a short introduction to design-expert

AshortintroductiontoDesign-Expert--apowerfulDOEsoftwareBiochemicalEngineeringBackgroundMyresearchDsbAassistedproteinrefoldinginvitroDsbA(DisulfidebondformationproteinA)SeparationpurificationFermentationRefoldingFermentationAmpIPTGTemperatureMediumvolumeInoculumsizeShakingspeedInductiontimeTESMgSO4CasaminoacidNH4ClKH2PO4Na2HPO4YeastextractNaClTryptoneDsbAIECPurificationStepsizeLoadingamountpHInitialsaltconcentrationTheamountofpurifiedDsbAHowtocracktohardnutsoftheseproblemsOnefactoratonetime?NODOE(designofexperiment)ResorttostatisticalapproachADesignofExperiment(DOE)isastructured,organizedmethodfordeterminingtherelationshipbetweenfactors(Xs)affectingaprocessandtheoutputofthatprocess(Y).Design-expertDesign-expertsoftwaremakesiteasyforyoutoperformstatisticaldesignofexperiments(DOE).CreatingaDOE-designAnalysisOptimizationClickonthenewicontocreateanewdesignTherearefourtypestotheleftofthescreenFourtypesofdesignFactorialDesigns–Identifythevitalfactorsthataffectyourprocessorproduct.Thenyoucanmakebreakthroughimprovements.ResponseSurfaceMethods(RSM)–Findtheidealprocesssettings.Achieveoptimalperformance.Mixturedesigntechniques–Discovertheoptimalformulation.Combineddesigns-Combineprocessvariables,mixturecomponentsandcategoricalfactorsinonedesign!AsimplifiedprocesstoillustrateAfiltrationprocessinvolvedformaldehydeasaprocessingaid.ToreducetheuseofformaldehydewhileachievingahighfiltrationRate.GOALAtwo-levelfactorialdesignfactorsunitesLowlevel(-)Highlevel(+)A.TemperatureDegC2435B.PressurePsig1015C.ConcentrationPercent24D.StirRateRPM530Response:FiltrationRateATTENTIONInthefollowingdiscussion,wewillfocusonhowthesoftwarefunctionsontheprocessofDOE,whilesomeunfamiliarconceptsindispensableinpracticaloperationmightbeskippedover.Tobeginbuildingthedesignclickonthewhitesquarelabeled24Optionsnumberofreplicatesofthedesign,thenumberofblocksandthenumberofcenterpoints.Leavethemattheirdefaults.PleaseclickontheContinuebuttontomovetothefactorsscreen.Factorscanbeoftwodistincttypes-NumericorCategoric.Numericdatacomesfromacontinuousscalesuchastemperatureorpressure.Categoricdata,suchascatalysttypes,occursindistinctlevels.Youcanchangethetypesoffactorfromthedropdownlist.YouhavenowcompletedthefirstphaseofDOE-creatingthedesign.SavethefilePrintthesunsheetPerformexperimentRecordtheresponsesorder,whichprotectsagainstanylurkingfactorssuchastime,temperature,humidityorthelike.Youcanchangetheformatontheresponse,theinputfactors,orchangenamesorlevels.ConvertbetweenthetwomethodsofdisplayingDesign-Expertprovidestwomethodsofdisplayingthelevelsofthefactorsinadesign:Actuallevelsofthefactors,andcodedas-1forlowlevelsand+1forlowlevelsand+1forhighlevels.Thengotothenextstep:AnalysisTheTransformscreendisplaysalistofmathematicalfunctionsthatyoumayapplytoyourresponse.Design-Expertnotesthattheresponserangeis2.4-fold,whichfallsbelowtheratioof3wherepowertransformshavelittleeffect.ClickonEffectstoselecteffectstobeincludedinthemodel.Theprogramdisplaystheabsolutevalueofalleffectsonahalf-normalprobabilityplot.(Color-codingprovidesdetailonwhethertheeffectsarepositiveornegative.)Youcanselecteffectsbysimplyclickingonthesquarepoints.NoticethatDesign-Expertadjuststhelinetoexcludethechoseneffects.Atthepointyoushouldstop,thislinejumpsup,leavinganoticeablegap.Clickheretoseeanotherformofdisplay.ClickonANOVAtolookattheanalysisofvariance.Checktheprobability(p-value)fortheModel.Bydefault,Design-Expertconsidersvaluesof0.05tobesignificant.Right-clickingandchoosingHelptoobtainthePop-updefinitions.Scrolldowntoseethecontentassociatedtostatistics.Lastlytheendofthereportshowsthepredictiveequationsbothincodedanduncodedform.ClickonDiagnosticstogenerateanormalprobabilityplotoftheresiduals.Ideallythenormalplotofresidualswillbeastraightline,indicatingnoabnormalities.InthiscasetheplotslooksOKsoyoucanmoveon.Design-Expertprovidesupperandlowerredlinesthataresimilarto95%confidencecontrollimitsonarun-chart.Clickitthenyoucantrackitonotherformsofdisplay.Itremainshighlighted.ClickonReportHereyouseethenumericalvaluesfordiagnosticstatisticsreportedcase-by-caseinstandardorder.Discrepantvalueswillbeflagged.Inthiscasenothingisdetectedasbeingabnormal.ClickModelGraphtolookatthesignificantfactoreffects.ThecomparisonofspreadsbetweenthepointsonThetwosidesindicatethattheeffectofHCHOconcentration(C)islesssignificantatthehighleveloftemperature(A).Onlythetermsthatareinthemodelareincludedinthislist.Adjustfactorsheretoseecorrespondingresponse.Theoptimumoutcome.FiltrationrateismaximumattheA+,D+,C-settings,whichalsocorrespondstothereducedformaldehydeconcentration.Fantastic!Clickheretorotatethegraphsoastoviewfromvariantangle.Thelastfunction:PointPredictionTheFactorsToolwilldefaulttothecenterpointsofeachfactor.A(Temperature)righttoitshighlevel(+)B(Pressure)leaveatdefaultlevelofcenterpointC(Concentration)lefttoitslowlevel(-)D(StirRate)righttoitshighlevel(+)Besidestheresponsepredictionvalues,severalotherpiecesofinformationareprovided.Thiswillprovidethehighestpredictedfilt