules(e.g.filtering orsorting)afterreadingbutbeforewriting. Pluginsaredynamicallyl oadedatruntime.This decreasestheoveralldiskandmemoryfootprintof OpenBabel,allowingexternaldeveloperstochoosepar- ticularfunctionalityneededfortheirapplicationand ignoreother,lessrelevantfeatures.Italsoallowsthe possibilityofathird-partydistributingpluginsseparately totheOpenBabeldistributiontoprovideadditional functionality. Open-SourceLicenseandOpenDevelopment OpenBabelisopen-sourcesoftware,whichoffersend usersandthird-partydevelopersarangeofadditional rightsnotgrantedbyproprietarychemistrysoftware. Open-sourcesoftware,atitsmostbasiclevel,grants userstherightstostudyhowtheirsoftwareworks,to adaptitforanypurposeorotherwisemodifyit,andto sharethesoftwareandtheirmodificationswithothers. Inthissense,OpenSourcefunctionsinsimilarwaysto theprocessesofopenpeerreview,publication,and citationinscience.Therightsgrantedbyopensource licenseslargelycoincidewiththenormsofscientific ethicstoenableverifiability,repeatability,andbuilding onpreviousresultsandtheories. Beyondtheserights,OpenBabel(likemostother open-sourceprojects)offersopendevelopment – thatis, alldevelopmentoccursinpublicforumsandwithpublic coderepositories.Thisresultsingreaterinputfromthe communityasanyusercaneasilysubmitbugreportsor featuresuggestions,getin volvedindiscussionsonthe futuredirectionofOpenBabelorevenbecomeadevel- operhim/herself.Inpractice,thenumberofactivecon- tributorshasincreasedovertimethroughthislevelof open,publicdevelopment(Figure3).Moreover,it meansthatthedevelopmentofthecodeiscompletely transparentandthequalityofthesoftwareisavailable forpublicscrutiny.Indeed,sinceitsinception,over658 bugshavebeensubmittedtothepublictrackerand fixed[41]. ValidationandTesting OpenBabelincludesanextensivetestsuitecomprising 60differenttestprogramseachwithtenstohundredsof tests.Inearly2010,anight lybuildinfrastructureand dashboardwasputinplacewithsupportfromKitware, Inc.Thishasgreatlyimprovedcodequalitybycatching regressions,andalsoensuresthatthecodecompiles cleanlyonallplatformsandcompilerssupportedby OpenBabel.Someexamplesofteststhatareruneach nightare: (1)TheMMFF94forcefieldcodeistestedagainstthe MMFF94validationsuite. Figure3 Numberofcontributorsovertime .Notethatthisgraph onlyincludesdeveloperswhodirectlycommitedcodetotheOpen Babelsourcecoderepository,anddoesnotincludepatches providedbyusers. O ’ Boyle etal . JournalofCheminformatics 2011, 3 :33 http://www.jcheminf.com/content/3/1/33 Page6of14 (2)TheOBAlignclass,whichwasdevelopedusing Test-DrivenDevelopment(TDD)methodology,is runagainstitstestsuite. (3)Handlingofsymmetryisvalidatedbyconverting severaltestcasesbetweenSMILES,2Dand3DSDF, andInChI(therearealsoseveraltestprogramswith unittestsfortheindividualstereoclassesinthe API). (4)TheSMARTSparseristestedusingover250 validandinvalidSMARTSpatterns,andthe SMARTSmatcheristestedusing125basic SMARTSpatterns. (5)TheLSSR(LeastSetofSmallestRings)codeis testedforinvarianceagainstchangingtheatom orderforaseriesofpolycyclicmolecules. Recentlythedevelopmentteamhasplacedamajor focusonincreasingtherobustnessoffileformattransla- tionparticularlyinrelationtothecommonlyused SMILESandMDLMolfileformats.Translatingbetween theseformatsrequiresaccuratestereochemistrypercep- tion,inferenceofimplicithydrogens,andkekulizationof delocalizedsystems.Whileitisdifficulttoensurethat anycomplexpieceofcodeisfreeofbugs,andOpen Babelisnoexception,validationprocedurescanbecar- riedouttoassessthecurrentlevelofperformanceand tofindadditionaltestcasesthatexposebugs.Thefol- lowingprocedurewasusedtoguidetherewritingof stereochemistrycodeinOpenBabel,aprojectthat beganinearly2009.Startingwithadatasetof18,084 3DstructuresfromPubChem3DasanSDFfile,we comparedtheresultof(a)conversiontoSMILES,fol- lowedbyconversionofthattoCanonicalSMILESto(b) conversiondirectlytoCanonicalSMILES.Thisproce- durecanbeusedtoflushouterrorsinreadingtheori- ginalSDFfile,reading/writingSMILES(eitherdueto stereochemistryerrorsorkekulizationproblems),andis alsoatest(tosomeextent)ofthecanonicalizationcode. Atthetimeofstartingthiswork(March2009),the errorratefoundwas1424(8%);byOct2009,combined workonstereochemistry,kekulizationandcanonicaliza- tionhadreducedthisto190(~1%),andcontinued improvementshavereducedthenumberoferrorsdown totwo(showninFigure4)forOpenBabel2.3.1 (~0.01%).Thefirstfailureisduetoakekulizationerror inapolycyclicaromaticmoleculeincorporatingheteroa- toms:(a)gavec1ccc2c(c1)c1[nH][nH]c3c4c1c(c2) ccc4cc1c3cccc1while(b)gavec1ccc2c(c1)c1nnc3c4c1c (c2)ccc4cc1c3cccc1.Thiserrorledtoconfusionover whetherornotthearomaticnitrogenshavehydrogens attached(theydonot).Thesecondfailureinvolvescon- fusionoverthecanonicalstereochemistryatabridge- headcarbon:(a)gaveC1CN2[C@@H](C1)CCC2while (b)gaveC1CN2[C@H](C1)CCC2.Thisisactuallya mesocompoundandsobothSMILESstringsarecor- rectandrepresentthesamemolecule.Howeverthe canonicalizationalgorithmshouldhavechosenone stereochemistryortheotherforthecanonical representation. Anotherareaoffocuswast hecanonicalizationalgo- rithm,whichcanbeusedtogeneratecanonicalSMILES aswellasotherformats.Thealgorithmcanbetestedby ensuringthatthesamecanonicalSMILESstringis obtainedevenwhentheorderofatomsinamoleculeis changed(whileretainingthesameconnectiontable). Theteststressesallareasofthelibrary,includingaro- maticityperception,kekuliz ation,stereochemistry,and canonicalization.Thedeve lopmentofthecanonicaliza- tioncodeinOpenBabelwasguidedbyapplyingthis testtothe5,151,179moleculesintheeMoleculescatalo- gue(dated2011-01-02)with10randomshufflesofthe atomorder.AtthetimeoftheOpenBabel2.2.3release, therewere24,404failuresofthecanonicalizationalgo- rithm;thishasnowbeenreducedtoonlyfour(shown inFigure5,0.001%).TheOpenBabelnightlytest suiteensuresthatthistestpassesforanumberofpro- blematicmolecules.Althoughthecanonicalizationalgo- rithmisstillnotperfect,webelievethatthecurrent levelofperformance(99.99992%successontheeMole- culescatalogue)isacceptableforgeneraluseandwith timeweintendtoimproveperformancefurther. Giventhattheerrorrateforcanonicalizationand handlingofstereochemistryisnowquitelow,thenext areaoffocusfortheOpenBabeldevelopmentteamis toimprovethehandlingofimplicitvalencefor “ unusual atoms. ” Thisisparticularlyimportantfororganometallic speciesandinorganiccomplexes. UsingOpenBabel Applications TheOpenBabelpackageiscomposedofasetofuser applicationsaswellasaprogramminglibrary.Themain commandlineapplicationprovidedis obabel (asmall upgradeontheearlier babel ),whichfacilitatesfilefor- matconversion,filtering(bySMARTS,title,descriptor value,orpropertyfield),3Dor2Dstructuregeneration, Figure4 Thetwofailuresfoundinthevalidationtestfor reading/writingSMILES . O ’ Boyle etal . JournalofCheminformatics 2011, 3 :33 http://www.jcheminf.com/content/3/1/33 Page7of14 conversionofhydrogensfro mimplicittoexplicit(and viceversa),andremovalofsmallfragmentsorofdupli- catestructures.Anumberoffeaturesareprovidedto handlemulti-moleculef ileformats(suchasSDFor MOL2)andtouseormanipulatetheinformationin propertyfieldsandmoleculetitles.Hereisanexample ofusing obabel toconvertfromSDFformattoSMILES: obabelinputmols.sdf-Ooutputmols.smi Amorecomplicatedusewouldbetoextractallmole- culesinanSDFfilewhosetitlesstartwith “ active": obabelinputmols.sdf-aT-ocopy-Oout- putmols.sdf – filter “ title= ’ active* ’” The copy formatspecifiedby “ -ocopy ” isautilityfor- matthatcopiestheexactcontentsoftheinputfile(for thefilteredmolecules)directlytotheoutput,without perceptionorinterpretation.The “ -aT ” indicatesthat onlythetitleoftheinputSDFfileshouldberead;full chemicalperceptionisnotrequired. TheOpenBabelgraphicaluserinterface(GUI)pro- videsthesamefunctionality.Figure6isascreenshotof theGUIcarryingoutthesamefilteringoperation describedinthe obabel exampleabove.Theleftpanel dealswithsettinguptheinputfile,therightpanelhan- dlestheoutputandthecentralpanelisforsettingcon- versionoptions.Dependingonwhetheraparticular optionrequiresaparameter,theavailableoptionsare displayedeitherascheckboxesorastextentryboxes. Theseinterfaceelements aregenerateddynamically directlyfromthetextdescriptionandhelptextprovided byeachformatplugin. ProgrammingLibrary TheOpenBabellibraryallowsuserstowritechemistry applicationswithoutworryingaboutthelow-leveldetails ofhandlingchemicalinformation,suchashowtoread orwriteaparticularfileformat,orhowtouseSMARTS forsubstructuresearching.Instead,theusercanfocus onthescientificproblemathand,oroncreatingamore easy-to-useinterface(e.g.aGUI)tosomeofOpen Babel ’ sfunctionality.TheOpenBabelAPI(Application ProgrammingInterface)isthesetofclasses,methods andvariablesprovidedbyOpenBabeltotheuserfor useinprograms.DocumentationonthecompleteAPI (generatedusingDoxygen[42])isavailablefromthe OpenBabelwebsite[43],orcanbegeneratedfromthe sourcecode. ThefunctionalityprovidedbytheOpenBabellibrary isrelieduponbymanyusersandbyseveralothersoft- wareprojects,withtheresultthatintroducingchanges totheAPIwouldcauseexistingsoftwaretobreak.For thisreason,OpenBabelstrivestomaintainAPIstabi- lityoverlongperiodsoftime,sothatexistingsoftware willcontinuetoworkdespitethereleaseofnewOpen Babelversionswithadditionalfeatures,fileformats andbugfixes.OpenBabelusesaversionnumbering systemthatindicateshowtheAPIhaschangedwith everyrelease:  Bugfixreleases(e.g.2.0.0versus2.0.1)donot changeAPIatall  Minorversionreleases(e.g.2.0versus2.1)willadd totheAPI,butwillotherwisebebackwards- compatible  Majorversionreleases(e.g.2versus3)arenot backwards-compatible,andhavechangestotheAPI (includingremovalofdeprecatedclassesand functions) Figure7showsanexampleC++programthatusesthe twomainclassesOBConversionandOBMoltoprint outthemolecularweightofallofthemoleculesinan SDFfile.Thiscouldbeused,forexample,toinvestigate differencesinthemolecularweightdistributionbetween twodatabases.ThesameprogramisshowninFigure8 butimplementedusingthePythonbindings. ExamplesofUse OpenBabelhasalreadybeenreferencedover400times forvarioususes.ThemostcommonuseofOpenBabel isthroughthe obabel commandlineapplication(orthe correspondinggraphicaluserinterface)fortheintercon- versionofchemicalfileformats.Suchconversionsmay alsoinvolvethecalculationorinferenceofadditional molecularinformationorapplicationofafilter.Some Figure5 Thefourfailuresfoundinthevalidationtestforcanonicalization . O ’ Boyle etal . JournalofCheminformatics 2011, 3 :33 http://www.jcheminf.com/content/3/1/33 Page8of14 publishedexamplesoftheseincludethefollowing:  interconversionofchemicalfileformatsorrepre- sentations[44-47]  additionofhydrogens[48-50]  generationof3Dmolecularstructures[51-53]  calculationofpartialcharges[54,55]  generationofmolecularfingerprints[56-59]  removalofduplicatemoleculesfromadataset[60]  calculationofMOL2atomtypes[61] Aninterestingexampletha tshowshowaparticular chemicalrepresentationmaybeusedtofacilitatea scientificstudyisthecrystallographicstudyofFábián andBrockwhousedOpenBabeltogenerateInChI stringsformoleculesintheCambridgeStructuralData- base[62].ExploitingthefactthatInChIsofenantiomers areidenticalexpectattheenantiomersublayer("/m0 ” Figure6 ScreenshotoftheOpenBabelGUI .Inthescreenshot,theOpenBabelGUIisrunningonBio-Linux6.0,anUbuntuderivative. Figure7 ExampleC++programthatusestheOpenBabel library .Theprogramprintsoutthemolecularweightofeach moleculeintheSDFfile “ dataset.sdf ” . Figure8 ExamplePythonprogramthatusestheOpenBabel library .Theprogramprintsoutthemolecularweightofeach moleculeintheSDFfile “ dataset.sdf ” . O ’ Boyle etal . JournalofCheminformatics 2011, 3 :33 http://www.jcheminf.com/content/3/1/33 Page9of14 Table1SoftwareapplicationsandlibrariesthatuseOpenBabelReferenceWebpageGUIformolecularmodellingandcomputationalchemistryG.HutchisonM.Hanwellhttp://avogadro.openmolecules.net/Parsecomputationalchemistryoutputfiles[72]http://cclib.sf.net/GUIforcomputationalchemistryJensThomashttp://www.cse.scitech.ac.uk/ccg/software/ManageachemicallaboratorydatabaseRémyDernathttp://chemaztech.sf.net/ChemSpotlightChemistryfileindexerforMacOSXG.Hutchisonhttp://chemspotlight.openmolecules.net/GUIforgeneratingcombinatoriallibrariesRuiAbreuhttp://www.esa.ipb.pt/~ruiabreu/chemt2Dmoleculardrawing[73]http://ruby.chemie.uni-freiburg.de/~martin/Libraryforhandlingandpreparingmulti-scalemulti-paradigm[74]http://web.mit.edu/mbuehler/www/research/CMDF/CMDF.htmSystematicallygenerateconformers[36]http://confab.googlecode.com/DockoMaticAutomatethepreparationandanalysisofAutoDockruns[75]http://sf.net/projects/dockomatic/DOVIS2.0AutomatethepreparationandanalysisofAutoDockruns[76]http://www.bhsai.org/dovis.htmlFAF-Drugs2ADMETfilteringofmoleculardatasets[77]http://www.mti.univ-paris-diderot.fr/fr/downloads.htmlLarge-scalechemicalgraphminingbasedonbackbonerefinementclasses[78,79]http://www.maunz.de/wordpress/bbrcGUIforcomputationalchemistryhttp://www.uku.fi/~thassine/projects/ChemistryUtils2Dchemicaleditor,3Dviewer,chemicalcalculatorandperiodictableforLinuxJeanBréforthttp://gchemutils.nongnu.org/MacOSXinterfacetoOpenBabelandotherOpenchemistrytoolsChrisSwainhttp://homepage.mac.com/swain/Sites/Macinchem/page65/ibabel3.htmlGUIshowinginformationontheperiodictableoftheelementsCarstenhttp://edu.kde.org/kalzium/LazyStructure-ActivityRelationshipsfortoxicityprediction[80]http://www.in-silico.de/software/GUIforcomputationalchemistryUgoVarettohttp://molekel.cscs.ch/molsKetch2DchemicaleditorHarmvanhttp://molsketch.sf.net/ChemistryextensiontotheMySQLdatabaseJ.Pansanelhttp://mychem.sf.net/NanoEngineer-Computer-aideddesignforthenanoscaleNanorex,Inc.http://nanoengineer-1.net/NanoHive-1Simulatorforthestudy,experimentation,anddevelopmentofnanotechentitiesBrianHelfrichhttp://www.nanohive-1.org/OpenSourcemoleculardynamicsengine[81]http://openmd.net/Open3DQSARHigh-throughputchemometricanalysisofmolecularinteractionfields[82,83]http://www.open3dqsar.org/Extractschemicalstructuresfromimages[84]http://osra.sf.net/ChemistryextensiontothePostgreSQLdatabasehttp://pgfoundry.org/projects/pgchemPharmacophorediscoveryandsearchingSilicosNVhttp://www.silicos.be/Pharmacophoresearching[85]http://smoothdock.ccbb.pitt.edu/pharmerShape-basedalignmentofmoleculesSilicosNVhttp://www.silicos.be/Libraryforhandlingandpreparingquantummechanicalmulti-scalesimulations[86]http://www.ipc.kit.edu/cfn-ysg/158.phpGUIforvirtualscreeningwithprotein-liganddockinghttp://pyrx.scripps.edu/GUIforanalysingresultsofquantumchemistrycalculations[72]http://qmforge.sf.net/ReactionMechanismGenerator[87]http://rmg.sf.net/evisualizationof3Dmodelsofscientificdata,suchasmolecularstructuresandsurfacesT.J.ODonnellhttp://sci3d.sf.net/FiltermoleculesfromdatasetsSilicosNVhttp://www.silicos.be/Generationoffragment-basedstructure-activityrelationships[88]http://www.karwath.org/systems/smirep.htmlExtractmolecularscaffoldsSilicosNVhttp://www.silicos.be/etalJournalofCheminformaticshttp://www.jcheminf.com/content/3/1/33Page10of14 ),theyusedtheInChIsaspartofaworkflowtoidentifykryptoracemates(aclassofracemiccrystalswheretheenantiomersarenotrelatedbyspace-groupsymmetry)inthedatabase.Toimplementnewmethods,oraccessadditionalmole-cularinformation,itisnecessarytousetheOpenBabellibrarydirectlyeitherfromC++orusingoneofthesup-portedlanguagebindings.Someexamplesofpublishedstudiesthathavedonethisincludethefollowing:etal.implementedmolecularcomplexitymeasuresbasedoninformationtheory[63].LanghamandJaindevelopedamodelforchemicalmutagenicitybasedonatompairfeatures[64].etal.implementedamethod,anchor-GRIND,thatusesananchorpointofamolecularscaffoldtocomparemolecularinteractionfieldswhendifferentsubstituentsarepresent[65].etal.havedevelopedapluginforOpenBabelthataddssupportfortheWebOntologyLan-guage(OWL)toallowautomatedreasoningaboutchemicalstructures[66].etal.(AstraZeneca)implementeda3-pointpharmacophorefingerprintcalledTRUST[67]. Table1SoftwareapplicationsandlibrariesthatuseOpenBabelToxichazardestimationusingdecisiontreesIdeaconsulthttp://toxtree.sf.net/VisualizeatomicstructuressuchascrystalsandgrainboundariesDamienCalistehttp://inac.cea.fr/L_Sim/V_Sim/index.en.htmlWebapplicationforfileformatconversionT.J.ODonnellhttp://webbabel.sf.net/2DmoleculareditorBryanHergerhttp://xdrawchem.sf.net/ExtensiontoAvogadroforcrystal-structureprediction[89]http://xtalopt.openmolecules.net/GUIformoleculargraphics,modelingandsimulationElmarKriegerhttp://www.yasara.org/GUIformolecularmodellinganddocking[90]http://www.zeden.org/ Table2WebapplicationsanddatabasesthatuseOpenBabelNameDescriptionReferenceWebpageDatabaseofsmallmolecules[91]http://cdb.ics.uci.edu/ChemicalstructureandpropertysearchengineCéondoLtdhttp://www.chemeo.com/Webapplicationforanalysingandclusteringsmallmolecules[92]http://chemmine.ucr.edu/Chemicalvendorsearchenginehttp://emolecules.com/FragmentStoreDatabaseforcomparisonoffragmentsfoundinmetabolites,drugsandtoxiccompounds[93]http://bioinf-applied.charite.de/fragment_store/FReeOnlinedruG3Dconformationgeneration[94]http://bioserv.rpbs.univ-paris-diderot.fr/cgi-hBarLabWebapplicationprovidingon-demandaccesstocomputer-aidedhBarSolutionshttps://www.hbar-lab.com/Databaseofhumandrugtargetsandtheirligands[95]http://www.iuphar-db.org/OpenCDLigWebapplicationforsharingresourcesaboutcyclodextrin/ligand[96]https://kdd.di.unito.it/casmedchem/Protein-Small-MoleculeDatabase[97]http://compbio.cs.toronto.edu/psmdb/Webapplicationforcalculationofburiedvolumeoforganometallic[98]https://www.molnac.unisa.it/OMtools/Databaseofmolecularscaffolds[99]http://202.127.30.184:8080/scafbank.htmlWebapplicationforpredictionofsitesofcytochromeP450mediatedmetabolism[100]http://www.farma.ku.dk/smartcyp/sMolExplorerWebapplicationforexploringsmall-moleculedatasets[101]http://www3a.biotec.or.th/isl/index.php/smol-SuperImposéWebapplicationforstructuralsimilaritybetweenligands,bindingsitesorproteins[102]http://farnsworth.charite.de/superimpose-Databaseoftoxiccompounds[103]http://bioinformatics.charite.de/supertoxic/Detailedinformationon,andcomparisonsof,protein-ligandbindingsites[104]http://bioinf-tomcat.charite.de/supersite/Databaseofnaturalandartificialsweeteners[105]http://bioinf-applied.charite.de/sweet/Chemical-proteininteractions[106]http://stitch.embl.de/VirtualComputationalChemistryLaboratory[107]http://www.vcclab.org/wwLigCSRreWebapplicationthatperformsligand-basedscreeningusing3D[108]http://bioserv.rpbs.univ-paris-diderot.fr/Help/wwLigCSRre.htmletalJournalofCheminformaticshttp://www.jcheminf.com/content/3/1/33Page11of14 Manyotherexamplesexist[68-71].ThevitalrolethatacheminformaticstoolkitplaysinthedevelopmentofscientificresourcesisshownbyTables1and2.Table1listsexamplesofstand-aloneapplicationsorprogramminglibrariesthatrelyonOpenBabel,eithercallingthelibrarydirectlyorviaoneofthecommand-lineexecutables.Table2containsexamplesofwebapplicationsanddatabasesthateitheruseOpenBabelontheserverorwhereOpenBabelwasusedinthepreparationofthedata.ConclusionsInNovember2011,OpenBabelwillmark10yearsofexistenceasanindependentproject,andforthefirsttime,wehavediscusseditsdevelopmentandfeatures.Asshownbymorethan400citations,ithasbecomeanessentialtoolforhandlingthemyriadofmolecularfileformatsencounteredindiversebranchesofchemistry.Whilemoreworkremainstobedone,throughvalida-tionprocessessuchasthosedescribedaboveandtherecentintroductionofanightlybuildandtestingframe-work,weaimtoimprovethequalityandrobustnessofthetoolkitwitheachnewrelease.Lookingforwardtothefuture,oneofthegoalsoftheprojectistoextendsupporttomoleculesthatcurrentlyarenothandledverywellbyexistingcheminformaticstoolkits.Typicallytoolkitsfocusonthetypesofmole-culesofprincipalimportancetothepharmaceuticalindustry,namelystableorganicmoleculescomprisingwhollyof2-center2-electroncovalentbonds.Moleculesoutsidethisset-suchasradicals,organometallicandinorganicmolecules,moleculeswithcoordinatebondsor3-center2-electronbonds-arepoorlysupportedingeneral.FuturereleasesofOpenBabelwillprovidesub-stantiallyimprovedhandlingofsuchspecies.Wealsoseektoimprovespeedandcoverageofimportantmeth-odssuchasstructuregeneration,kekulizationandOpenBabelisfreelyavailablefromhttp://openbabel.org,andnewcommunitymembersareverywelcome(users,developers,bugreporters,featurerequesters).ForinformationonhowtouseOpenBabel,pleaseseethedocumentationathttp://openbabel.org/docsandtheAPIdocumentationathttp://openbabel.org/api.AvailabilityandRequirementsProjectName:OpenBabelProjecthomepage:Operatingsystem(s):Programminglanguage:C++,bindingstoPython,Perl,Ruby,Java,C#Otherrequirements(ifcompiling):CMake2.4+GNUGPLv2Anyrestrictionstousebynon-academics:AcknowledgementsandFundingWewouldliketothankallusersandcontributorstotheOpenBabelprojectoveritshistory,includingOpenEyeScientificSoftwareInc.fortheirinitialOELibcode.WealsothanktheBlueObeliskMovementforideas,commentsonthismanuscript,andsupport.WethankSourceForgeforprovidingresourcesforissuetrackingandmanagingreleases,andKitwareforadditionaldashboardresources.NMOBissupportedbyaHealthResearchBoardCareerDevelopmentFellowship(PD/2009/13).AuthordetailsAnalyticalandBiologicalChemistryResearchFacility,CavanaghPharmacyBuilding,UniversityCollegeCork,Co.Cork,Ireland.DepartmentofChemistry,TechnischeUniversitätMünchen,GarchingD-85747,Germany.eMolecules,Inc.,420StevensAve#120,SolanaBeach,CA92075,USA.OpenBabeldevelopmentteam.UniversityofPittsburgh,DepartmentofChemistry,219ParkmanAvenue,Pittsburgh,PA15217,USA.GRHistheleaddeveloperoftheOpenBabelproject.CAJ,CM,MB,NMOB,andTVaredevelopersofOpenBabel.Allauthorsreadandapprovedthefinalmanuscript.CompetinginterestsTheauthorsdeclarethattheyhavenocompetinginterests.Received:27June2011Accepted:7October2011Published:7October20111.WeiningerD:SMILES,achemicallanguageandinformationsystem.1.Introductiontomethodologyandencodingrules.JChemInfComputSci2.Murray-RustP,RzepaH:Chemicalmarkup,XML,andtheWorldwideWeb.1.Basicprinciples.JChemInfComputSci3.Murray-RustP,RzepaHS:ChemicalMarkup,XMLandtheWorld-WideWeb.2.InformationObjectsandtheCMLDOM.JChemInfModel4.Murray-RustP,RzepaH,WrightM:Developmentofchemicalmarkuplanguage(CML)asasystemforhandlingcomplexchemicalcontent.NewJChem5.Murray-RustP,RzepaH:ChemicalMarkup,XML,andtheWorldWideWeb.4.CMLSchema.JChemInfComputSci6.HollidayGL,Murray-RustP,RzepaHS:ChemicalMarkup,XML,andtheWorldWideWeb.6.CMLReact,anXMLVocabularyforChemicalJChemInfModel7.DaylightTheory::,SMARTShttp://www.daylight.com/dayhtml/doc/theory/8.FogelK:ProducingOpenSourceSoftware:HowtoRunaSuccessfulFreeSoftwareProjectReillyMedia,Inc.Sebastopol,CA;2005.9.CitationsweregeneratedbyGoogleScholar:[http://scholar.google.com/10.Aselectionofsuchprojectsisincludedbelow.:,Thefulllistisavailableat:11.OpenBabel::[http://openbabel.org/].12.OpenBabelReportFormat::[http://openbabel.org/docs/2.3.0/FileFormats/13.OpenBabelFingerprintFormat::[http://openbabel.org/docs/2.3.0/14.OpenBabelFastsearchFormat::[http://openbabel.org/docs/2.3.0/15.MolPrint2DFormat::[http://openbabel.org/docs/2.3.0/FileFormats/16.BenderA,MussaHY,GlenRC,ReilingS:MolecularSimilaritySearchingUsingAtomEnvironments,Information-BasedFeatureSelection,andaNaïveBayesianClassifier.JChemInfModel17.MNAFormat::[http://openbabel.org/docs/2.3.0/FileFormats/etalJournalofCheminformaticshttp://www.jcheminf.com/content/3/1/33Page12of14 18.FilimonovD,PoroikovV,BorodinaY,GloriozovaT:ChemicalSimilarityAssessmentthroughMultilevelNeighborhoodsofAtoms:DefinitionandComparisonwiththeOtherDescriptors.JChemInfModel19.PDBFormatv3.2::[http://www.wwpdb.org/documentation/format32/v3.2.20.PDB:CrufttoContent::[http://www.daylight.com/meetings/mug01/Sayle/21.MorganHL:TheGenerationofaUniqueMachineDescriptionforChemicalStructures-ATechniqueDevelopedatChemicalAbstractsJChemDocum22.Nauty::[http://cs.anu.edu.au/~bdm/nauty/].23.McKayBD:Practicalgraphisomorphism.CongressusNumerantium24.GakhA,BurnettM:ModularChemicalDescriptorLanguage(MCDL):Composition,connectivity,andsupplementarymodules.JChemInfComputSci25.TrepalinSV,YarkovAV,PletnevIV,GakhAA:AJavaChemicalStructureEditorSupportingtheModularChemicalDescriptorLanguage(MCDL).26.GakhAA,BurnettMN,TrepalinSV,YarkovAV:ModularChemicalDescriptorLanguage(MCDL):Stereochemicalmodules.JCheminf27.HalgrenT:Merckmolecularforcefield.1.Basis,form,scope,parameterization,andperformanceofMMFF94.JComputChem28.HalgrenT:Merckmolecularforcefield.2.MMFF94vanderWaalsandelectrostaticparametersforintermolecularinteractions.JComputChem29.HalgrenT:Merckmolecularforcefield.3.MoleculargeometriesandvibrationalfrequenciesforMMFF94.JComputChem30.HalgrenT,NachbarR:Merckmolecularforcefield.4.ConformationalenergiesandgeometriesforMMFF94.JComputChem31.HalgrenT:Merckmolecularforcefield.5.ExtensionofMMFF94usingexperimentaldata,additionalcomputationaldata,andempiricalrules.ComputChem32.AndronicoA,RandallA,BenzRW,BaldiP:Data-drivenhigh-throughputpredictionofthe3-Dstructureofsmallmolecules:reviewandprogress.JChemInfModel33.RappeA,CasewitC,ColwellK,GoddardWIII,SkiffWM:UFF,afullperiodictableforcefieldformolecularmechanicsandmoleculardynamicsJAmChemSoc34.WangJ,WolfRM,CaldwellJW,KollmanPA,CaseDA:Developmentandtestingofageneralamberforcefield.JComputChemWangJ,WangW,KollmanPA,CaseDA:Automaticatomtypeandbondtypeperceptioninmolecularmechanicalcalculations.JMolecGraph36.OBoyleNM,VandermeerschT,FlynnCJ,MaguireAR,HutchisonGR:-Systematicgenerationofdiverselow-energyconformers.JCheminf37.CMake::[http://www.cmake.org/].38.MartinK,HoffmanB:MasteringCMake:ACross-PlatformBuildSystem.Kitware,Inc.,CliftonPark,NY;,52010.39.CDashDashboardforOpenBabel::[http://my.cdash.org/index.php?40.OBoyleN,MorleyC,HutchisonGR:Pybel:aPythonwrapperfortheOpenBabelcheminformaticstoolkit.ChemCentJ41.OpenBabelBugTracker::[https://sourceforge.net/tracker/?42.Doxygen::[http://www.doxygen.org/].43.OpenBabelAPI::[http://openbabel.org/api].44.MyersJ,AllisonT,BittnerS,DidierB,FrenklachM,GreenW,HoY,HewsonJ,KoeglerW,LansingC,etalAcollaborativeinformaticsinfrastructureformulti-scalescience.ClusterComputing45.LindP,AlmM:ADatabase-CentricVirtualChemistrySystem.JChemInf46.AminiA,ShrimptonPJ,MuggletonSH,SternbergMJE:Ageneralapproachfordevelopingsystem-specificfunctionstoscoreprotein-liganddockedcomplexesusingsupportvectorinductivelogicprogramming.Struct,Funct,Bioinf47.ArborS,MarshallGR:Avirtuallibraryofconstrainedcyclictetrapeptidesthatmimicsallfourside-chainorientationsforoverhalfthereverseturnsintheproteindatabank.JComput-AidedMolDes48.HuangZ,WongCF:AMiningMinimaApproachtoExploringtheDockingPathwaysofp-NitrocatecholSulfatetoYopH.BiophysJ49.HillAD,ReillyPJ:AGibbsfreeenergycorrelationforautomateddockingofcarbohydrates.JComputChem50.ArmenRS,ChenJ,BrooksCLIII:AnEvaluationofExplicitReceptorFlexibilityinMolecularDockingUsingMolecularDynamicsandTorsionAngleMolecularDynamics.JChemTheoryComp51.LiuL,MaH,YangN,TangY,GuoJ,TaoW,JaaDuan:ASeriesofNaturalFlavonoidsasThrombinInhibitors:Structure-activityrelationships.ThrombRes52.WallachI,JaitlyN,LilienR:AStructure-BasedApproachforMappingAdverseDrugReactionstothePerturbationofUnderlyingBiologicalPLoSOnePailaYD,TiwariS,SenguptaD,ChattopadhyayA:Molecularmodelingofthehumanserotonin1Areceptor:roleofmembranecholesterolinligandbindingofthereceptor.MolecularBioSystems54.MelvilleJL,HirstJD:TMACC:InterpretableCorrelationDescriptorsforQuantitativeStructureActivityRelationships.JChemInfModel55.PenchevaT,LagorceD,PajevaI,VilloutreixBO,MitevaMA:AutomatedMolecularMechanicsOptimizationtoolforinsilicoBMCBioinformatics56.SchietgatL,RamonJ,BruynoogheM:AnEfficientlyComputableGraph-BasedMetricfortheClassificationofSmallMolecules.Proceedingsofthe11thInternationalConferenceonDiscoveryScienceSpringer-VerlagBerlin,Heidelberg;2008,197-209.57.KrierM,HutterMC:BioisostericSimilarityofMoleculesBasedonStructuralAlignmentandObservedChemicalReplacementsinDrugs.ChemInfModel58.WangX,HuanJ,SmalterA,LushingtonGH:Applicationofkernelfunctionsforaccuratesimilaritysearchinlargechemicaldatabases.59.ChengT,LiQ,WangY,BryantSH:BinaryClassificationofAqueousSolubilityUsingSupportVectorMachineswithReductionandRecombinationFeatureSelection.JChemInfModel60.MihalevaVV,VerhoevenHA,deVosRCH,HallRD,vanHamRCHJ:AutomatedprocedureforcandidatecompoundselectioninGC-MSmetabolomicsbasedonpredictionofKovatsretentionindex.61.BasDC,RogersDM,JensenJH:VeryfastpredictionandrationalizationofpKavaluesforprotein-ligandcomplexes.Proteins:Struct,Funct,Bioinf62.FabianL,BrockCP:Alistoforganickryptoracemates.ActaCryst63.DehmerM,BarbariniN,VarmuzaK,GraberA:ALargeScaleAnalysisofInformation-TheoreticNetworkComplexityMeasuresUsingChemicalPLoSOne64.LanghamJJ,JainAN:AccurateandInterpretableComputationalModelingofChemicalMutagenicity.JChemInfModel65.FontaineF,PastorM,ZamoraI:Anchor-GRIND:Fillingthegapbetweenstandard3DQSARandtheGRid-INdependentDescriptors.JMedChem66.KonykM,DeLeonA,DumontierM:ChemicalknowledgeforthesemanticDataIntegrationintheLifeSciencesSpringer-VerlagBerlin,Heidelberg;2008,169-176.KogejT,EngkvistO,BlombergN,MuresanS:MultifingerprintBasedSimilaritySearchesforTargetedClassCompoundSelection.JChemInf68.ReynèsC,HostH,CamprouxA-C,LacondeG,LerouxF,MazarsA,DeprezB,FahraeusR,VilloutreixBO,SperandioO:DesigningFocusedChemicalLibrariesEnrichedinProtein-ProteinInteractionInhibitorsusingMachine-LearningMethods.PLoSComputationalBiology69.LagorceD,PenchevaT,VilloutreixBO,MitevaMA:DG-AMMOS:ANewtooltogenerate3DconformationofsmallmoleculesusingDistanceGeometryandAutomatedMolecularMechanicsOptimizationforinsilicoScreening.BMCChemicalBiologyetalJournalofCheminformaticshttp://www.jcheminf.com/content/3/1/33Page13of14 70.GómezMJ,PazosF,GuijarroFJ,deLorenzoV,ValenciaA:environmentalfateoforganicpollutantsthroughtheglobalmicrobialMolecularSystemsBiology71.KaziusJ,NijssenS,KokJ,BäckT,IJzermanAP:SubstructureMiningUsingElaborateChemicalRepresentation.JChemInfModel72.OBoyleNM,TenderholtAL,LangnerKM:cclib:Alibraryforpackage-independentcomputationalchemistryalgorithms.JComputChem73.BrüstleM:Chemtool-MolekülezeichnenmitdemPinguin.ausderChemie74.BuehlerM,DodsonJ,vanDuinA:TheComputationalMaterialsDesignFacility(CMDF):Apowerfulframeworkformulti-paradigmmulti-scaleMaterialsResearchSocietysymposiumproceedings75.BullockCW,JacobRB,McDougalOM,HampikianG,AndersenT:Dockomatic-automatedligandcreationanddocking.BMCResearch76.JiangX,KumarK,HuX,WallqvistA,ReifmanJ:DOVIS2.0:anefficientandeasytouseparallelvirtualscreeningtoolbasedonAutoDock4.0.CentJ77.LagorceD,SperandioO,GalonsH,MitevaMA,VilloutreixBO:FreeADME/toxfilteringtooltoassistdrugdiscoveryandchemicalbiologyprojects.BMCBioinformatics78.MaunzA,HelmaC,KramerS:Efficientminingforstructurallydiversesubgraphpatternsinlargemoleculardatabases.MachineLearning79.MaunzA,HelmaC,KramerS:Large-scalegraphminingusingbackbonerefinementclasses.Proceedingsofthe15thACMSIGKDDInternationalConferenceonKnowledgeDiscoveryandDataMining(KDD2009)ACMParis;2009,617-626.80.HelmaC:Lazystructure-activityrelationships(lazar)forthepredictionofrodentcarcinogenicityandSalmonellamutagenicity.MolDiversity81.MeinekeMA,VardemanCF,LinT,FennellCJ,GezelterJD:OOPSE:anobject-orientedparallelsimulationengineformoleculardynamics.ComputChem82.ToscoP,BalleT:Brute-forcepharmacophoreassessmentandscoringwithOpen3DQSAR.JCheminf3(Suppl1)83.ToscoP,BalleT:Open3DQSAR:anewopen-sourcesoftwareaimedathigh-throughputchemometricanalysisofmolecularinteractionfields.MolModel84.FilippovIV,NicklausMC:OpticalStructureRecognitionSoftwareToRecoverChemicalInformation:OSRA,AnOpenSourceSolution.JChemInfModel85.KoesDR,CamachoCJ:Pharmer:EfficientandExactPharmacophoreJChemInfModel86.JacobCR,BeyhanSM,BuloRE,GomesASP,GötzAW,KiewischK,SikkemaJ,VisscherL:PyADF-AscriptingframeworkformultiscalequantumJComputChem87.GreenHWilliam,AllenWJoshua,AshcraftWRobert,BeranJGregory,ClassACaleb,GaoConnie,FranklinGoldsmithC,HarperRMichael,JalanAmrit,MagoonRGregory,MatheuMDavid,MerchantSShamel,MoDJeffrey,PetwaySarah,RamanSumathy,SharmaSandeep,SongJing,VanGeemMKevin,WenJohn,WestHRichard,WongAndrew,WongHsi-Wu,YelvingtonEPaul,YuJoanna:RMG-ReactionMechanismGenerator2011[http://rmg.sourceforge.net/].88.KarwathA,DeRaedtL:SMIREP:PredictingChemicalActivityfromSMILES.JChemInfModel89.LonieDC,ZurekE:XTALOPT:Anopen-sourceevolutionaryalgorithmforcrystalstructureprediction.ComputPhysCommun90.ZontaN,GrimsteadIJ,AvisNJ,BrancaleA:Accessiblehaptictechnologyfordrugdesignapplications.JMolModel91.ChenJH,LinsteadE,SwamidassSJ,WangD,BaldiP:ChemDBupdatefull-textsearchandvirtualchemicalspace.92.BackmanTWH,CaoY,GirkeT:ChemMinetools:anonlineserviceforanalyzingandclusteringsmallmolecules.NucleicAcidsResServerissue)93.AhmedJ,WorthCL,ThabenP,MatzigC,BlasseC,DunkelM,PreissnerR:acomprehensivedatabaseoffragmentslinkingmetabolites,toxicmoleculesanddrugs.NucleicAcidsRes94.MitevaMA,GuyonF,TufferyP:Frog2:Efficient3Dconformationensemblegeneratorforsmallcompounds.NucleicAcidsRes95.SharmanJL,MpamhangaCP,SpeddingM,GermainP,StaelsB,DacquetC,LaudetV,HarmarAJ,NC-IUPHAR:IUPHAR-DB:newreceptorsandtoolsforeasysearchingandvisualizationofpharmacologicaldata.NucleicAcids96.EspositoR,ErmondiG,CaronG:OpenCDLig:afreewebapplicationforsharingresourcesaboutcyclodextrin/ligandcomplexes.JComput-AidedMolDes97.WallachI,LilienR:Theprotein-small-moleculedatabase,anon-redundantstructuralresourcefortheanalysisofprotein-ligandbinding.98.PoaterA,CosenzaB,CorreaA,GiudiceS,RagoneF,ScaranoV,CavalloL:SambVca:AWebApplicationfortheCalculationoftheBuriedVolumeofN-HeterocyclicCarbeneLigands.EurJInorgChem99.YanB-b,XueM-z,XiongB,LiuK,HuD-y,ShenJ-k:ScafBank:apubliccomprehensiveScaffolddatabasetosupportmolecularhopping.PharmacologicaSinica100.RydbergP,GloriamDE,OlsenL:TheSMARTCypcytochromeP450metabolismpredictionserver.101.IngsriswangS,PacharawongsakdaE:sMOLExplorer:anopensource,web-enableddatabaseandexplorationtoolforSmallMOLeculesdatasets.102.BauerRA,BournePE,FormellaA,FrommelC,GilleC,GoedeA,GuerlerA,HoppeA,KnappEW,PoschelT,etalSuperimpose:a3Dstructuralsuperpositionserver.NucleicAcidsRes103.SchmidtU,StruckS,GrueningB,HossbachJ,JaegerIS,ParolR,LindequistU,TeuscherE,PreissnerR:SuperToxic:acomprehensivedatabaseoftoxiccompounds.NucleicAcidsRes104.BauerRA,GuntherS,JansenD,HeegerC,ThabenPF,PreissnerR:dictionaryofmetaboliteanddrugbindingsitesinproteins.NucleicAcids105.AhmedJ,PreissnerS,DunkelM,WorthCL,EckertA,PreissnerR:aresourceonnaturalandartificialsweeteningagents.NucleicAcidsRes106.KuhnM,SzklarczykD,FranceschiniA,CampillosM,vonMeringC,JensenLJ,BeyerA,BorkP:STITCH2:aninteractionnetworkdatabaseforsmallmoleculesandproteins.NucleicAcidsRes107.TetkoIV,GasteigerJ,TodeschiniR,MauriA,LivingstoneD,ErtlP,PalyulinVA,RadchenkoEV,ZefirovNS,MakarenkoAS,etalComputationalChemistryLaboratory-DesignandDescription.Comput-AidedMolDes108.SperandioO,PetitjeanM,TufferyP:wwLigCSRre:a3Dligand-basedserverforhitidentificationandoptimization.NucleicAcidsResdoi:10.1186/1758-2946-3-33Citethisarticleas:etalOpenBabel:AnopenchemicalJournalofCheminformatics W. Jeffery Hurst,The Hershey Company. http://www.chemistrycentral.com/manuscript/ etalJournalofCheminformaticshttp://www.jcheminf.com/content/3/1/33Page14of14