文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > Relational subgroup discovery for gene expression data mining
侵权投诉

Relational subgroup discovery for gene expression data mining

Abstract: We propose a methodology for predictive classification from gene expression data, able to combine the robustness of highdimensional statistical classification methods with the comprehensibility and interpretability of simple logic-based models. W

RELATIONALSUBGROUPDISCOVERYFORGENE

EXPRESSIONDATAMINING

ˇˇep´F.Zelezn´y ,J.Tolar ,N.Lavraˇc andO.Stˇankov´a

GerstnerLaboratory,DepartmentofCybernetics,

CzechTechnicalUniversity,Technick´a2,

16627Prague,CzechRepublic

DivisionofHematology,Oncology,BloodandMarrowTransplant,Departmentof

Pediatrics,UniversityofMinnesotaMedicalSchool420DelawareStreet,55455Minneapolis,USA

DepartmentofKnowledgeTechnologies,JoˇzefStefanInstitute

Jamova39,1000Ljubljana,Slovenia

and

NovaGoricaPolytechnic,Vipavska13,5000NovaGorica,Slovenia

zelezny@fel.cvut.cz,tolar003@umn.edu,nada.lavrac@ijs.si,step@labe.felk.cvut.cz

Abstract:Weproposeamethodologyforpredictiveclassi cationfromgeneexpressiondata,abletocombinetherobustnessofhigh-dimensionalstatisticalclassi cationmethodswiththecomprehensibilityandinterpretabil-ityofsimplelogic-basedmodels.We rstcon-structarobustclassi ercombiningcontribu-tionsofalargenumberofgeneexpressionval-ues,andthensearchforcompactsummariza-tionsofsubgroupsamonggenesassociatedintheclassi erwithagivenclass.Thesubgroupsaredescribedbymeansofrelationallogicfea-turesextractedfrompubliclyavailablegeneannotations.Thecurseofdimensionalityper-tainingtothegeneexpressionbasedclassi- cationproblemduetothelargenumberofattributes(genes)isturnedintoanadvantageinthesecondarysubgroupdiscoverytask,asheretheoriginalattributesbecomelearningexamples.Introduction

Manytasksofautomatedknowledgediscoveryfromgeneexpressionmicroarraydatabydatamin-ingalgorithmsaimatconstructingclassi ersabletodiagnoseacancertypefromageneexpressionpro le.Seeeg.theseminalpapers[1,2]forreference.

Thistaskischaracterizedbytheabundanceofattributes(eg.simultaneouslymeasuredgeneexpres-sionvalues)ononehandconfrontedwiththeshort-ageoftheavailablesamples(eg.patients/tissuessub-jecttomeasurements)ontheotherhand.Itisknownfromscienti cdiscoveryliteraturethatsuchdomainsarepronetoover tting:over ttedclassi ersarechar-acterizedbysigni cantlydecreasedpredictiveaccu-racyonunseensamplescomparedtothetrainingsetaccuracy.

Informally,indomainscharacterizedbyasmallnumberofexamplesandalargenumberofat-tributes,over ttingoccursbecausesomeartefacts(“ ukes”)ofactuallyirrelevantattributecombina-tionscanemergesimplybymeansofchanceandappearsigni cantwithrespecttotheexamplesavail-abletothemachinelearningalgorithm.

Toavoidtheover ttingpitfall,state-of-artap-proachesconstructcomplexclassi ersthatcombinerelativelyweakcontributionsofuptothousandsofgenes(attributes)toclassifyadisease.Real-valuedsupportvectormachine(SVM)[3]modelsarecur-rentlyspeci callypopularinthegeneexpressiondataminingdomain.However,classi ersbasedonmanyreal-valuedattributeshaveanimportantdraw-back:theyarenotappropriateforexpertinterpreta-tion.Thecomplexityofsuchclassi erslimitstheirtransparencyandconsequentlythebiologicalinsighttheycanprovide.Althoughitispossibletoextracttheattributeswithmaximalcontributionweight,thelogicalconnectionsamongtheextractedattributesarethenlost.

Inourpreviouswork[4],wehavetestedthefeasibilityofconstructingsimpleyetrobustlogic-basedclassi ers,intheformofpropositionalrulesamenabletodirectexpertinterpretation,byanin-novativealgorithmicmethodologyofsubgroupdis-covery.Suchrulestypicallyincludetwoto vegeneexpressionattributesand,incontrasttomarkersobtainedfromSVMschemes,theserulesexplicitlystresstheimportanceofthecorrelationoftheactiv-ity(ornon-activity)ofanarrowgeneset.Followingourwork,furtherpapersappeared[5,6]proposingmethodsforconstructingverysimple,interpretable

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email