Data mining techniques for cancer detection using serum proteomic profiling

《Data mining techniques for cancer detection using serum proteomic profiling》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、ArtificialIntelligenceinMedicine(2004)32,71—83Dataminingtechniquesforcancerdetectionusingserumproteomicprofilinga,aaabLihuaLi*,HongTang,ZuobaoWu,JianliGong,MichaelGruidl,bbaJunZou,MelvynTockman,RobertA.ClarkaDepartmentofRadiology,CollegeofMedicine,H.LeeMoffittCa

2、ncerCenterandResearchInstitute,UniversityofSouthFlorida,Tampa,FL33612-4799,USAbDepartmentofInterdiciplinaryOncology,H.LeeMoffittCancerCenterandResearchInstitute,UniversityofSouthFlorida,Tampa,FL33612-4799,USAReceived29August2003;receivedinrevisedform30January2

3、004;accepted9March2004KEYWORDSSummaryObjective:PathologicalchangesinanorganortissuemaybereflectedinProteomics;Cancerproteomicpatternsinserum.Itispossiblethatuniqueserumproteomicpatternscoulddetection;Datamining;beusedtodiscriminatecancersamplesfromnon-canceron

4、es.DuetothecomplexityofStatisticaltesting;proteomicprofiling,ahigherorderanalysissuchasdataminingisneededtouncoverthedifferencesincomplexproteomicpatterns.Theobjectivesofthispaperare(1)toGeneticalgorithm;brieflyreviewtheapplicationofdataminingtechniquesinproteo

5、micsforcancerSupportvectormachinedetection/diagnosis;(2)toexploreanovelanalyticmethodwithdifferentfeatureselectionmethods;(3)tocomparetheresultsobtainedondifferentdatasetsandthatreportedbyPetricoinetal.intermsofdetectionperformanceandselectedproteomicpatterns

6、.Methodsandmaterial:ThreeserumSELDIMSdatasetswereusedinthisresearchtoidentifyserumproteomicpatternsthatdistinguishtheserumofovariancancercasesfromnon-cancercontrols.Asupportvectormachine-basedmethodisappliedinthisstudy,inwhichstatisticaltestingandgeneticalgor

7、ithm-basedmethodsareusedforfeatureselectionrespectively.Leave-one-outcrossvalidationwithreceiveroperatingcharacteristic(ROC)curveisusedforevaluationandcomparisonofcancerdetectionperformance.Resultsandconclusions:Theresultsshowedthat(1)dataminingtechniquescanb

8、esuccessfullyappliedtoovariancancerdetectionwithareasonablyhighperformance;(2)theclassificationusingfeaturesselectedbythegeneticalgorithmconsistentlyoutperformedthoseselectedbystatisticalt