用于藥物關(guān)系學(xué)習(xí)的上下文感知分層融合 Context-Aware Hierarchical Fusion for Drug Relational Learning_第1頁
用于藥物關(guān)系學(xué)習(xí)的上下文感知分層融合 Context-Aware Hierarchical Fusion for Drug Relational Learning_第2頁
用于藥物關(guān)系學(xué)習(xí)的上下文感知分層融合 Context-Aware Hierarchical Fusion for Drug Relational Learning_第3頁
用于藥物關(guān)系學(xué)習(xí)的上下文感知分層融合 Context-Aware Hierarchical Fusion for Drug Relational Learning_第4頁
用于藥物關(guān)系學(xué)習(xí)的上下文感知分層融合 Context-Aware Hierarchical Fusion for Drug Relational Learning_第5頁
已閱讀5頁,還剩28頁未讀, 繼續(xù)免費(fèi)閱讀

下載本文檔

版權(quán)說明:本文檔由用戶提供并上傳,收益歸屬內(nèi)容提供方,若內(nèi)容存在侵權(quán),請進(jìn)行舉報(bào)或認(rèn)領(lǐng)

文檔簡介

Context-AwareHierarchicalFusion

forDrugRelationalLearning

YijingxiuLu,YinhuaPiao,SangseonLee,SunKimSeoulNationalUniversity

Outline

?Background

?Motivation

?Method

?Experiments

?Summary

Background

DrugRelationalLearning

oCo-administrationofdrugsisacommonpracticeintreatingdiseases.

oChemicalandphysicalreactionsbetweendrugscanaltertheintendedfunctionalityofdrugs.

oComplexbiochemicalmechanismswithinthehumanbodycouldfurtherleadtoadversedrugreactions.

oDiscoveringallpossibledrugcombinationsusing

traditionallaboratory-basedmethodsischallenging.

Synergeticeffect

ondestroyinga

specifictypeof

lungcancercells

Unwanted

chemicalUnexpected

interactionspolypharmacy

sideeffects

Background

DrugRelationalLearning

1.Drug-druginteractionsarecontext-dependent

oE.g.TheconcomitantintakeofTylenolandalcoholcanleadtoliver

damageduetocompetitionforthesamemetabolicenzyme.Tylenol(acetaminophen)Alcohol(Ethanol)

CYP2E1

compete

CYP2E1

NAPQI(toxic)

!

Glutathione

cysteineandmercapturic

acidconjugates

(nontoxic)

Acetaldehyde

insufficient

Unexpectedpolypharmacy

sideeffects

Background

DrugRelationalLearning

2.Drugrelationshipscanchangewithcontext

oE.g.Cabazitaxalandzoledronicacidexhibitsynergyinlungcancercelllinesbutactantagonisticallyinbreastcancertreatment.

Complexmechanismsaffectedbycontextchanges:

oTumormicroenvironments.

oAntagonisminbreastcancer.

oDrugtransportandmetabolism.

Synergeticeffect

ondestroyinga

specifictypeof

9lungcancercells

BackgroundCurrentWorks

Currentworksondrugrelationallearningcanbecategorizedintonetwork-basedandstructure-based.

oNetwork-basedmethods:

oIntegratemulti-omicsdatatoconstructheterogeneousnetworksforinferringDDI.

oStructure-basedmethods:

oDirectlylearnchemicalpropertiesandbiologicalactivitiesfrommolecularstructure.

Network-basedMethods

Structure-basedMethods

BackgroundCurrentWorks

Currentworksondrugrelationallearningcanbecategorizedintonetwork-basedandstructure-based.

oNetwork-basedmethods:

oIntegratemulti-omicsdatatoconstructheterogeneousnetworksforinferringDDI.

oStructure-basedmethods:

oDirectlylearnchemicalpropertiesandbiologicalactivitiesfrommolecularstructure.

Howtocombinetheadvantagesofbothandbuildamodelsuitablefornewdrugs?

Network-basedMethodsStructure-basedMethods

Method

HierarchicalInformationFusion

Context-awaredrug-drugrelationallearning:

oInformationfusionbetweendrugs.

oInformationfusionbetweendrug-context.

oDrugfeatureencoderlearnscontext-awarerelationknowledge.

oInferunknownrelationship.

drugi

contextc

drugj

Rdi→c

HiⅡRdj→di

?--------->

Lsup

HjⅡRdi→dj

Rdj→c

Hi

Hc

Hj

Method

ProblemDefinition

Context

oConsiderasetofannotateddrug-drug-contexttriplet

drugi

tuples(di,dj,c,y),wheredi,dj∈D,c∈C,andyisthetargetvariablebelongingtoY.

drugj

oD={d1,d2,...,dn}representacollectionofndrugs,andC={c1,c2,...,cm}denoteasetofmcontexts.

drugk

oHere,yisascalarvalue,rangingfromnegativetopositiveinfinityinregressiontasks,andtakingbinaryvalues(0or1)inclassificationtasks.

relation(e.g.whethertwodrugsi,jexhibitsynergyinaspecificcelllinec)

drugi

drugj

contextc

Method

Context-AwareHierarchicalFusion

1.DrugEncoderandContextEncoder

Weemploy:

oGraphIsomorphismNetwork(GIN)asgraphencoder.

?=MLP(??1+??1)

u∈N(v)

oMulti-LayerPerceptron(MLP)ascontextencoder.

?c=MLP(xc)

contextc

Hc

Method

2.Drug-DrugCrossFusion

Context-AwareHierarchicalFusion

oweemployanatom-wiseinteractionmaptocalculatethe

Hi,Hj

directionalrelationshipRdi→djbetweenapairofdrugsiandj.

Iij=sim

Rdi→dj=I·Hj

Hi∥Rdi→dj

oweupdatetherepresentationofdrugias:

H=concat

3.Drug-ContextCrossFusion

oSimilarly,wecomputetherelationshipsbetweendrugsandcontext:

Iic=simH,HcRdi→c=I·H

Rdi→c

Method

Context-AwareHierarchicalFusion

4.TripletRelationPredictor

oWefeedthefinalhiddenrepresentationofthedrug-drug-contexttripletintoMLPforrelationprediction:

hdi,dj,c=concat(HcⅡRdi→cⅡRdj→c)di,dj,c=MLP(hdi,dj,c)

drugi

c

context

drugj

Rdi→c

HiⅡRdj→di

__--------->

Rdj→c

HjⅡRdi→dj

Hi

Hc

Hj

Lsup

Outline

?Background

?Motivation

?Method

?Experiments

?Summary

Results

BenchmarkDatasets

weconsiderthethreemostpopulartasksindiseasetreatment:

oDrug-DrugSynergytask:

opredictswhetherapairofdrugsdi,djexhibitsynergyinaspecificcelllinec.

oDrug-DrugPolypharmacySideEffecttask:

opredictswhetherapairofdrugsdi,djleadstoaspecificadverseeventc.

oDrug-DrugInteractiontask:

opredictswhetherapairofdrugsdi,djleadstoaparticularreactionc.

Results

Performance

oOurmodelsconsistentlyoutperformthebaselinesacrossalltasks,underscoringtheeffectivenessofourarchitectureinlearningcomplexdrugrelationsacrossdiversetasks.

Results

AblationStudy

Oneofthemostnoteworthydistinctionsbetweenourmodelandotherbaselinesisthatourmodelexplicitlylearnsdrugrelationshierarchicallythroughthedrug-drug-contexttriplet.

Thereisasignificantdropwhenrelationsarenotexplicitlymodeled.

Withouthierarchy,themodel’sperformancedropsbyaround3.3%inAUROC.

suggestingthatthehierarchicalarchitectureeffectivelyfiltersoutfeaturesthatareirrelevanttomodelprediction.

Removingeithersideofthefusionresultsinadropinperformance.

Results

Performanceundercold-drugsetting

Toassessthegeneralizationabilityofourmodelinpredictingrelationshipsbetweenunknowndrugpairs,weadoptedacold-drugsettingbypartitioningasmallsubsetofdrugsfromtheoriginaldataset.

oOurmodeloutperformedothermodelsbyasignificantmarginonDrugBankDDI,andachievecomparableperformancetothebestbaselineonDrugComb.

oInsuchacontext-richenvironment,theabilityofmodelstolearncontextualinformationismorecriticalforperformance.

Summary

MainchallengesinDrugRelationalL

溫馨提示

  • 1. 本站所有資源如無特殊說明,都需要本地電腦安裝OFFICE2007和PDF閱讀器。圖紙軟件為CAD,CAXA,PROE,UG,SolidWorks等.壓縮文件請下載最新的WinRAR軟件解壓。
  • 2. 本站的文檔不包含任何第三方提供的附件圖紙等,如果需要附件,請聯(lián)系上傳者。文件的所有權(quán)益歸上傳用戶所有。
  • 3. 本站RAR壓縮包中若帶圖紙,網(wǎng)頁內(nèi)容里面會有圖紙預(yù)覽,若沒有圖紙預(yù)覽就沒有圖紙。
  • 4. 未經(jīng)權(quán)益所有人同意不得將文件中的內(nèi)容挪作商業(yè)或盈利用途。
  • 5. 人人文庫網(wǎng)僅提供信息存儲空間,僅對用戶上傳內(nèi)容的表現(xiàn)方式做保護(hù)處理,對用戶上傳分享的文檔內(nèi)容本身不做任何修改或編輯,并不能對任何下載內(nèi)容負(fù)責(zé)。
  • 6. 下載文件中如有侵權(quán)或不適當(dāng)內(nèi)容,請與我們聯(lián)系,我們立即糾正。
  • 7. 本站不保證下載資源的準(zhǔn)確性、安全性和完整性, 同時(shí)也不承擔(dān)用戶因使用這些下載資源對自己和他人造成任何形式的傷害或損失。

評論

0/150

提交評論