高级应用程序MOEQSAR1讲义资料.ppt

AdvancedApplicationinMOEQSAROutlineQSAROverviewDescriptorcalculationDescriptorselection(PCA)DerivingQSARmodelsModelValidationQSARQuantitativeStructure-ActivityRelationship(QSAR)applicationscorrelateexperimentaldata().*)TheartofderivingaQSARmodelliesin:plexdimensionalityofrealityintoasfewdimensionsaspossiblewhilestillbeingabletogiveusefulpredictionsofspecificpropertiesfor&QualitativeQSARModelsinMOEBesidestheselectionofmostappropriatedescriptorsandameaningfulseparationofavailabledataintotrainingandtestsets,:QuantitativeapproachesincludelinearregressionmethodssuchasPartialLeastSquares(PLS)andponentRegression(PCR).Qualitativeapproachesincludeanon-linearbinaryfilterbasedonBayesianstatisticsaswellasadevelopmentModelvalidationDescriptorselectionDescriptorcalculationDescriptorCalculationInitialStepsinUnderstandingaDatasetTheinitialstepsininterpretinganexperimentaldatasetinvolve:BuildingpreliminaryStructureActivityRelationships Commonfragmentstoactives/inactivesLookingforpatternswithinthedatastructure Areclusterspresentinthedata?Evaluatingtherelativeimportanceofdescriptorsforapotentialmodel monality,anddiversitywithinthedata RobustnessinchemicalspaceStep1indataanalysis::yes/noflagsforpresenceorabsenceofcertainfeatures(likebitsinfingerprints–seebelow).Quantitative:numericalmeasuresofphysico-chemistryonly(2D)oralsoonconformation/3Dgeometry(3D).Fingerprintstypicallyconsistsofbitstringsofseveralhundredsoreventhousandsofindividualyes/encod