基于群體智能的特征選擇算法在SELDI質(zhì)譜數(shù)據(jù)分析中的研究.pdf

1、江南大學(xué)碩士學(xué)位論文基于群體智能的特征選擇算法在SELDI質(zhì)譜數(shù)據(jù)分析中的研究姓名：張蓉申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：計(jì)算機(jī)應(yīng)用與技術(shù)指導(dǎo)教師：馮斌20090801AbstractApplyingfe甜ureselection(FS)techniquesinbioinformaticshasbecomearealprerequlslteformodelbuildingInparticularthehighdimensionalandsmall

2、s鋤plesizesnatureso士m姐ymodelingtasksinbioinformatics，goingfromsequence觚alysisovermlcroa仃aYanalyslstospectralanalysesandliteraturemininghasgivenrisetoawealthoffeatureselectiontechniquesbeingpresentedinthefieldSmalls鋤plesiz

3、esa11dtheirinheremnsko士lmpreclslonandoverfittingposeagreatchallengeformanymodelingproblemsinb101nfornlatlcsSpecificapplicationsinbioinformaticshaveledtoawealthofnewlyproposedMassspectrometry(MS)technologyisusedtomeasuret

4、hemixtureofproteins／peptidesofblol091caltissuesorfluids，suchasserumorurineSuchmeasurementscaIlbeusedto1dentlfydiseaserelatedpattems，whichholdpotentialforearlydiagnosis，prognosis，monltonngdlseaSeprogression，responsetotrea

5、tmentanddrugtargetresearchuomprehenslVeanalysesonSurface。enhancedlaserdesorption／ionizationtimeofflightmassspectrometry(SELDITOFMS)dataanalysesaremainlydiscussedinour、vorka11dmeappllcatlonofswarmintelligencealgorithmcomb

6、inedwithSVMinbiomarkefselectionisalsostudiedIntheworkThemaincontentsofthisdissertationareaSfollo、j姻：【1)Fhethesisr骼earchedfundamentalprincipleofSELDITOFMStecllllologya11dsummanzedVariousmethodsofitstwomainphases：preproces

7、singandbiomarkerselectionAnditsshortcomingsandprogressarediscussedhere【2)Kese般honfundamentalprincipleofAntColonyOptimizationAIgori廿1111(ACO)ParncleSwanllOptimizationAlgorithm(PSO)andtheirimprovedmethodspro、，idestheoretic

8、alprinciplesforfurtherlearning(3)Newmethodisraisedusingweightingfactoraspriorinfo蛐ationintheantcolonvop‘1蛐za‘10nse眥hingprocessCombinedwithsupportvectormachines(SVM)，it嘲appiledto1den缸fyrelevantserumproteomicbiomarkersExpe

9、rimentsproposedmethodhasstrongpowerindistinguishingcancerpatientsfromhealthyindivid砌s(41CombinedSVMwithQPSO，ACOandPSO，andusingthemodelsbiom破ersselectiontheexperimentsshowthatmodelbuiltbyQPSOachievednotonlyhighprediction∞