资源描述:
《Interpretation of Nonlinear QSAR Models Applied to Ames Mutagenicity》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、J.Chem.Inf.Model.2009,49,255125582551InterpretationofNonlinearQSARModelsAppliedtoAmesMutagenicityDataLarsCarlsson,*ErnstAhlbergHelgee,andScottBoyerSafetyAssessment,AstraZenecaResearch&Development,43183Mo¨lndal,SwedenReceivedJune24,2009Amethodforlocalinter
2、pretationofQSARmodelsispresentedandappliedtoanAmesmutagenicitydataset.Intheworkpresented,localinterpretationofSupportVectorMachineandRandomForestmodelsisachievedbyretrievingthevariablecorrespondingtothelargestcomponentofthedecision-functiongradientatanypo
3、intinthemodel.Thiscontributiontothemodelisthevariablethatisregardedashavingthemostimportanceatthatparticularpointinthemodel.ThemethoddescribedhasbeenverifiedusingtwosetsofsimulateddataandAmesmutagenicitydata.Thisworkindicatesthatitispossibletointerpretnonl
4、inearmachine-learningmethods.Comparisontoaninterpretablelinearmethodisalsopresented.1.INTRODUCTIONTheremainderofthispaperisorganizedasfollows.Insection2methodsarepresenteddescribingtheretrievalofQuantitative-structureactivity-relationship(QSAR)modelsthelo
5、calbehavior,bothforanalyticalanddiscreteformula-arewidelyusedasawayofapproximatingafunctionaltionsofQSAR-modelfunctions.InSection3,twoexamplesrelationshipbetweenasetofdescriptorsandagivenhavebeenconstructedusingsimulateddataoverapredefinedendpoint.Modelint
6、erpretationisveryimportant,withoutitgrid;themethodhasalsobeenappliedtoAmesmutagenicitythepossibilityofmodifyingcompoundsisrestricted.Oftendata.Thepaperisconcludedbythediscussionandconclu-chemistsresorttomodelalgorithmsthatareeasytosionsinSection4.understa
7、nd,suchaslinearmodels.Nonlinearmodelsarecommonlyviewedashardtointerpret,andthesemodel2.METHODalgorithmsormodelsaresometimesreferredtoasblack-1,2boxmodels.SomephenomenawillmostlikelybemodeledAnyQSARmodelisanapproximationtoarelationbestbynonlinearmodels,and
8、thereisaneedforQSARbetweenbiologicalactivityandcompoundcharacteristicsandmodelsthatareeasytointerpret.canbeviewedasamathematicalfunction.Differentmachine-Forsomemodelalgorithmstherearewaystoextractlearningmethodshav