Giving Molecules an Identity. On the Interplay Between QSARs and Partial Order Ranking.docx

GivingMoleculesanIdentity.OntheInterplayBetweenQSARsandPartialOrderRankingMolecules2004,9,1010-1018moleculesISSN1420-3049parisonwithexperimentallywe11-characterized,structurallysimilarcompounds.ItisdisclosedthatexperimentallyWeIl-CharaCteriZedcompoundsmayserveassubstitutesforhighlytoxiccompoundsinexperimentalstudieswithoutexhibitingthesameextremetoxicity,whilefromanoverallviewpointtheyexhibitanalogousenvironmentalcharacteristics.Keywords:Noise-deficientQSARs：PartialOrderRanking;HasseDiagrams;Organo-phosphates;Nerveagents.IntroductionThelackofdataforthevastmajorityofexistingchemicalsiswel1knownandconstitutesobviouslyasignificantprobleminrelationtoe.g.,riskassessment.Thus,accordingtotheEuropeanCommissiononlyinthecaseofapproximately14%oftheHPV(HighProductionVolume)chemicalsontheEINECSlist,comprising100,116entries,theminimumrequireddataforevaluatingthechemicalswereavai!able.Forapproximately21%ofthecompoundsnodataatallconcerningtheirpotentialimpactontheenvironmentandhumanhealthwerefound1.InastudybytheDanishEPA2itwasconcludedthateveninmajorsourcesoftestdata,informationonselectedecotoxicologicaleffectscouldonlybefoundforverylimitednumberofthecompoundsontheEINECSlist(acutetoxiceffect:10.5%,reproductivedamage:2.2%,geneticdamage:3.2%,carcinogeniceffect:1.6%,effectontheaquaticenvironment:3.5%).Sinceintensiveandexperimentalevaluationsofchemicalsarcrathercostly3,andreferencestherein,QSRderiveddataforphysico-chemicalaswel1astoxicologicalMolecules2004,91011endpointsappearasanattractivealternative.However,althoughthelackofdatacanberemediedtoacertainextentthroughQSARmodeling,thiswillleaveuswiththepossibilityofcharacterizingthesinglemoleculesbasedonsingleparameters,suchassolubility,octanol-waterpartitioning,vaporpressure,biodegradation-andbioaccumulationpotential.However,toestablishanidentityforagivenmolecule,e.g.,asapotentialPBTsubstancerequirestakingseveralparametersintoaccountsimultaneously,i.e.,Persistence,BioaccumulationandToxicity.Inthepresentstudytheadvantageoususeofso-callednoise-deficientQSARs,developedusingdatafromexperimentallywe11-characterizedcompoundsasthetrainingset,asapreprocessingtooltoderivethedesiredendpointsforsubstanceswhereexperimentaldataarenotavailable.Subsequently,theseendpointswillbeappliedasdescriptorsinestablishingapartialorderingofcombinedsetsofcompounds,herebygivingtheexperimentallynotinvestigatedcompoundsanidentitybycomparingtostructurallyrelated,experimentallywel!-characterizedcompounds4,53.MethodsQSARInthepresentstudytheend-pointsaregeneratedthroughQSARmodeling,theEPISuitebeingtheprimarytool6.Togeneratenew1inearnoise-deficientQSARmodels,EPIgeneratedvaluesfor,e.g.,logSol,logK0W,logVPandlogH1.CarefurthertreatedbyestimatingtherelationshipsbetweentheEPIgenerateddataandavai!ableexperimentaldata7fortheaseriesofexperimentallywell-characterizedcompoundsinthetrainingset,thegeneralformulafortheend-points,Di,tobeusedbeingDi=aiDEPI+bi(1)DEPIistheEPIgeneratedend-pointvalueandaiandbibeingconstants.ThelogKOWvaluesgeneratedinthiswayaresubsequentlyusedtogeneratelogBCFvaluesaccordingtotheConnellformula8logBCF=6.910-3(logKow)1.8510-1(log4K)3+1.55(logKow)2ow4.181ogKow+4.72(2)Themodelwassomewhatmodified.Thus,a1ineardecreaseoflogBCFwithlogKOWwasassumedintherange1logK0W2.33,thelogBCF=0.5forlogKOW1,thelattervaluebeinginaccordancewithBCFWin6.SubsequentlydatafornotcharacterizedcompoundsarecalculatedbasedontheseformulaeandtheappropriateEPIgenerateddata.Inthepresentstudyatrainingsetconsistingofupto65organophosphorus(OP)insecticidesareapplied.Duetothelackofexperimentaldataforthetrainingsetcompoundswithregardstotheirbiodegradation,theaboveprocedurewasnotapplicabletothebiodegradationpotential,BDP3.Thus,dataonBDP3areusedasestimatedbytheappropriatemodulesintheEPlSuite.Molecules2004,91012PartialOrderRankingThetheoryofpartialorderrankingispresentedelsewhere9anditsapplicationinrelationtoQSRispresentedinpreviouspapers1013.Inbrief,PartialOrderRankingisasimpleprinciple,whichaprioriincludesastheonlymathematicalrelation.Ifasystemisconsidered,whichcanbedescribedbyaseriesofdescriptorspi,agivencompound,characterizedbythedescriptorspi（八）canbecomparedtoanothercompoundB,characterizedbythedescriptorspi(B),throughcomparisonofthesingledescriptors,respectively.Thus,compoundwillberankedhigherthancompoundB,i.e.,BA,ifatleastonedescriptorforAishigherthanthecorrespondingdescriptorforBandnodescriptorforislowerthanthecorrespondingdescriptorforB.If,ontheotherhand,pi()pi(B)fordescriptoriandpj()pj(B)fordescriptorj,AandBwillbedenotedincomparable.InmathematicaltermsthiscanbeexpressedasBApi(B)pi（八）foralli(3)Obviously,ifalldescriptorsforAareequaltothecorrespondingdescriptorsforB,i.e.,pi(B)=pi（八）foralli,thetwocompoundswillhaveidenticalrankandwillbeconsideredasequivalent.ItfurtherfollowsthatifABandBCthenAC.IfnorankcanbeestablishedbetweenandBthesecompoundsaredenotedasincomparable,i.e.theycannotbeassignedamutualorder.Inpartialorderrankingincontrasttostandardmultidimensionalstatisticalanalysis-neitherassump