hibit2017 conference bookhibit2017.ii.metu.edu.tr/.../uploads/hibit2017_conference_book.pdf ·...
TRANSCRIPT
1
organizedbyInformaticsInstitute
MiddleEastTechnicalUniversity(METU)
atMETUNorthernCyprusCampus
99738Kalkanli,Guzelyurt/NorthernCyprus
June28–30,2017
Motivation
The International SymposiumonHealth Informatics andBioinformatics, (HIBIT),nowinitstenthyear(HIBIT2017),aimstobringtogetheracademics,researchersandpractitionerswhoworkinthesepopularandfulfillingareasandtocreatethemuch-needed synergy among medical, biological and information technologysectors. HIBIT is one of the few conferences emphasizing such synergy. HIBITprovidesaforumfordiscussion,explorationanddevelopmentofboththeoreticaland practical aspects of health informatics and bioinformatics and a chance tofollowcurrentresearchinthisareabynetworkingwithotherbioinformaticians.
2
HIBIT2017COMMITTEESHonoraryChairs
NeşeYalabık METUNazifeBaykal President,METUNorthernCyprusCampus
ConferenceCo-chairs
NurcanTunçbağ METUTuncaDoğan METU
OrganizationCommittee
AybarCanAcar METUÇağdaşSon METUNurcanTunçbağ METURengülAtalay METUTolgaCan METUTuncaDoğan METUYeşimAydınSon METU
ProgramCommittee
AbdullahKahraman UniversityofZurichAhmetSacan DrexelUniversityAhmetRaşitÖztürk UMassMedicalSchoolAlperKucukural UMassMedicalSchoolAnthonyGitter UniversityofWisconsin-MadisonArzucanOzgur BogaziciUniversityAttilaGursoy KoçUniversityAybarCanAcar METUBaharTaneri EasternMediterraneanUniversityBarışSüzek MuğlaSıtkıKoçmanUniversityBilgeKaracali İYTEBurakErman KoçUniversityBurcuBakir-Gungor AbdullahGulUniversityCerenSucularli HacettepeUniversityCagdasDevrimSon METUCesimErten AntalyaInternationalUniversity
3
ElifErsonBensan METUEmreGüney UniversityofPompeuFabraErcümentÇiçek BilkentUniversityErdemKarabulut HacettepeUniversityErnestFraenkel MassachusettsInstituteofTechnologyGökhanKarakülah IBGHasanOğul BaşkentUniversityHilalKazan AntalyaInternationalUniversityJensAllmer İYTEMaria-JesusMartin EuropeanBioinformaticsInstitute(EMBL-EBI)MehmetÖzturk IBGMehmetSomel METUMesutMuyan METUNurcanTunçbağ METUOgünAdebali TheUniversityofNorthCarolinaatChapelHillOmerSinanSarac IstanbulTechnicalUniversityÖzgürGümüş EgeUniversityÖzgürŞahin BilkentUniversityÖzlemKeskin KoçUniversityÖzlenKonu BilkentUniversityÖznurTaştan BilkentUniversityRabieSaidi EuropeanBioinformaticsInstitute(EMBL-EBI)RalfHofestädt BielefeldUniversityRengülAtalay METUSukruTuzmen EasternMediterraneanUniversityTolgaCan METUTuğbaSüzek MuğlaSıtkıKoçmanUniversityTuncaDoğan METUUfukNalbantoğlu ErciyesUniversityUğurSezerman AcıbademUniversityVildaPurutcuoglu METUVolkanAtalay METUYeşimAydınSon METUZerrinIşık DokuzEylülUniversity
4
HIBIT2017PROGRAM*
MainConference–Day1(June28,2017)
09:00–10:00
OpeningSession
§ WelcomebyAssoc.Prof.Dr.MohammadShikakhwa(AdvisortothePresident,METUNCC)
§ WelcomebyProf.Dr.NeseYalabik(METU,HIBIT2017HonoraryChair)
10:00–11:00
Keynote–Dr.ErnestFraenkel,MassachusettsInstituteofTechnology
Title:BeyondGenomics:OpportunitiesandChallengesforUsingtheOtherOmicsin
PrecisionMedicine
11:00–11:15 Coffeebreak&CancerGenomicsCloud(CGC)demobySevenBridgesGenomics
11:15–11:45Dr.EmreGuney,UniversityofPompeuFabra(InvitedTalk)
Title:Challengesandopportunitiesinsystemspharmacology
11:45–12:15Dr.ErcumentCicek,BilkentUniversity(Invited)
Title:GenomicDataSharingandPrivacyRisks
12:15–13:30 Lunch
13:30–14:00Dr.MehmetOzturk,InternationalBiomedicineandGenomeInstitute(Invited)
Title:CancerGenomeAnalysesandPersonalizedMedicine
14:00–14:30Dr.BedirhanUstun,KocUniversity(Invited)
Title:SemanticInteropretability:Howtoavoidane-TowerofBabel?
14:30–14:45 Coffeebreak&CancerGenomicsCloud(CGC)demobySevenBridgesGenomics
5
14:45–15:30
SessionIStructuralBioinformatics(Amfi3)Chair:Dr.ÇağdaşSon
SessionIIHealthInformatics(Amfi2)Chair:Dr.YesimAydinSon
EzgiKaraca,JoãoRodrigues,Andrea
Graziadei,AlexandreBonvinandTeresa
Carlomagno.
Title:AnIntegrativeFrameworkfor
StructureDeterminationofMolecular
Machines
KorayAçıcı,ÇağatayBerkeErdaş,Tunç
Aşuroğlu,MünireKılınçToprak,Hamit
ErdemandHasanOğul.
Title:WearableSolutionsforParkinson’s
DiseaseMonitoring
SılaÖzdemir,AttilaGursoyandOzlem
Keskin.
Title:Analysisofsingleaminoacid
variationsinhotspotandhotregion
residuesofprotein-proteininteraction
interfaces
İlknurBuçanKırkbir,BurçinKurtand
KemalTurhan.
Title:AComputerAidedDiagnosisSystem
forHeartAttackUsingDecisionTree
SerenaMuratcıoğlu,HyunbumJang,Attila
Gursoy,OzlemKeskinandRuthNussinov.
Title:Interactionoffarnesylated,butlikely
depalmitoylated,RasisoformswithPDEδ
EmrahAkkoyun,AybarCanAcar,Byron
ZambranoandSeungikBaek.
Title:CardiovascularModellingfor
AbdominalAorticAneurysms
6
15:30–16:15
SessionIIICancerSystemsBiology(Amfi3)Chair:Dr.MesutMuyan
SessionIVGenomics(Amfi2)Chair:Dr.OgünAdebali
OguzhanBegik,MerveOyken,TunaCinkilli
Alican,TolgaCanandAyseElifErsonBensan.
Title:Alternativepolyadenylationpatternsfor
cancerclassification
ArifYılmazandYeşimAydınSon.
Title:LosslessPruningofAHPSNP
PrioritizationTreeUsingRandom
ForestVariableImportances
TundeAderinwaleandHilalKazan.
Title:IntegratingMultipleDataTypesfor
CancerSubtypeDiscovery
FatihKaraoglanoglu,MarziehEslami
RasekhandCanAlkan.
Title:DiscoveringLargeGenomicInver-
sionsUsingLong-rangeInformation
MonaShojaei,EceAkhan,AybarCanAcarand
RengülÇetinAtalay.
Title:IdentificationofGeneMutations
InvolvedInDrugResistanceInLiverCancer
UsingRNA-SeqDataAnalysis
ElifBozlak,YetkinAlıcı,EvrimFer,
MelikeDönertaş,RasmusNielsen
andMehmetSomel.
Title:Neighbouringsequencecontext-
basedfixationbiasofmutationsin
chimpanzeegenome
16:15–16:30 CoffeeBreak&CancerGenomicsCloud(CGC)demobySevenBridgesGenomics
16:30–18:30 POSTERSESSION
19:00–20:30 WelcomeReception
7
MainConference–Day2(June29,2017)
09:00–09:30 Tea/coffee,cookies
09:30–10:30Keynote–Dr.MariaMartin,EuropeanBioinformaticsInstitute(EMBL-EBI)
Title:Bigdatainfrastructureforadvancingbiomedicalresearch
10:30–10:45 Coffeebreak&CancerGenomicsCloud(CGC)demobySevenBridgesGenomics
10:45–11:15Dr.AybarCanAcar,METU(Invited)
Title:ProbabilisticProgramminginBioinformatics
11:15–11:45Dr.OgunAdebali,UniversityofNorthCarolinaatChapelHill(InvitedTalk)
Title:GenomicsofDNADamage,RepairandMutagenesis
11:45–12:15
SessionVMulti-omics(Amfi3)Chair:Dr.EmreGüney
SessionVIComputationalDrugDiscovery(Amfi2)Chair:Dr.TolgaCan
CesimErten,EvisHoxha,HilalKazanandEsra
Tepe.
Title:Identificationofdysregulatedpathways
acrossmultiplecancertypes
RemziÇelebi,ErkanYaşar,Özgür
GümüşandOğuzDikenelli.
Title:RDFGraphEmbeddingsfor
PredictionofDrug-DrugInteractions
GuvanchOvezmyradov.
Title:Integrativebioinformaticsanalysisof
multi-omicsdatafacilitatesfunctional
characterizationofthecandiatetumor
suppressorproteinCTCF
HandanMelikeDönertaşandJanet
Thornton.
Title:DrugRepurposingforAgeing:A
ConnectivityMapApproach
12:15–13:30 Lunch
13:30–14:00Dr.TugbaSuzek,MuglaSitkiKocmanUniversity(InvitedTalk)
Title:Bridgingcheminformaticsandbioinformaticswithpublictools
8
14:00–14:30
SessionVIIProteinFunctionPrediction(Amfi3)Chair:Dr.BarisSuzek
SessionVIIIInteractomics&miRNA(Amfi2)Chair:Dr.AybarCanAcar
RabieSaidi,MaryamAbdollahyanand
MariaMartin.
Title:ASelf-trainingApproachforFunctional
AnnotationofUniProtKBProteins
ErdemTurkandBarisSuzek.
Title:LeveragingTaxonomicRanksfor
ImprovingMirrorTree-basedProtein
InteractionPrediction
AhmetSüreyyaRifaioğlu,MariaMartin,
RengulCetin-Atalay,VolkanAtalayand
TuncaDogan.
Title:InvestigationofMulti-taskDeepNeural
NetworksinAutomatedProteinFunction
Prediction
SedefErkunt,NecatiAltindis,
AhmetEfeKoseogluandCemalUn.
Title:UseofmiRNAinstrugglingof
Varroaparasite
14:30–14:45 Coffeebreak&CancerGenomicsCloud(CGC)demobySevenBridgesGenomics
14:45–17:00
SessionIX–DataSharing
(Amfi3)
Chair:Dr.RengulCetin-Atalay
SevenBridgesGenomics
Title:CancerGenomicsCloudandCaseStudies
Dr.RichardDixon,QIAGEN
Title:HumangenomesequenceanalysisandinterpretationwithQIAGENBioinformatics
Dr.ElifErsonBensan,METU
Title:Resolvingthecomplexityofthecancertranscriptomethrough3’UTRs
Dr.NurhanOzlu,KocUniversity
Title:ProteomicAnalysisofEpithelialMesenchymalTransition
Dr.HilalOzdag,AnkaraUniversity
9
Title:SporadicColorectalCancerinTurkey:FromGWAStoValidation
Dr.OzgurSahin,BilkentUniversity
Title:SystemsbiologyapproachestoconstructRNAinteractionnetworksregulating
chemotherapyresistanceandmetastasisintriplenegativebreastcancer
Dr.MesutMuyan,METU
Title:Proteininteractionapproachestoassignafunctionforanestrogenresponsive
geneprotein:CXXC5
17:00–17:15 Closingremarks
ISCBRSGTurkeyStudentSymposium–Day3(June30,2017)
09:00–09:15 OpeningSession
09:15–10:00Keynote–Dr.UğurSezerman,AcıbademUniversity
Title:WANTED:Bioinformaticians
10:00–10:20 CoffeeBreak
10:20–11:20 StudentSession–ScienceSlam
11:20–11:50Dr.EmreGüney,UniversityofPompeuFabra(InvitedTalk)
Title:ShouldIstayorshouldIgoNOW?
11:50–12:10 Closingremarks
*Subjecttochanges
10
KEYNOTES
Dr.ErnestFraenkel
MassachusettsInstituteofTechnology
http://fraenkel.mit.edu/
Prof. Fraenkel received his A.B. in Chemistry and Physics from HarvardCollegeandhisPh.D.inBiologyatthelaboratoryofProfessorCarlPabofromatMIT.Hecontinuedhispost-doctoralresearchasafellowatthelaboratoryof Professor Stephen Harrison at Harvard University. He was a Whitehead
FellowandaPfizerComputationalBiologyFellowattheWhiteheadInstitute.Prof.FraenkeljoinedMITasaResearchAffiliateattheMITComputerScienceandArtificialIntelligenceLaboratory.HebecameanAssistant Professor at the Department of Biological Engineering in 2006. The Fraenkel laboratory isdeveloping computational and experimental approaches to search for new therapeutic strategies fordiseases.Newexperimentalmethodsmakeitpossibletomeasurecellularchangesacrossthegenomeand proteome. These technologies include genome-wide measurements of transcription, of protein-DNA interactions (ChIP-Seq), of genetic interactions, and of protein modifications. Each data sourceprovidesaverynarrowviewofthecellularchanges.However,bycomputationallyintegratingthesedatathe group can reconstruct signaling pathways and identify previously unrecognized regulatorymechanismsthatcontributetotheetiologyofdiseaseandmayprovidenewapproachesfortreatment.
Dr.MariaMartin
EuropeanBioinformaticsInstitute(EMBL-EBI)
https://www.ebi.ac.uk/about/people/maria-jesus-martin
Maria’s team provides the bioinformatics infrastructure for the databasesandservicesoftheUniversalProteinResource(UniProt).Herteamcomprisessoftware engineers and bioinformaticians who are responsible for theUniProt,theGeneOntologyAnnotationandtheEnzymeportalsoftwareand
database development, and who study novel automatic methods for protein annotation andrepresentation. The team’s user experience analyst coordinates the user request gathering process,which informs the design and development of the web site. Maria’s team is responsible for themaintenance and development of tools for UniProt curation, and works in a fully complementaryfashionwithClaireO’Donovan’sUniProtContentteamtoprovideessentialresourcesforthebiologicalcommunity,asthedatabaseshavebecomeanintegralpartofthetoolsresearchersuseonadailybasisfortheirwork.
11
KEYNOTETALKS
BeyondGenomics:OpportunitiesandChallengesforUsingtheOtherOmicsinPrecisionMedicine
ErnestFraenkelMassachusettsInstituteofTechnology
Rapidadvancesinhigh-throughputtechnologies,includingnext-generationsequencing,proteomics,andmetabolomics,areprovidingexceptionallydetaileddescriptionsofthemolecularchangesthatoccurindiseases.However,ithasbeendifficulttousethesedatatodiscovernewtherapeuticinsights.Despitetheir power, each of these methods still only captures a small fraction of the cellular response.Moreover,whendifferentassaysareappliedtothesameproblem,theyprovideapparentlyconflictinganswers.Iwillshowhowspecificnetworkmodelingapproachesrevealtheunderlyingconsistencyofthedata by identifying small, functionally coherent pathways linking the disparate observations. Thesepatient-specificnetworksmayprovidecriticalinsightsfortargetedtherapies.Currently,weareapplyingtheseapproachestoarangeofdiseases,includingbraintumorsandneurodegenerativedisorders.Iwilloutlinehowweareusingthesemethods inthe largeststudyeverstudyofALS,andwilldiscusssomeopenproblems.
Bigdatainfrastructureforadvancingbiomedicalresearch
MariaJesusMartinEuropeanMolecularBiologyLaboratory,EuropeanBioinformaticsInstitute(EMBL-EBI)
Over the last decade, life science research has become a data driven scientific field. Large-scalegenomics and proteomics research generate an enormous amount of information, which is mostlystored in knowledgebases, and accessed and analyzed using bioinformatics tools. Bioinformatics andsupporting databases hold significant interest in the scientific community because of its potential tomovescientific research forwardmorequicklyandat lessexpense than traditional laboratory testing.The European Bioinformatics Institute (EMBL-EBI) (http://www.ebi.ac.uk/) hosts many internationaldatabases and bioinformatics tools in major research areas, allowing data to be shared and freelyavailabletoall.Iwilldiscussthevitalroleofpublicdatabasesinlife-scienceresearchanddemonstratetheirusefromlabdatagenerationtosubmission,addedvalueanddissemination.Oneofworldleadingknowledgebases forproteinresearch isUniProt (http://www.uniprot.org/)whichprovideshighqualityand comprehensive protein information including annotation of variants with functional effects anddisease associations derived fromboth the scientific literature and large experimental data sets. Thisprovidesbiomedical researcherswithaplatformto investigateandvisualize functional informationofproteinsalongwithgenomicalterations thatcancontribute tomodifications in the translatedproteinandtheirsignificanceinadiseaseorsyndrome.Bioinformaticsintegrativeapproachesareessentialfortoday’shealthanddiseaseresearch-databasesareatthecoreprovidingdataandtoolsforbuildingnewplatformse.g.forclinicalanddrugdiscovery.
12
INVITEDTALKS
Challengesandopportunitiesinsystemspharmacology
EmreGuneyUniversityofPompeuFabra
Thepastfewdecadeshavewitnessedaburstinthehigh-throughputmoleculardatasets,sheddinglightonthemechanismsunderlyinghumangeneticdiseases.Despitetherecenteffortsonsequencingandgenomewideassociationstudiescharacterizinggeneticvariantsassociatedwithpatientsubgroups,mostclinicaltrialsfailduetothelackofefficacyandsafety.Accordingly,repurposingexistingdrugsfornewuseshasattractedconsiderableattentionoverthepastyears.Toidentifypotentialcandidatesthatcouldberepositionedforanewindication,manystudiesmakeuseofchemical,target,andsideeffectsimilaritybetweendrugstotrainclassifiers.Despitepromisingprediction accuraciesof these supervised computationalmodels, their use inpractice,suchasforrarediseases, ishinderedbytheassumptionthattherearealreadyknownandsimilardrugs for a given condition of interest. In this talk, using publicly available data sets and data-drivenmodeling, I will challenge such assumptions and explain my recent efforts on interactome-basedcharacterizationofdrugeffect.
GenomicDataSharingandPrivacyRisks
ErcumentCicekBilkentUniversity
Genomic datasets are often associatedwith sensitive phenotypes and leak ofmembership information ismajorprivacyconcern.GenomicBeacons,whichwereproposedbyGlobalAllianceforGenomicsandHealthcoalition, aim to provide a secure and standardized interface for data sharing. Beacons allow only yes/noquestions on the presence of specific alleles in the dataset. However, recent studies have shown thatinformationleakispossiblebyrepeatedlyqueryingthebeaconforcertainSNPsandtheyalsoproposedsomecountermeasures.Inthistalk,Iwillpresentanovelre-identificationattackandshowthattheprivacyriskismore serious thanpreviously thought. In thismodel, even if the attackerdoesnothave access to the fullgenome of the victim (e.g., regions covering SNPswith lowminor allele frequencies are hidden), it is stillpossible to infer these alleles and infer beacon results with high confidence. We use the linkagedisequilibriumandahigh-orderMarkovchainbasedalgorithmfortheinference.Weshowthatinabeaconwith65 individuals fromtheCEUpopulation,wecan infermembershipof individualswith95%confidencewith only 2 queries, even though SNPs with MAFs less than 0.05 are hidden. We also show thatcountermeasures such as setting a query budget for the user would fail to protect the privacy of theparticipants.
CancerGenomeAnalysesandPersonalizedMedicine
MehmetÖzturkInternationalBiomedicineandGenomeInstitute
13
SemanticInteropretability:Howtoavoidane-TowerofBabel?
BedirhanÜstünKocUniversity
Today many health systems utilize computers to capture and analyze health information digital format,however, exchange of information among different agents is very difficult because of use of differentterminologies,technicalformatsandlackofrelevantstandards:thissituationhasresultedinane-TowerofBabel.
Health Information emanates from different scientific activities using a large set of terms, definitions andotherformulationsrangingfromanatomytogenetics…;frommentalfunctionstosocialimpacts…andsoon.Therepresentationofthisbroadknowledgebaseforuseincomputerizedhealthinformationsystemsposeasetofchallengessuchas:thevastamountofconcepts;theirdifferentinterpretations;andvariationsinusebydifferentpeople includingprovidersandconsumers.Hence, the“standardization”ofhealth informationanditsterminologicalrepresentation.
Althoughstandardizationmaycreatea“commonlanguage”,itsemulationincomputersystemsrevealmanyshort-comings.“Meaningofmeaning”wasconceptualizedbyOgdenandRichardsin1923asthe“semantictriangle”dissectingtheknowledgerepresentationamongthethings,thoughtsandterms.Today,weneedtotransform the existing complex health knowledge from its analog format to digital format for use incomputerized health information systems. To achieve this aim, it is necessary to systematically create anexplicit and operational description of the concept to represent the knowledge. In computer science“ontologies”areusedforthispurposetorepresenteachconcept.Thetermontologyhereisuseddifferentthanitsuseinphilosophy,simplyasaknowledgerepresentationscience.
Oncethere isanagreementonthe“commonontology”thenthecomputersystemscantalktoeachotherusing this common framework and exchange information. Similarly, the same framework could be usedamongpeopleaswell toexchange information.This iscalled interoperabilityandhastwocomponents: (a)technical interoperability: one agent can send and receive messages from the other; (b) semanticinteroperability:themeaningofexchangedinformationiscommon;i.e.,understoodassamebetweentwoagents.
Nevertheless,conversionof“analog”healthrecordsinto“digital”willnotsufficetosolvetheproblem;andwilllikelyendupintheclassicalcomputeradage:“Garbage-in:Garbage-out”.Thereisaneedtomakeuseofthis information in an intelligent manner through computational processing. If we create appropriateaggregations, thenwecanprocessdata toanswerquestions like: (a)HowmanypatientsdohavediabetesmellitustypeIIinCyprus?(b)howmanyhavenormalHbA1cresults?(c)howmanyneedtreatmentneedfordiabetesmellitus?(d)HowdoesCyprusregionAcompareagainstregionB;orCyprusagainstrestofEuropeortheWorld?
HealthInformaticsofferagreatopportunitytomakeon-lineepidemiology:“e-pidemiology”oron-linehealthservicesresearch. Thisusagemustberegulatedbeyondandabovetechnicalusageregarding itsutility forpublichealth.Mostimportantlyappropriateuseofthisdatarespectingprivacyofpeopleisessential.Ontheotherhand, there isnext tononeregulatory framework tomandateanyorganizedactivity tomakeuseofthisdataforpublichealth.
14
ProbabilisticProgramminginBioinformatics
AybarCanAcarMETU
GenomicsofDNADamage,RepairandMutagenesis
OgünAdebaliUniversityofNorthCarolinaatChapelHill
OurDNAiscontinuallydamagedbycellularandenvironmentalfactors.Someofthemostcommonmutagens,including UV light and platinum-based drugs, cause bulky adduct formation; these are almost exclusivelycorrected through nucleotide excision repair. Although the basic mechanisms of excision repair arereasonablywellunderstood,wepreviouslydidnotknowwhichgenomiclociarepreferentiallydamagedandrepaired.Usingthenoveltechnologies,Damage-seqandXR-seq,wewereabletodecipherthegenome-widekineticsand locationofdamage formationandexcisionrepair followingultraviolet irradiationandcisplatintreatments. These single-nucleotide resolution maps revealed cellular components affecting the DNAdamageformationandrepair.Thedamagesthatescapetherepairmachinerycanresult inmutagenesis. Inclinics, there are currently no accurate ways to determine the consequences of a novel mutation. Weproposeanewcomputationalperspectivetopredictingpathogenicityofsinglepointmutationsbybuildingthepreciseevolutionaryhistoriesof genes.Taken together,weapplygenomic technologies tounderstandhowDNAisdamagedandrepaired,aswellaswhetherunrepairedmutagenicdamagesleadtopathogenicityornot.
Bridgingcheminformaticsandbioinformaticswithpublictools
TuğbaSüzekMuglaSitkiKocmanUniversity
Typicaldrugdevelopmentpipelinetakesonaverage15yearswithextremelyhighmonetarycostswithverylow number of drugs successfully coming out of the process. Although not very well known bybioinformaticians, in-silico compound bioactivity mining tools offer a cost-effective alternative to leadcandidateidentification.Inthistalk,Iwillpresentanoverviewofonlinetoolsforin-silicodrugminingandanexampleonhowpubliccheminformatics resourcescanbeutilizedtobridgebetweencheminformaticsandbioinformaticsdataspace.
15
INVITEDTALKS(DATASESSION)
CancerGenomicsCloudandCaseStudies
YağızCanŞişman
SevenBridgesGenomics
TheCancerGenomicsCloud (CGC),poweredbySevenBridges, isoneof threepilotsystems fundedbytheNationalCancerInstitutetoexploretheparadigmofcolocalizingmassivegenomicsdatasets,likeTheCancerGenomics Atlas (TCGA), alongside secure and scalable computational resources to analyze them. The CGCmakesmorethanapetabyteofmulti-dimensionaldataavailableimmediatelytoauthorizedresearchers.YoucanaddyourowndatatoanalyzealongsideTCGAusingpredefinedanalyticalworkflowsoryourowntools.Everyexecutionisfullyreproducible,andcollaboratingwithyourteamissimpleandsecure.
HumangenomesequenceanalysisandinterpretationwithQIAGENBioinformatics
RichardDixonQIAGEN
Resolvingthecomplexityofthecancertranscriptomethrough3’UTRs
ElifErsonBensanMETU
Advancements insequencingandtranscriptomeanalysismethodscontributedtoabetterunderstandingofthe complexity of cancer. These findings are paving the way toward the development of improveddiagnostics, prognostic predictions, and targeted treatment options. In an effort to have a morecomprehensiveunderstandingofcancer,wefocusonthe3’UTRs(untranslatedregions)ofgenesaswecametoappreciatethevalueofthenoncodingregionsofourgenomes,partlyduetomicroRNAs.The3’UTRshavelongbeenknowntohaveimportantrolesinmaintainingthestability,localization,andhalf-lifeofmRNAsbuta detailed mechanistic explanation as to how these properties are regulated or the consequences ofderegulation are just beginning to be appreciated. Our group is interested in the regulation andconsequencesof3’UTRlengthchangesinmRNAisoformsinbreastcancers.3’UTRlengthiscontrolledmainlyby the position of the polyadenylation (poly(A)) signal. Interestingly, majority of human genes harbormultiplepoly(A) signals on their 3’UTRs that canbedifferentially selectedon thebasis of thephysiologicstate of cells, resulting in alternative mRNA isoforms. Hence, deregulation of alternative polyadenylation(APA) has increasing interest in cancer research, because APA generates mRNA isoforms with potentiallydifferentprotein functions. Inthis talk,ourgroupsefforts to identifyandcharacterizenovelcancerrelatedgeneswill be discussed based on our combinatorial approach to reveal APA events. Overall, detection ofderegulated APA-generated isoforms in cancer may implicate some proto-oncogene activation cases of
16
unknowncausesandmayhelpthediscoveryofnovelcases;thus,contributingtoabetterunderstandingofmolecularmechanismsofcancer.
ProteomicAnalysisofEpithelialMesenchymalTransition
NurhanÖzlüKocUniversity
Epithelial-Mesenchymal-Transition(EMT)playsanimportantroleduringcarcinogenesisandtumorformation.Itsmajorcontributiontothetumorformationisthroughprovidingepithelialcellswiththeabilityof invasionandmetastasis. By thisway, epithelial cells gain the capacity to disseminate from primary tumors to allowthemtogrowatadistantlocation.Cancerprogressionthroughtheprocessofmetastasishasbeenthefocusofextensiveresearchforyears.However,themolecularplayersoftheEMTprocessarenotfullyunderstood.Inthisregard,examiningtheradicalbiochemicalchangesatbothproteinandpost-translationmodificationlevelsduring EMT are critically important to identify regulatory factors effecting EMT andmetastatic behavior ofcarcinomas.Inthisstudy,wetakeadvantageofthecomparativeproteomicsandphosphoproteomicsmethodsthat we developed in our previous studies to comprehensively evaluate the biochemistry of a cell and itsphosphorylationeventsasitundergoesEMT.
SporadicColorectalCancerinTurkey:FromGWAStoValidation
HilalÖzdağAnkaraUniversity
ColorectalcancersshowthethirdhighestmortalityrateinWesterncountriesandTurkey.Overthepast10years, in most of the OECD countries mortality from colorectal cancer (CRC) has significantly decreased.However,CRCmortalityinTurkeyhasunfortunatelyincreased.Studiesshowedthat85-90%ofCRCcasesaresporadicwhile10-15%ofcasesarefamilialorhereditary.Todate,mostofthegenescausinghereditaryandfamilial CRC have been identified. Meanwhile, many studies have recently been conducted in order toidentify genes responsible of predisposition to sporadic CRC bymeans of genomewide association studies(GWAS).
OurgrouphasconductedthefirstfamilybasedGWASinTurkishpopulation.Withintheframeofthisstudy,primarily anovel approachpreviouslydevelopedbyour groupadapting SPRT (Sequential ProbabilityRatioTest) analysis to family based association analysis was used. SPRT contrary to TDT (TransmissionDisequilibriumTest)which requiresat least200 trios inorder toconducta familybasedassociationstudy,hastheabilitytoconductareliableassociationanalysisusinglessthan200trios.Thus,250KSNParraydataof 51 trios (sporadic colorectal cancer cases and their healthy parents)whichwere included in this familybasedGWASwere analyzed by SPRT. The analysis revealed that 75 SNPs showed genetic associationwithCRC.
Thevalidationofthis75candidateSNPswasconductedbykompetitiveallelespesificamplification(KASP)in1019sporadiccolorectalcancercasesand948healthy individualswhichwerecollected from45centres in
17
Turkey.Besides these75SNPs,7SNPsassociatedwithCRC(p<1.5x10-10) inEuropeanpopulationswasalsoinvestigatedinourstudygroup.Statisticalanalysisshowedthat5SNPswhicharelocalizedin1p32.2,1q43,4p15.2,5q11.2,3q25.32areassociatedwithCRC(p<0.05).Furthermoreamong7EuropeanCRCassociatedSNPs we have found that only one, namely rs6983267, was associated with CRC (p<0.0018) in Turkishpopulation.
To the best of our knowledge this study is the first and only GWAS and its validation results in Turkishpopulation. This study has clearly showed that CRC genetic predisposition profile of Turkish population isdifferentfromEuropeanpopulations.
SystemsbiologyapproachestoconstructRNAinteractionnetworksregulatingchemotherapyresistanceandmetastasisintriplenegativebreastcancer
ÖzgürŞahinBilkentUniversity
Triplenegativebreastcancer(TNBC)isthemostaggressivetypeofbreastcancerwithhighincidencerateoflungmetastasis,andcurrenttherapiesarelimitedtochemotherapy.Notonlyproteincodingtranscripts,butalso non-coding transcriptome, such as microRNAs (miRNAs) and long non-coding RNAs (lncRNAs), playcritical roles in cancer progression, therapy resistance andmetastasis. To this end,we aimed to integratetranscriptomics and functional genomics to discover RNA interaction networks modulating chemotherapyresistanceandmetastasisinTNBCs.
To identify chemotherapy resistance-associatedmiRNA-mRNA regulatorynetwork in TNBCs,wedevelopeddoxorubicin resistance animalmodels in vivo and performed next generation sequencing (both RNA- andsmallRNA-seq)comparingdoxorubicinsensitiveandresistanttumors.UsingmiRNAtargetfilterandpathwayenrichmentanalysesandnetworkbiologyapproaches,weincorporateddifferentiallyexpressedmiRNAsandmRNAs into an interaction network. Later, miRNA and mRNA expression based survival analysis ofchemotherapy-treatedTNBCpatientshelped to strengthen the clinical relevanceof theproposednetworkandto findthecoremechanisms involved indoxorubicinresistance inTNBCs.Wearecurrently testingtheroleofcriticalnodestoovercomechemoresistance.
InordertoconstructthefirstmRNA-miRNA-lncRNAcompetingendogenousRNA(ceRNA)networkcontrolingmetastasis in TNBCs, we established primary tumors, and human-in-mouse (HIM) and mouse-in-mouse(MIM)lungmetastasismodelsusingTNBCcelllines.WeperformedbothRNAandsmallRNAsequencingandobtaineddifferentiallyexpressedmiRNAs,mRNAsandlncRNAsbetweenprimaryandmetastatictumors.Wethenintegratedthesethreelayersofdatawithfunctionalenrichments,pathwaymapsandtargetpredictionsto construct the ceRNA network controlling lung colonization in TNBCs. Currently, we are testing thefunctionalrolesofcandidatelncRNAsininvitroandinvivometastasisassays.
Ultimately, these studies will uncover novel non-coding RNAs and proteins that can be used as potentialtargetstoovercomeresistanceorblockmetastasis.
18
Proteininteractionapproachestoassignafunctionforanestrogenresponsivegeneprotein:CXXC5
MesutMuyan
METU
Estradiol(E2),themainestrogenhormoneincirculation,is involvedinthephysiologicalregulationofmanyorgansandtissues,includingmammarytissue.E2alsoplaysacriticalroleintheinitiationanddevelopmentofbreastcancer.EffectsofE2oncellsaremediatedbyestrogenreceptor(ER),whichisaligand-dependenttranscription factor. Upon binding to E2, ER regulates the expression of target genes through genomicsignalingpathwaysthatresultinalterationsofcellularresponses.TheinteractionoftheE2-ERcomplexwithspecific DNA sequences, estrogen response elements (EREs), constitutes the initial stage of the ERE-dependentgenomic signalingpathway.The functional interactionofE2-ERwithother transcription factorsbound to their cognate response elements signifies the ERE-independent genomic signaling route. Theexpression of estrogen responsive genes mediated by E2-ER encompasses proteins involved in themetabolism of nucleic acid/proteins, transcription factors, membrane signaling cascade and receptorproteins.Theseproteins inturnparticipate intheregulationofsecondarygeneexpressionsresponsibleforDNArepair,cellcycleanddivisionand,consequently,intheinitiationofE2-mediatedcellularproliferation.
TheidentificationofE2responsivegenes,characterizationofmechanismofgeneexpressionsanddiscerningfunctions of protein products could lead to important gains towards understanding of the physiology andpathophysiologyofestrogensignaling.Ourpreviousstudies indicatedthatCXXC5isanestrogenresponsivegene. Although, the structure and function of CXXC5 are largely unknown, due to the presence of a zinc-fingerdomain-CXXCdomain (ZF-CXXC),CXXC5 ispresumed tobeamemberof theZF-CXXCprotein family.ThebindingofZF-CXXCproteinstoCpGdinucleotidesissuggestedtopreventcytosinemethylationleadingtotheformationofanucleationsiteforthedirectorindirectrecruitmentofhistonemodifyingproteinstoDNAfor transcriptionregulation.WethereforepredictedthatCXXC5,synthesizedas theprimaryresponsegeneproduct,participatesasanepigeneticfactorintheregulationofE2-mediatedcellularprocesses.
OurrecentstudiesfurtherindicatedthattheexpressionoftheestrogenresponsiveCXXC5geneismediatedbytheE2boundERthroughadirectinteractionwithanEREsequencepresentattheupstreamregionofthefirstencodingmethionine in theCXXC5gene locusthatresults inalterations inCXXC5protein levels in thenucleus of a cell model. Moreover, we found that CXXC5 is indeed a CpG dinucleotide binding protein.Furthermore, our studies using proximity biotinylation assay, BioID, coupled to LC-MS/MS revealed thatCXXC5putativelyinteractswithanumberofproteinsinvolvedinthechromatinremodeling,DNArepairandcellcycle.
Ourongoingstudiesaimedattheverificationofprotein-partnerinteractionswiththeuseofvariousinvitroand in cellula approaches would be supportive for our prediction that CXXC5 as an epigenetic factor isinvolvedinE2-mediatedcellularevents.
19
SELECTEDTALKS
SESSIONI:StructuralBioinformatics
AnIntegrativeFrameworkforStructureDeterminationofMolecularMachines
EzgiKaraca*,JoãoRodrigues,AndreaGraziadei,AlexandreBonvinandTeresaCarlomagno
*EMBL&iBG-izmir{[email protected]}
Classical structural studiesofmolecularmachinesarehamperedby technical limitationsofhigh-resolutionmethods (NMR spectroscopy and X-ray crystallography), due to the large size of these systems and theirhighly flexible nature. Luckily, even for such challenging systems, we can access low-resolution structuralinformation from complementary biochemical and biophysical experiments. Most of the time, however,these data are rather sparse and contain limited information. In order to translate such data into viablestructuralmodels,integrativecomputationaltoolsarerequired.
During the past decade, a handful of integrativemethods have been proposed. Though, asmost of themweredeveloped inacase-specificmanner,there isstillanongoingdemandforastandardizedtool.Tothisend,weproposeageneral integrative structuredetermination framework that candealwithmoleculesofdifferentnature(e.g.proteins,nucleicacids)andmakeuseofdiversedistanceand/orproximityinformationtodeterminethestructuresofmolecularmachines.
Inmytalk, Iwill first introducetheconceptof integrativemodelingandour integrativeframework.Then, Iwillillustratetheperformanceofourapproachwithtwoproof-of-principleexamples.Finally,Iwillunderpinits strengthwith a real case application, theBoxC/Dmolecularmachine,whereweuse a combinationofNMR distance information and complementary SAXS/SANS shape data to elucidate how this machineregulatesribosomalRNAmethylation.
20
Analysisofsingleaminoacidvariationsinhotspotandhotregionresiduesofprotein-proteininteractioninterfaces
SılaÖzdemir*,AttilaGursoyandOzlemKeskin
KoçUniversity{[email protected]}
Single amino acid variations (SAVs) on protein-protein interaction (PPI) sites play critical roles in diseases.Distribution of SAVs within PPI site (interface) residues may be related to their classification as disease-causingorbenign. Interfaceshavea small subsetof residues calledhot spots that contributemore to thebinding energy, and they may form clusters called hot regions. We performed statistical and structuralanalysesofSAVswithexperimentalthermodynamicdata,anddemonstratethatSAVsdestabilizingPPIsaremore likely to be found in hot regions and hot spots rather than energetically less important interfaceresidues.Incontrast,non-hotspotresiduesaresignificantlyenrichedinneutralSAVswhichdonotaffectPPIstability. Hot spots also tend to be enriched in destabilizing and disease-causing SAVs, while neutral andbenignSAVssignificantlyoccurinnon-hotspotresidues.Also,analysisoftemperaturefactorshowedthathotspotSAVresiduesaregenerallylessflexiblecomparedtooverallSAVresiduesandhotspotresidueswithoutSAVs.FordistributionofSAVswithinsecondarystructureelements(SSEs),weobservedthatdisease-causinghotspotSAVstendtobelocatedincoil,turnandbridges,whilebenignhotspotSAVsaremorelikelytooccurinhelices.Ourworkdemonstrates thatdistributionofSAVswithindifferent interface residuesand regionscanbeimportanttopredicttheeffectofSAVs.
Interactionoffarnesylated,butlikelydepalmitoylated,RasisoformswithPDEδ
SerenaMuratcıoğlu*,HyunbumJang,AttilaGursoy,OzlemKeskinandRuthNussinov
KoçUniversity{[email protected]}
PDEδshuttlestheK-Ras4Bisoformfromtheendoplasmareticulumtotheplasmamembrane.ThisisacriticalstepsinceRasneedstobemembrane-attachedthroughitshypervariableregion(HVR)torelythesignal.Thehypervariableregion(HVR)ofK-Ras4B isusually farnesylated,butcanbegeranylgeranylated.Experimentaldata indicate thatPDEδbindsK-Ras4BaswellasN-Ras;butnotK-Ras4A,which isa splicevariantofKRASgene.TheHVRsofK-Ras4BandK-Ras4Aaresimilar,botharehighlypositivelycharged.N-Ras,ontheotherhand,isnearyneutral.K-Ras4Bisfarnesylated;K-Ras4AisfarnesylatedandpalmitoylatedasisN-Ras.Unlikefarnesylation,palmitoylation is reversible.Herewe investigated thebindingofPDEδwith farnesylatedandgeranylgeranylated K-Ras4B, and analyzed the interactions.We observe that themajor contributor to theinteraction is the farnesyl/geranylgeranyl moiety through docking into PDEδ’s hydrophobic pocket.Importantly, we also observe that the flexible depalmitoylated K-Ras4A HVR can also be accommodated,corroborating with the experimental observation that depalmitoylated N-Ras can interact with PDEδ.Wethuspropose thatPDEδcanbind todepalmitoylatedK-Ras4A inoncogenic cells suchasadenocarcinomas,which are K-Ras4B cancers, and that drugs blocking the shuttling of K-Ras4B may also hinder that ofdepalmitoylatedK-Ras4A.
21
SessionII:HealthInformatics
WearableSolutionsforParkinson'sDiseaseMonitoring
KorayAçıcı,ÇağatayBerkeErdaş,TunçAşuroğlu,MünireKılınçToprak,HamitErdemandHasanOğul
Parkinson'sDisease(PD)isaneurodegenerativeproblem,whichoccursduetothelackofenoughdopamineinthebrain.Whilethereisnotcurrentlyaknowncureforthedisease,thecrucialissueformedicaltreatmentof the patients with PD is constant monitoring of the disease symptoms such as tremor, bradykinesia,dyskinesiatomaintainand/orimprovetheirqualityoflife[1].SincethePDpatientsusuallysufferfromlosingbasicmotor abilities, their remotemonitoring is a recent challenge to serve a satisfactory home care andclinicalsupport.Currentcomputationalresearchonthetopichasbeenfocusedondevelopingsystemsandalgorithmsforrecognizingtheactivitiesofdailylivingusingmotionsensors[2-4].Theclinicalstudiescanonlybenefit from these computational efforts formaking binary decisions about disease state [5-7]. Here, weoffer a signal-processing-based solution for quantitatively monitoring PD motor symptoms in addition torecognizing daily living activities throughwireless on-body sensors. Specifically, our approach can provideeffectivediagnosis andprognosis solutions fromgait analysis viamultiple foot-worn sensors thatmeasureground reaction force (GRF).We argue that a regression-based analysis of GRF signals acquired from gaitsensors can accurately predict the PD symptoms in terms of Universal Parkinson Disease Rating Scale(UPDRS).Anexperimentalanalysisconductedonarealdatasetacquiredfrom73healthycontrolsand93PDpatients has justifiedour argument.Main advantageof this solution is the fact that an exact valueof thesymptom in UPDRS can be inferred rather than a categorical result that defines the severity of motordisorders.Anotheradvantage is thefact thatthemethodcanadditionallyreport theactivityperformedbythepatientwhenthecurrentsymptomanalysisisbeingcarriedout.
Acknowledgement:ThisstudywassupportedbytheScientificandTechnologicalResearchCouncilofTurkey(TUBITAK)undertheProject115E451.
References1. Patel, S., Lorincz, K., Hughes, R., Huggins, N., Growdon, J., Standaert, D., Akay, M., Dy, J., Welsh, M.,Bonato,P.:MonitoringmotorfluctuationsinpatientswithParkinson'sdiseaseusingwearablesensors. IEEETransactionsonInformationTechnologyinBiomedicine13(6),864-873(2009).2. Aminian, K., Najafi, B.: Capturing human motion using body-fixed sensors: outdoor measurement andclinicalapplications.ComputerAnimation&VirtualWorlds15(2),79–94(2004).3. Erdaş Ç.B., Atasoy I., Açıcı K., and Oğul H., Integrating features for accelerometer-based activityrecognition,”ProcediaComputerScience98,522-57(2016).4. Asuroglu T., Acici K, Erdas C.B., Ogul H., Texture of Activities: Exploiting Local Binary Patterns forAccelerometerDataAnalysis,12thInternationalConferenceonSignal-ImageTechnologyandInternet-BasedSystems,Naples(2016).5.Daliri,M.R.:Chi-squaredistancekernelof thegaits for thediagnosisofParkinson’sdisease.BiomedicalSignalProcessingandControl8(1),66–70(2013).6.Jane,Y.N.,Nehemiah,H.K.,Arputharaj,K.:AQ-backpropagatedtimedelayneuralnetworkfordiagnosingseverityofgaitdisturbancesinParkinson’sdisease.JournalofBiomedicalInformatics60,169–176(2016).7.Ertugrul,O.F.,Kaya,Y.,Tekin,R.,Almali,M.N.:DetectionofParkinson’sdiseasebyShiftedOneDimensionalLocalBinaryPatternsfromgait.ExpertSystemswithApplications56,156-163(2016).
22
AComputerAidedDiagnosisSystemforHeartAttackUsingDecisionTree
İlknurBUÇANKIRKBİR1,BurçinKURT1,KemalTURHAN1
1DepartmentofBioistatisticalandMedicalInformatics,KaradenizTechnicalUniversity,Trabzon,Turkey
[email protected],[email protected],[email protected]
Purpose-Myocardialinfarction(MI)whichisgenerallynamedasheartattackthatoccursasaresultofthecloggingofbloodvessels.Inthiscase,theheartmusclecellswhichcannotgetenoughoxygenbecauseofnotgettingenoughblood,damagemayoccurandifheartmusclestaywithoutoxygenforalongtime,deathmayoccur.50%ofdeathswhicharefromtheheartattackhappeninfirstonehourandthisratioswellsto80%infirsttwentyfourhours.Durationoftreatmentandtransportationofpatienttothehospitalplayabigroleindeaths which are from the heart attack. Therefore, a computer aided diagnosis (CAD) system has beendevelopedtohelpdoctorsinthediagnosisoftheheartattackinthisstudy.
Methods- Inthescopeofthestudy,datasof350patientswhichhadappliedwithchestpaincomplaintatFarabihospitalemergencyserviceofmedicinefacultyofKaradenizTechnicalUnıversity in2014-2016,werediagnosed and non-diagnosed with heart attack, have been used. These datas have been obtained byexaminingresultsofthebiochemistrylaboratorytest,epicrisisreportsandresultsoftheangiographyreportsin consultancy with specialist physician. Gender, creatine kinase-MB (CK-MB), high sensitive troponin I,changeofSTsegmentandchangeofECGparametershavebeenusedasinputs.ForthedevelopmentofCADsystem,thedecisiontreemodelwhichisaoneofthemachinelearningmethods,hasbeenused.InthisCADmodel,205ofthe350datausedastrainingsetandtheremaining145wereusedastestset.
FindingsandResults-Consequently,aquitesuccessfulCADsystemforheartattackhasbeendevelopedwith91.7%sensitivity,97.7%specificity.Inadditiontothis,AUCvalueof0.94andKappavalueof0.86havebeenobtainedwhichareverysatisfactoryaccordingtothesimilarstudiesintheliterature.
Keywords–Heartattack,computeraiddiagnosis,decisiontree.
CardiovascularModellingforAbdominalAorticAneurysms
EmrahAkkoyun*,AybarCanAcar,ByronZambranoandSeungikBaek
MiddleEastTechnicalUniversity{[email protected]}
Anaorticaneurysmisdiagnosedwhenthereisalocalexpansionoftheaorta,carryingbloodcomingfromtheheart.Whentheexpandedportionis50%largerthanthenormalvesseldiameter(30mmormore)thelesionisdefinedasananeurysm.Althoughaneurysmsare seen indifferent regionsof theaorta, theycommonlypresentintheabdominalregion.Mostsmallaneurysmshavenosymptomandwereconsideredsafe,whilelargeaneurysmsmaybe fatal in thecaseof rupture,whichcausesmassive internalbleeding.This requiresimmediate interventionand, even then, generally resultsdeath (a90%mortality rate). It is the13thmostcommonkiller in theU.S. (Wilminketal,1998;Ernstetal,1993).Aneurysmrepairprior to rupture is thusvital.Studiesshowthatboththelongtermmonitoringofpatientspriortoanysurgicalintervention;aswellas the EVAR intervention required, if surgery is selected, has their own risks. Therefore, physicians arerequiredtoassessbothriskscarefullytoselectappropriatewayforward.Thestateoftheartistomakethisdecisionbasedonthediameteroftheaneurysmorannualaneurysmgrowth.Scientificresearchhasshownthatthecriterion isnotaccurateforpredictinganeurysmrisk (Vorpetal,2007;Limetetal.1991)and it isnecessarytoconsiderotherparametersforassessingrisk.Furthermore,visualizingthe3Dmodelwithacolor
23
mapisveryusefultosurgeonsbeforetheoperation.Inthisstudy,biomechanicalbehavioroftheaneurysmisanalyzedinadetailedwaywithinthecontextofhemodynamicforcessuchaswallshearstressandvelocitypatterninordertobetterunderstandthereasonsforgradualaneurysmgrowthandpotentialrupture.
Thereisaconsiderableamountofliteratureonriskassessmentofcardiovasculardiseasesaswellasfortheircomputationalmodeling. Furthermore, a numberof open-source software tools are available for differentpurposes such as medical image enhancement, automatic and manual segmentation of arteries/veins,creating a proper vascular model, mesh generation and applying computational flow analyses. Hence, aconstructed computational model, based on patient’s scan images (e.g., CT and ultrasound imaging), canvisualizethestatusoflesions,andbiomechanicalsimulations(e.g.,bloodvelocity)canbeinformedtotrainedphysicians and aid them for decision-making. We present an end-to-end procedure that can be used toconstruct 3D models of the aneurysm and run hemodynamics simulations with realistic choices for flowparametersandprofiles.Ifnotdonesystematicallyandcarefully,suchsimulationsareusuallynotconsistentwithrealityand/ortakeunnecessarylongtimestocompute.
Todemonstrate, threepatientseach from the rupturedandnon-rupturedaneurysmgroups (e.g., total sixpatients)havebeenselectedandthesimulationswereperformedseparatelywithourprocedure.Theresultsforeachgroupwereobtained,comparedwiththereferencevaluesandverified.WeobservedhighWSSontherupturedaneurysmgroup,asexpected,withtheexceptionofonepatient.Figure1showsanexampleofWSSandvelocityvisualizationforbothgroupsduringthesystolicphaseatthesecondcardiaccycle.Theleftaneurysmbelongstonon-rupturedgroupwhiletherightonebelongstorupturedgroup.
More closely, when we investigate the wall shear stress distribution over the maximum values during acardiac cycle (ruptured: 6,62 and un-ruptured: 2,31), we can find a single threshold value where we candiscriminatetwoclasses fromeachotherwithoutanyexception.Therearesignificantdifferencesbetweenthegroupson these featuresevenall theotherhemodynamic forces investigatedareclose toeachother.Therefore,we are hypothesizing that distributions ofwall shear stress during a cardiac cyclemight be anindicatorofthepredictinggrowthrateandspeedofapproachtorupture.Nevertheless,thehypothesiswillbetestedfurtherinourongoingresearch,withlargerdatasets.
Insummary,inthissubmissionwedemonstratehowtoperformcardiovascularmodelingfromrawCTimagesto3Dmodeloftheaneurysmwherethecolormapshowsthedistributionofthehemodynamicforcesovertheaneurysmsurface.Theapproachcanfurtherbegeneralizedforuseinsimilarproblemsincardiovascularresearchinvolvingcoronary,cerebralandcarotidarteries.
24
SessionIII:CancerSystemsBiology
Alternativepolyadenylationpatternsforcancerclassification
OguzhanBegik,MerveOyken,TunaCinkilliAlican,TolgaCan*andAyseElifErsonBensan
*DepartmentofComputerEngineering{[email protected]}
Certainaspectsofdiagnosis,prognosisandtreatmentofcancerpatientsarestillimportantchallengestobeaddressed. Therefore, we propose a pipeline to uncover patterns of alternative polyadenylation (APA), ahiddencomplexityincancertranscriptomes,tofurtheraccelerateeffortstodiscovernovelcancergenesandpathways.Weanalyzedexpressiondata for1,045 cancerpatients and founda significant shift inusageofpoly(A) signals in common tumor types (breast, colon, lung, prostate, gastric and ovarian) compared tonormal tissues. Using machine-learning techniques, we further defined specific subsets of APA events toefficientlyclassifycancertypes.Overall,ourstudyoffersacomputationalapproachforuseofAPAinnovelgene discovery and classification in common tumor types, with important implications in basic research,biomarkerdiscoveryandprecisionmedicineapproaches.
IntegratingMultipleDataTypesforCancerSubtypeDiscovery
TundeAderinwaleandHilalKazan*
AntalyaInternationalUniversity{[email protected]}
Cancerisaheterogeneousdiseaseandidentificationofcancersubtypesiscriticalforpersonalizedtreatmentanddrugdevelopment.Recently,cancergenomeprojectshaveproducedmultipletypesofhigh-throughputdata for thousands of cancer patients. Exploiting the complimentary information between different datatypescanimprovefindingsubtypes.Here,weproposeanewmethodthatintegratesmultipledatatypesforcancer subtype discovery.We focus on five cancer types: Breast Invasive Cancer, Colon adenocarcinoma,Glioblastoma multiform , Kidney renal clear cell carcinoma and Lung squamous cell carcinoma. We usegenome-widemeasurementsofgeneexpression,DNAmethylationandmiRNAexpressionofcancerpatientsfrom The Cancer Genome Atlas (TCGA) database. An intermediate integration method was used bytransformingeach typeof dataset into a radial basis function kernel andusingmultiple kernel learning toinfertheoptimalweightsforkernelcombination.Combiningkernelswithlearnedweightsgiveclusterswithhigher silhouette score compared to combining the kernels uniformly. Moreover, Kaplan-Meier analysisshowsthatourdiscoveredclustershavedistinctsurvivalprofileswithstatisticallysignificantlog-ranktestp-values.
25
IdentificationofGeneMutationsInvolvedinDrugResistanceinLiverCancerUsingRna-SeqDataAnalysis
MonaShojaei,EceAkhan,AybarCanAcarandRengülÇetinAtalay
Cancer is the leading causeof deathworldwide and the risk increaseswith aging.A significant concern incancerresearchisthedetectionofcancerdrugresistanceassociatedsomaticmutations.AccordingtoGlobalCancerStatistics(GCS)livercanceristhe5thmostcommonand2nddeadliestcancerintheworld.Overthepast two decades, the death rate for liver cancer increased by 2.5% per year and incidentswere 3 timeshigherinmenthaninwomen(AmericanCancerSociety).Appropriatetreatmentofhepatocellularcarcinoma(HCC, primary liver cancer) depends on the disease stage, patient’s age and overall health and individualpriorities.Targetedtherapiesareknowntoblockcancer-associatedproteinsorpreventcellproliferationandinvasion. Unlike conventional chemotherapy, which affects the whole normal and cancerous fast-growingcells,targeteddrugsattackspecificmoleculesincancercellsandhavemuchlessimpactonhealthytissues.Withtargetedtherapyinmind, inthisstudy,therelationshipbetweenmutationstatusanddrugtreatmentresponseofwell-differentiatedHuh7andpoorly-differentiatedMahlavulivercancercellswereanalyzed.TheRNA-Seq data of each cancer cell line (as control) was compared to “sorafenib” and “PI3K/Akt Pathwayinhibitors” treated samples. Somatic mutations associated with drug resistance were comparativelyidentifiedwithMuTecttool(Cibulskis,2013).
The results were then filtered to distinguish the missense mutations. The common genes among drug-resistant sets were found to be associated with liver cancer perseverance and aggressiveness. SLC39A5,FRG1,PPHLN1andSRP9genemutationswere found tobe themost significant, sharedamong threedrugtreated sets. The sets were further investigated in detail to discover the liver cancer associated survivalgenes.Usingourresults,appropriatetargetscanbedefinedthatplaycriticalrolesincancerouscellgrowthfordrugdevelopmentpurposes.Drugswithspecifictargetscanfindtheappointedplaceandturnthetargetgeneofftodisturbthecancercells’proliferation.Accordingly,themutatedgenesactivitiesarestoppedandthediseaseprogressionisprevented.
Themutatedgenesthatweidentifiedduringthechemicalknockdownstudiescanbefurtherstudiedingeneexpression vs. patient survival data. These genes are being analyzed in laboratory to test whether theirsilence (knocking-down founded gene mutations which are correlated with liver cancer under drugtreatments) decreases cancer growth or not. In our success we can target these genes in future cancertreatments.
Cibulskis, K. et al. Sensitive detection of somatic point mutations in impure and heterogeneous cancersamples.NatBiotechnology(2013).doi:10.1038/nbt.2514
26
SessionIV:Genomics
LosslessPruningofAHPSNPPrioritizatonTreeUsingRandomForestVariableImportances
ArifYılmazandYeşimAydınSon*
METUInformaticsInstitute{[email protected]}
SubjectivityisanoldyetunsolvedprobleminMultipleCriteriaDecisionMakingincludingAnalyticHierarchyProcessing (AHP). Here,we have proposed amachine learning based analytic hierarchy process (ML-AHP)method to address expert judgment uncertainty in decision making system design. It is accomplished bytraining a classifier algorithmaccording toAnalyticHierarchyProcess inputdata andevaluating categoriesbased on variable importance values. As a comparative case study Single Nucleotide Polymorphism(SNP)prioritization in the bioinformatics domain is presented. Variable Importance figures provided by theemployedmachinelearnerareusedtoevaluatetheimportanceofAHPcategories.InthisstudyweselectedRandom Forest. It was discovered that most of the expert defined weights of categories were zero andperformancewas identicalafterpruning.AnalysisontheProstateCancerandtheType2DiabetesMellitusdisease data were performed to demonstrate the benefits of the proposed approach, where pairwisecomparisonsofcategoriesrequirednoexpertevaluation.Hence,subjectivity,uncertaintyandimprecisionisavoided. Implementationof theproposedmethodcanenhanceevaluationof the categoryweights inAHPandalsoinothermultiplecriteriadecisionmakingmethods.
DiscoveringLargeGenomicInversionsUsingLong-rangeInformation
FatihKaraoglanoglu*,MarziehEslamiRasekhandCanAlkan
BilkentUniversity{[email protected]}
Nextgenerationsequencing(NGS)hasbeenprovidingevercheapermeansforobtaininglargeamountsof genomic sequencing data over the last decade. These data have been utilized to call a varietygenome-levelvariations.MostwidelyusedformofNGSdatatypicallyconsistofpaired-endshortreads(~100-150bp),withlowsequencingerrorrate.Thislowerrorrateallowsforrelativelyeasydiscoveryofsmall genomic variations. However the discovery of larger variations (>50 bp), known as structuralvariations (SV), is a much harder problem. This is due to the fact that most of short reads wouldcompletelyliewithinanSVregion,andthusprovidenoinformationabouttheexistenceofsaidSV.Evenwhen short reads capture SVsbreakpoints, correctlymapping these short reads isnot trivial since SVbreakpointstypicallylieinduplication-orrepeat-richregions.
We present VALOR (VAriation using LOng Range information), an algorithm for discovering largeinversions. VALOR utilizes the linked-read sequencing data generated using the new 10X Genomicsplatform,which tagspaired-end short-reads thatoriginate from the same largeDNAmoleculewith aunique barcode. VALOR uses this long-range information to discover large inversions. It is alsostraightforwardtoextendVALORforthediscoveryofotherformsoflargeSV,andincorporatesplitreadandlocalassemblysequencesignaturestoimproveaccuracyandbreakpointestimates.
27
Neighbouringsequencecontext-basedfixationbiasofmutationsinchimpanzeegenome
ElifBozlak*,YetkinAlıcı,EvrimFer,MelikeDönertaş,RasmusNielsenandMehmetSomel
InformaticsInstitute,METU{[email protected]}
The rate at which different types of mutations occur in genome, and the rate they spread throughoutpopulationsandfix,canvarydramatically.Forexample,transitionsoccurmorefrequentlythantransversions,andCtoTmutationsatCpGsitesarethemostcommonmutationtypeinmammalians.Inadditiontothesemutationratebiases,GC-biasedgeneconversionisawellknownfixationbias,suchthatmutationsfromA/TtoG/Cinrecombinationhotspotscanspreadthroughoutpopulationsandconsequentlygetfixedathigher-than-neutralrates.Previouslywestudiedpossiblefixationbiasesinthehumangenometakingneighbouringsequence context into account, analysing 5bp sequences with themutated base in themiddle.We havefoundthatmutationsthatextendthesamebasesequences(e.g.AGCGG->AGGGG)arefixedmorerapidlyincomparisontomutationsthatdisruptsuchsequences(e.g.AGGGG->AGCGG).Thechimpanzeeisthemostclosely relatedspecies tohuman,andhas thebeststudiedgenomeamongprimates,besideshuman.Herewe study whether this fixation bias exists also in the chimpanzee lineage. Using published chimpanzeepopulation genomic data from 75 individuals, and comparative genomic data of chimpanzee-human-orangutangenomealignment,wedetermineancestral stateof chimpanzeebases throughoutgenomeandclassifymutationsonthatlineageaspolymorphicorfixed.UsingtheMcDonald-Kreitmantestwethencheckforafixationbiasofallsinglenucleotidemutationsintheir5mersequencecontext,relativetotheirreversemutations.
SessionV:Multi-omics
Identificationofdysregulatedpathwaysacrossmultiplecancertypes
CesimErten*,EvisHoxha,HilalKazanandEsraTepe
AntalyaInternationalUniversity{[email protected]}
A major challenge in cancer genomics is to distinguish driver mutations that are responsible for cancerdevelopmentfrompassengeralterationsthatareobservedduetochance.Recentstudiesshowthatthereisa limited overlap between the driver mutations across multiple samples of the same cancer type. Thismutationalheterogeneityisexplainedbythefactthatdrivermutationscantargetmultiplegenesinthesamefunctionalpathway.Also,alterationsof thealternativedrivergenes in the samepathwayexhibitamutualexclusivitypatternduetoreducedselectivepressure.
Inthisstudy,weproposeamethodthataimstoidentifydrivereventsofcancerbyconsideringfunctionallyrelatedgeneswithmutationprofilessubjecttotheabove-mentionedprinciples.OurmethodisbasedontheexistingHotnet2algorithm[1].Hotnet2takesasinputahumanPPInetwork,appliesarandomwalkstrategyimplementingaheatdiffusionprocesswhereheatscorrespondtomutationfrequenciesacross3110samplesfrom12cancertypes.Sucharandomwalkprovidesthesocalled‘exchangedheatmatrix’whereanentry(p,q)correspondstotheextentofthe influenceonqofthemutationsoccurringatp.Oncethe lowinfluencepairs are removed from consideration, the remaining “exchanged heatmatrix” corresponds to a directedgraphfromwhichstronglyconnectedcomponentsareidentifiedasoutputmodules.
28
We extended the existing Hotnet2 algorithm to incorporate mutual exclusion and coverage. Namely, weconstructed an edgeweighted PPI networkwhere each edgeweight between a pair of genes reflects thedegreeofmutual exclusionand the coverageof thepair across all the samples.Wemodified the randomwalksothattheedgeweightsaretakenintoconsiderationappropriately.TherestofthestepsarethesameasHotnet2 algorithm.When the total numberof genes in theoutputmodules is fixed, ourmodules havehigherscoresforconnectivity,mutualexclusionandcoveragecomparedtoHotnet2-identifiedmodules.ThenumberofknowncancerdrivergenesselectedbetweenourmethodandHotnet2iscomparable.Moreover,thesetofmodulesprovidedbyourmethodhasbetterGOconsistencyscoresthanthatofHotnet2.
In summary, ourmethod combines the knowledge in theexisting interactionnetworkwith cancer-specificprinciples (i.e., mutual exclusivity, coverage) to identify driver pathways. The identified modules will beinstrumentalinprovidingdirectionsfornewdiagnosticandtherapeuticstrategiesincancerbiology.
[1] Leiserson et al, Pan-cancer network analysis identifies combinations of rare somatic mutations acrosspathwaysandproteincomplexes.NatureGenetics,47(2):108(2015).
[2]Vogelstein,Betal.Cancergenomelandscapes.Science,339,1456(2013).
Integrativebioinformaticsanalysisofmulti-omicsdatafacilitatesfunctionalcharacterizationofthecandidatetumorsuppressorproteinCTCF
GuvanchOvezmyradov
Dept.ofBiostatisticsandMedicalInformatics,RegenerativeandRestorativeMedicineResearchCenter(REMER),IstanbulMedipolUniversity{[email protected]}
TheCCCTC-bindingfactor(CTCF)isamultifunctionalzinc-fingerprotein,knownas“themasterweaverofthegenome” for its unique role in coordinating the 3D organization of the genome and regulating geneexpression.CTCFproteinactsasatranscriptionfactorandoccupiesamyriadoftargetsitesacrossthehumangenome. Considering the genome-wide significance of CTCF, is this regulatory protein also implicated incancer? The answer remains elusive, partly becauseof the versatile natureof this protein and the lackofexperimental studies addressing this question. Current evidence only loosely links CTCF to cancer as acandidatetumorsuppressorgeneandmoreworkisneededtoclarifytheCTCF'sinvolvementincancer.Vastdata generated by numerous experimental studies, which is available in multiple databases, offersunprecedentedpossibilitiestoaddressthis issue.Inthisstudy,multipletypesofomicsdataareanalyzedinthe context of CTCF and cancer using dataminingmethods. The integrative bioinformatics approach linksCTCF-associated transcriptomics, interactomics and functional annotation data by reconstructing andanalyzing the CTCF protein-protein interaction network. The resulting putative CTCF interaction mapimplicates CTCF with distinct biological processes and co-expression pattern of CTCF interaction partnersprovidesindicationsabouttheunderlyingcellularmechanisms.Preliminaryresultsfromthismeta-analysisofmulti-omicsdatacomplementsexperimentalresearchfindingsandprovidenovelinsightsintoimplicationofthiscandidatetumorsuppressorproteinincancer.
29
SessionVI:ComputationalDrugDiscovery
RDFGraphEmbeddingsforPredictionofDrug-DrugInteractions
RemziÇelebi,ErkanYaşar,ÖzgürGümüşandOğuzDikenelli
EgeUniversityComputerEng.Dept.{[email protected]}
Linked Open Data (LOD) is a technique for publishing, describing, and linking data. Linked open data is apotential sourceofbackgroundknowledge formodelingpredictivemachine learningandbuildingcontent-based recommender systems. LOD identify resourceswithUniformResource Identifiers (URI) and throughstandardssuchastheRDF(ResourceDescriptionFramework).RDFisapowerfuldatamodeltodescribeandexchangeresourcesontheWeb.Bio2RDFisanopen-sourceprojectthat integratesnumerousLifeSciencesdatabases available on different websites. Bio2RDF provides a service for data integration service fromvarioussourcesasaresourceforscientificresearchers.Bio2RDFscriptsconvertheterogeneouslyformatteddatasuchasSQLandXMLintoRDFformat.Bio2RDFcreatesalargeRDFgraphthatinterlinksdatafrommajorbiologicaldatabases,includingDrugbank,KEGGandSIDER.
RDF2VecisarecentlypublishedmethodologythatadaptsthelanguagemodelingapproachofWord2VectoRDFGraphEmbeddings.Word2Vectrainsaneuralnetworkmodelto learnvectorrepresentationofwords,calledwordembeddings.Itmapseachwordtoavectoroflatentnumericalvaluesinwhichsemanticallyandsyntactically closer words will appear closer in the vector space. The hypothesis which underlies thisapproachisthatcloserwordsinwordsequencearestatisticallymoredependent.RDF2VecappliesasimilarapproachtoRDFGraphconsideringtheentitiesandrelationsbetweenentitiesbyconvertingthegraphintoset of sequences of entities (walks or paths) and trains the same neural network model to learn vectorrepresentationofentitiesintheRDFgraph.
Drug-drug interactions (DDIs) are a very important topic in drug discovery and public health. Drug-druginteraction(DDI)mayoccurwhenmultipledrugsareco-prescribed,andtheseinteractionshavethepotentialtoleadtopatientdeathordrugwithdrawal,apotentialwhichmakesittheobjectofgreatinterestfrombothacademia and industry. Prediction of potential drug drug interaction helps reduce unanticipated druginteractions, drug development costs and optimizes the treatments in the drug design. The underlyingassumption of the similarity based DDI prediction approach is that similar drugs may interact with samedrugs. There are already several similarity-based approaches, the most common of these are structural,therapeutic, phenotypic and genomic similarities [1]. The studyby Zhang,Wen, et al. [2] collected awidevarietyofdrugdataandthuspredicteddrug-druginteractionsbyintegratingchemical,biological,phenotypicandnetworkdata.Gottliebetal. [3]builtanovel framework, INDI,whichconsideredsevenkindsofdrug-drug similarities. The Villar et al. study [4] designed a novelmolecular fingerprint similarity based on DDIprofilesanddevelopedausefulinsilicomodeltopredictnewdruginteractions.Chengetal.[5]presentedaHNAI framework to predict drug interactions utilizing the drug phenotypic, therapeutic, structural, andgenomic similarities. They applied five machine-learning-based predictive models (decision tree, logisticregression,naiveBayes,k-nearestneighbor,supportvectormachine)onadrug-druginteractiondataset.
In thiswork,we have applied RDF2Vec to extract feature vector representation of linked open drug datafromsubsetofBio2RDFdatasetstopredictpotentialdrug-druginteractions.WegeneratedwalkstobeusedasinputforvectorrepresentationofdrugsonRDFgraphdata.Asimilarityorrelatednessscorebetweenanytwodrugscouldbecalculatedbytakingthecosineofthosedrugvectorrepresentations.Weextendedourprevious work [1] by integrating the RDF Graph Embedding based drug similarities to train a logistic
30
regressionclassifierforDDIprediction.Ourpreliminaryresultssuggestthatdrugvectorrepresentationbasedsimilaritiescouldenhanceexistingpharmacologicalsimilarity-basedDDIpredictionmethods.TheAUCvaluehasbeenincreasedfrom0.67to0.76basedonfivefoldcross-validationwiththesenewsimilarities.
References
1- Çelebi, R.,Mostafapour, V., Yasar, E., Gümüs, Ö., & Dikenelli, O. "Prediction of Drug-Drug InteractionsUsingPharmacological SimilaritiesofDrugs."DatabaseandExpert SystemsApplications (DEXA), 201526thInternationalWorkshopon.IEEE,2015.
2-Zhang,W.,Chen,Y.,Liu,F.,Luo,F.,Tian,G.,&Li,X.(2017).Predictingpotentialdrug-druginteractionsbyintegratingchemical,biological,phenotypicandnetworkdata.BMCbioinformatics,18(1),18.
3-Gottlieb A, Stein GY, Oron Y, Ruppin E, Sharan R. INDI: a computational framework for inferring druginteractionsandtheirassociatedrecommendations.MolSystBiol.2012;8:592.
4-Vilar,S.,Uriarte,E.,Santana,L.,Tatonetti,N.P.,&Friedman,C.(2013).Detectionofdrug-druginteractionsbymodelinginteractionprofilefingerprints.PloSone,8(3),e58321.
5-Cheng F, Zhao Z. Machine learning-based prediction of drug-drug interactions by integrating drugphenotypic,therapeutic,chemical,andgenomicproperties.JAmMedInformAssoc.2014;21(e2):e278–86.
DrugRepurposingforAgeing:AConnectivityMapApproach
HandanMelikeDönertaşandJanetThornton
EMBL-EBI{[email protected]}
Ageing is broadly defined as time-dependent functional decline and loss of physiological integrity inbiological systems leading to reduced homeostasis and susceptibility to many pathologies includingneurodegenerative disorders. Genetic experiments on model organisms proved the possibility oflifespanextensionupto10-fold.Whilenotaspowerfulasthegenetic interventions,severaldrugsarealso tested in model organisms and shown to modulate lifespan or improve health during ageing.Identification of new drug candidates, however, is not straightforward mainly due to molecularmechanismsofageingprocessbeingmostlyuncovered.
Inthisstudy,wehaveusedadrugrepurposingapproach,i.e.applicationofexistingdrugstotargetnewbiologicalprocesses,allowingidentificationofdrugstargetingthesamepathwaysthatchangeinageing.Usingmultipleage-seriesgeneexpressiondatasets(microarrayandRNA-Seq)thatareavailableinpublicdatabases,weidentifiedgenesshowingaconsistentchangewithageinginhumanbrain.Wecomparedtheir expression changes during ageing with the drug perturbed gene expression changes. Using theConnectivityMapdatabase,whichincludesdrugperturbationexperimentsfor1309drugs,weidentifiedasetofdrugswhichcanpotentiallymodulateageingprocessandextend lifespan.Thetopcandidateswe identified include drugs such as rapamycin, which are known to modulate lifespan in modelorganisms.Theapproachhasadvantagesthatitstartswithhumanageingdata,increasingthepossibilitythat these drugs can also be functional in humans and it is not biased toward previously identifiedageingrelatedpathways,makingitpossibletodiscovernewproteintargetsforageingstudies.
31
SessionVII:ProteinFunctionPrediction
InvestigationofMulti-taskDeepNeuralNetworksinAutomatedProteinFunctionPrediction
AhmetSureyyaRifaioglu1,*,MariaJesusMartin2,RengülÇetin-Atalay3andMehmetVolkanAtalay1,TuncaDoğan2,3
1DepartmentofComputerEngineering,MiddleEastTechnicalUniversity,Ankara,Turkey,2EuropeanBioinformaticsInstitute(EMBL-EBI),Hinxton,Cambridge,CB101SD,UK,3CanSyL,GraduateSchoolof
Informatics,MiddleEastTechnicalUniversity,Ankara,06800,Turkey*Towhomcorrespondenceshouldbeaddressed:[email protected]
INTRODUCTION
Functional annotation of proteins is an active research field for understanding molecular mechanisms ofliving-beingsandalsoforbiomedicalpurposes.SeveralGeneOntology(GO)basedproteinfunctionpredictionmethodshavebeenproposed inthe lastdecade.However,consideringthepredictionperformancesoftheproposedmethods,itcanbestatedthatthereisstillroomforsignificantimprovementsinthisarea(1).Deeplearningtechniquesbecamepopularinrecentyearsandtheyturnedouttobeanindustrystandardinseveralareas such as computer vision and speech recognition. To the best of our knowledge, as of today, deeplearningalgorithmshavenotbeenappliedtothelarge-scaleproteinfunctionpredictionproblem.Here,wepropose a hierarchical multi-task deep neural network architecture, DEEPred, as a solution to proteinfunctionpredictionproblem.Firstofall,weinvestigatedthepotentialofemployingdeeplearningmethodsforproteinfunctionprediction.Forthispurpose,wemeasuredtheperformanceofourmodelsatdifferentparametersettings.Furthermore,weexaminedtherelationshipbetweentheperformanceofthesystemandthesizeofthetrainingdatasets.
METHODS
For training, we created a dataset using GO term annotations from UniProtKB/SwissProt with manualexperimentalevidencecodes.WedividedthisdatasetaccordingtodifferentlevelsinGOhierarchy(i.e.9to12 levelswith respect todifferentGOcategories).Theobjectivehere is tocreateamulti-taskdeepneuralnetwork model for GO terms at each level of a GO category. The models were trained with drop-outtechniquetoavoidoverfitting.Forthegenerationofproteinfeaturevectorsattheinput,weusedamodifiedversionofsubsequence-basedfeatureextractionmethodcalledSPMap(2).Ageneralviewofthemethodisshown in Figure 1A. In the proposedmethod, a task corresponds to a GO term, therefore,when a querysequence is fed to ourmodels as input, DEEPred calculates a score for each trained GO termwithin themodel (Figure 1B).We determined a threshold value for each GO termwithin a level to provide reliablepredictions.
Level specificmodels consider the GO termswithin a level of GO directed acyclic graph and they do notconsider the GO terms between different levels. Therefore, we used a method to provide predictionsconsideringalllevelsofGOhierarchy.Inthismethod,inordertoprovideapositiveGOtermpredictionforaqueryprotein,we also check theprediction scores of its parents.Wepresent the correspondingGO termprediction, if the target GO term andmajorıty of its parents are over the predetermined thresholds.Wecreatedseveralmulti-taskfeed-forwarddeepneuralnetworkmodelswithseveralparametersfornumberofhiddenlayers,numberofneurons,learningrateanddrop-outrateandusedthebestperformingmodelsforeachlevel.
32
RESULTS
WecreatedsixtrainingdatasetsforeachGOcategorywhereeachdatasetcorrespondstoGOtermshavingprotein associations with numbers greater than the specified thresholds; (i.e. > 50,100,200, 300,400,500annotatedproteins).Alevelspecificevaluationwasperformedforperformancecalculations.Here,ourfirstobjectivewas to demonstrate the relation between the GO levels (e.g. generic/specific) and classificationperformance. The second aim was to show the relation between performance and size of the trainingdatasetsthatwereusedforeachGOterm.Theresultsshowedthatthereisageneraltrendofperformanceincreasewith the increasing number of training sampleswhichmeans that includingGO termswith smallnumberofproteinassociationsdecreasestheperformance(Table1).However,weobservedthatthereisnocorrelation between GO levels and performance. Overall system performancewas evaluated byMathewsCorrelationCoefficient(Table1).Theresultsshowedthatthesystemperformancewassatisfactoryanddeeplearning can be employed to improve the performance for hard to predict GO categories such as thebiologicalprocessandthecellularcomponent.Asaconclusion,weshowedthatdeeplearningtechniqueshasa significant potential in protein function prediction. We plan to further optimize the models to provideDEEPredasanopen-accesstooltotheresearchcommunity.
FIGURE
Figure1:OverviewofDEEPredarchitecture:(A)Trainingdatasetandmodelconstruction;(B)Modeltrainingforalevel
Table1:Performanceresultsforlevelspecificandoverallsystemevaluation.
LevelSpecificPerformance(F-Score) OverallPerformance
(MathewsCorrelationCoefficient)50 100 200 300 400 500MolecularFunction 0.69 0.72 0.71 0.76 0.82 0.82 0.75
BiologicalProcess 0.42 0.42 0.47 0.47 0.50 0.51 0.49
CellularComponent 0.63 0.65 0.66 0.69 0.70 0.73 0.63
REFERENCES
1.Jiang,Y.,Oron,T.R.etal.(2016).Anexpandedevaluationofproteinfunctionpredictionmethodsshowsanimprovementinaccuracy.Genomebiology,17(1),184.
2.Saraç,Ö.S.,Atalay,V.,andCetin-Atalay,R.2010GOPred:GOmolecularfunctionpredictionbycombinedclassifiers.PLoSONE,8pp.1:11
33
ASelf-trainingApproachforFunctionalAnnotationofUniProtKBProteins
RabieSaidi*,MaryamAbdollahyanandMariaMartinUniProt,EuropeanBioinformaticsInstitute,Cambridge{[email protected]}
1.INTRODUCTIONAutomaticannotationsystemsareessentialtoreducethegapbetweentheamountofproteinsequencedataand functional information in public databases such as UniProtKB (1). These systems rely on manuallyannotated (also called labelled) data to learn rules for predicting annotations.Manually labelled data are,however,oftenscarceortimeconsumingtoobtainastheyhavetobereviewedbyexperthumancurators.Ontheotherhand,unlabelleddataareabundantandcomparativelyeasytogather.Inthiswork,wepresenta self-training (2) automatic annotation approach that utilises unlabelled data in order to improve theaccuracy of predictions.Weevaluatedour systemon a set of entries inUniProtKB/Swiss-Prot. The resultsshow improvement indifferentperformancemetricswhen self-training isused. ThegeneratedmodelwasthenusedtopredictmetabolicpathwayinvolvementofUniProtKB/TrEMBLproteins.Asaresult, itcovered86%of theproteins currentlyannotatedbyUniProtpipelines,butalsocouldannotate6.7millionproteinsthatlackedanypreviouspathwayannotations.2.MOTIVATIONANDMETHODSIn a previous work, we introduced the Association-Rule-Based Annotator (ARBA), a multiclass annotationsystem for automatic classification and annotation of UniProtKB proteins. The system was evaluated onUniProtKB/Swiss-Prot prokaryotic data where it achieved very promising results (3). However, 89.7% OfUniProtKB/Swiss-Prot entries have been automatically annotated using HAMAP annotation rules (4). Thisleadstothefollowingquestion:canARBAstillpredictannotationsintheabsenceofannotationsprovidedbyHAMAPandusingonlymanualannotations?Toanswerthisquestion,weconsideredthetaskofpredictingmetabolicpathways inUniProtKBbacterialdatausing twodifferentapproaches: first, theoriginalARBAasdefinedin(3)andsecond,ARBAcombinedwithself-trainingasintroducedbelow.In order to dealwith small amount of labelled data, ARBAwas self-trainedonUniProtKB/Swiss-Prot data.Proteins that containpathwayannotations constitute the labelleddatasetwhile those thatdonot containany pathway annotation constitute the unlabelled dataset. The systemperforms self-training in twomainsteps. In the first step, annotations arepropagated from the labelled to theunlabelled setbasedon theirsimilarity. The similarity criterion is defined as a Boolean that is true if the two proteins have the sameattributes; e.g., signatures. In the second step, ARBA iteratively learns from these data, retrains itself andaddstothelabelledinstancesuntiladesiredperformancelevelisreached.Theoutputofself-trainingisthefinal learningdatasetusingwhich theannotationmodelwasbuilt. Thismodelwas validated in two2-foldcross-validationruns.Theresults,averagedoverthetworuns,areshowninTable1.Finally,weappliedthismodel to predict metabolic pathways in UniProtKB/TrEMBL bacterial data which are poorly covered,currently3.5%.Acomparisonbetweentheannotationcoverageofthesystembeforeandafterself-trainingisshowninFigure1.3.RESULTSANDDISCUSSIONResults indicate thatwhile the systemperformswell in its original form (Precision = 98.4%, Recall=71.2%,AUC=86.2%), itsperformanceis improvedwithself-training,as illustratedbynoticeable increases inrecallandAUC(Precision=99.7%,Recall=89.4%,AUC=95.8%).
34
Figure1providesstatistics fromtheUniProtKB/TrEMBLproteinsannotatedbyARBA.Withoutself-training,ARBAcovered4,564,250entries,where3,083,501proteins(denotedbyNew)lackedanypreviouspathwayannotations, resulting in an increase in pathway coverage from 3.5% to 7.4%, and 1,480,749 proteins(denotedbyOverlap)hadbeenpreviouslyannotatedbyUniProtpipelines.1,224,264proteins (denotedbyMissing) thathadbeenpreviouslyannotatedwerenotcoveredbyARBA. Incomparison,withself-training,thenumberofproteinsexclusivelyannotatedbyARBAwasincreasedto6,687,267,resultinginacoverageof12.1%. The number of missed entries was notably reduced to 394,205. These results demonstrate thebenefits of using self-training algorithms to provide functional annotations for UniProtKBwheremanuallycuratedannotationsarerare.4.AVAILABILITYModels for pathway prediction, generatedwith andwithout self-training based on release 2017_02 alongwith a Java Archive (JAR) package for applying them to UniProtKB bacterial data are available athttp://www.ebi.ac.uk/~rsaidi/arba/self-training/.5.REFERENCES1. The UniProt Consortium. 2017. UniProt: the universal protein knowledgebase. Nucleic Acids Research45(D1):D158-D169.2.ZhuX.2005.Semi-supervisedlearningliteraturesurvey(ReportNo.1530).ComputerSciences,UniversityofWisconsin-Madison.3.BoudelliouaI.etal.2016.PredictionofMetabolicPathwayInvolvementinProkaryoticUniProtKBDatabyAssociationRuleMining.PLOSONE11(7):e0158896.4.PedruzziI.etal.2015.HAMAPin2015:updatestotheproteinfamilyclassificationandannotationsystem.NucleicAcidsResearch43(D1):D1064-D1070.
35
SessionVIII:Interactomics&miRNA
LeveragingTaxonomicRanksforImprovingMirrorTree-basedProteinInteractionPrediction
ErdemTurk*andBarisSuzek
MuglaSitkiKocmanUniversity{[email protected]}
Proteins as one of themainstays of living organisms. Proteins’ functional subunits, namely domains,physically interactwitheachotherand this is critical toprotein’s functional and/or structural roles inbiological systems. Identification of domain-domain interactions (DDIs) and, consequently, proteininteractionsisimportantinunderstandingofrespectiveprotein’sfunctionalroles.
Using computational methods to predict DDIs is a fast and inexpensive way to complementexperimental studies. These methods typically identify potential interacting proteins for furtherexperimentalvalidation.Inliterature,thereareseveralmethodsforDDIpredictionincludingDPEA(Rileyetal.,2005),PE(Guimarãesetal.,2006),DIPD(Zhaoetal.,2009),MLE(Dengetal.,2002),DIMA(Pagelet al., 2004), RDFF (Chen and Liu, 2005) and RCDP (Jothi et al., 2006). RDCP is a MirrorTree-basedmethod used to predict protein-protein interactions based on coevolution with the premise thatinteractingproteinsevolvetogetherandtheytendtohavesimilarphylogenetictrees.Thus,similarityofphylogenetic trees, computed as correlation coefficient between similarity matrices for respectivedomains,isusedtoanswerwhethertwodomains(andproteins)areinteractingwitheachotherornot.Unfortunately MirrorTree is known to be sensitive to taxonomic diversity of domains (Zhou andJakobsson,2013).
Inthiswork,weproposeanewDDIpredictionmethod.OurmethodisbasedonMirrorTree’sbutcopeswithtaxonomicdiversityproblembyintroducinganewtaxonomy-rankbasedstepinsamplingdomainsthatwillbeusedincomputationofsimilaritymatrices.Insteadofusingdomainsrepresentingspecies(orlower)leveltaxonomicranks,weselectahigherleveltaxonomicrank(e.g.family)andrandomlysampledomains representing each possible taxa at this level. The similarity matrices are computed for thesampledsetsandguaranteedtomaintainacertainleveloftaxonomicdiversity.
Wehavetestedourmethodwithdifferenttaxonomicranksrangingfromspeciestoclassanddifferentsample sizes (10 to 100 proteins). For comparison with other methods, we used the 6,634experimentally verified DDIs from DOMINE (Yellaboina et al., 2011) database as a positive set andgeneratedanequalsizenegativesetbyrandomlypairingdomainswithnoknowninteractions.Witha71.0%sensitivityand63.0%specificityourmethodoutperformsalltheexistingDDIpredictionmethodswith the exception of RDFF, which has rather limited coverage within DOMINE gold set (148 DDIpredictions). Theagreementbetweenourmethodand theotherpredictionmethods rangesbetween(67and75%)basedonDOMINE’sgoldset.
In conclusion, our method’s performance positions it as a viable alternative to existing predictionmethods. Furthermore, given the overlap between predictionsmade by ourmethod and others, ourmethodcanbeusedtocomplementothermethodsinameta-DDIpredictor.
36
UseofmiRNAinstrugglingofVarroaparasite
SedefErkunt*,NecatiAltindis,AhmetEfeKoseogluandCemalUn
EgeUniversity{[email protected]}
Beekeepinghasan important role inworldeconomybycontributing toagriculturalproductionandhumanfood production. In terms of number of bee colonies, Turkey ranks second in the world with 4.5 millioncoloniesafterChina,andabout50thousandfamiliesareearnedalivinginthiswork.Therearemanyfactorsthataffect this importanteconomicactivitynegatively.Thebiggestproblem inbeekeeping for thepast50yearsisVarroa(Varroadestructor)parasite.
InvestigationsinNewZealandhavereportedthatapproximately$400-900millionbudgetover35yearswerespentagainstforVarroaandonly55-70milliondollarsinNewZealandwereusedforeradicationofparasites.AlsoinTurkey,thereferencesfromBeekeepersAssociationindicatethatatotalof22.5millionTurkishLirasperyearwerespentfor4.5millionhivesinthefightagainstVarroaparasite.
Varroa,whilesuckingthebloodoftheadultbeesdirectly,causestheyoungindividualswhowillbeleavingtobeweakandwinglessandalsocausestheextinctionofthehiveinextremecases.ManymethodsofstruggleagainstVarroahavebeenused,withtheresultthatpesticideshavegainedmostweight.However,mostofthemethodshavenotbeensuccessfulenoughandpesticideshaveharmedbothhumanbeingsandhumanhealthbyleavingresidueonthehoney.
TheaimofthestudyistoproduceanewbiologicalproductforstruggleofVarroaparasite.Forthispurpose,Varroagenomewas targeted forending theviabilityofparasite.5differentgene regions (ATPsynthase8,cytochromeb,cytochromeoxidaseI,sodiumchannelprotein,NADHdehydrogenasesubunit2)wereselectedfromVarroa genome andmiRNA sequences targeting these regionswere designed by using bioinformatictools. It has been confirmed that the designed miRNA sequences do not target the bee and the humangenome. The synthesizedmiRNA sequences were applied to hives by spraying and the expression of thetargetedgeneregionswasmeasuredinVarroasamplescollected.
Afterthefirstapplicationinthestudy,2generegionswiththehighestdecreaseinexpressionamountweredeterminedandthesegeneregionswereusedinthesecondapplication.AsaresultofsecondapplicationtheexpressionanalysisshowedthatthemiRNAsequencewhichtargetsNADHdehydrogenasesubunit2hasthehighestinfluenceonVarroaparasite.
37
POSTERSESSION
Poster# Authors Title
1 AbdulahadBayraktar AStudyofCommonPathwaysofNeurodegenerativeDiseases
2 AbubakhariSserwaddaandÖmerSinanSaraç
Investigatingtheeffectofdifferentfeatureselectionstrategiesforclassificationofgeneexpressionsignaturesoftumorcells
3AhmetEfeKoseoglu,MortezaHaghi,IsmailKarabozandCemalUn
DetectionofHeatShockProtein(DnaK,DnaJandGrpE)HorizontalGeneTransfersAmongAcanthamoebapolyphaga,AcanthamoebaPolyphagaMimivirus(APMV),Amoeba-InfectingBacteriaandSputnikVirophage
4AhmetEfeKoseoglu,MuhammetUslupehlivan,ZarifaOsmanliandCemalUn
PosttranslationalModificationsofα-ConotoxinandTheirEffectonFeedingTypeofConusSpecies
5AhmetFarukAcar,JohanSanmartinBerglundandPeterAnderberg
SpokenDialogueInterfacefortheSmart4mdApplication
6*
AhmetSüreyyaRifaioğlu,MariaMartin,RengulCetin-Atalay,VolkanAtalayandTuncaDoğan
InvestigationofMulti-taskDeepNeuralNetworksinAutomatedProteinFunctionPrediction
7
AlperenDalkiran,AhmetSüreyyaRifaioğlu,TuncaDogan,VolkanAtalay,MariaMartinandRengulCetin-Atalay
PredictionofEnzymaticPropertiesofProteinSequencesBasedontheECNomenclature
8* ArifYılmazandYeşimAydınSon LosslessPruningofAHPSNPPrioritizatonTreeUsingRandomForestVariableImportances
9 AydanurŞentürk,ErdemSanalandNurhanÖzlü GlobalPhosphoproteomeAnalysisofXenopus
laeviseggextracts
38
10AyhanSerkanŞik,YeşimAydınSon,ErginSoysalandArsevUmurAydınoğlu
AConceptualDesignforGeneticInformationExchangeCodingStandardsinTurkey
11AyşeDeryaCavga,MehmetTardu,AttilaGürsoy,ÖzlemKeskinandHalilKavaklı
Cryptochromemutationsinp53mutantmiceincreaseanti-carcinogenicpathwaysanddamageresponses
12AyşegülTombuloğlu,HülyaÇöpoğlu,TülinGürayandYeşimAydınSon
ChangesinGeneExpressionProfileofHumanHepatocellularCarcinomaCellLine(Hepg2)InducedbyBoricAcidatHalfMaximalInhibitoryConcentration(IC50)
13 BesteMimaroğluAltınayandYeşimAydınSon
ADataMiningApproachforDirectingthePatienttoAppropriateTreatmentMethodinPainClinic
14BunyaminKasap,SerbulentUnsal,ZelihaKasap,AybarCanAcarandKemalTurhan
AMethodProposalforRevealingEpigeneticImmune-EscapeMechanismsUsing450kMethylationArrayData.
15 BurçinKurt,SerbülentÜnsal,İlknurBuçanKırkbirandKemalTurhan
SurvivalandEnrichmentAnalyzesofCisplatin-BasedDrugResistance-RelatedGenesforNon-SmallCellLungCancer
16
BurcuKarakaya,GizemKars,NeginRazizadeh,ÇağlaEceOlgun,PelinYaşar,GamzeAyazandMesutMuyan
Assessingputativeproteinpartnersofselectedestrogenresponsivegeneproductsbyayeast-two-hybridapproach#
17 BurcuTuranlı-YıldızandMeteYılmaz MolecularPhylogeneticsofHemeOxygenase
inCyanobacteria
18 BurcuYaldizandYesimAydinSon StructuralPropertiesofMethylatedHumanPromoterRegionsinTermsofDNAHelicalRise
19
ÇağlaEceOlgun,GamzeAyaz,BurcuKarakaya,GizemKars,NeginRazizadeh,PelinYaşar,NurcanTuncbağandMesutMuyan
TowardsproteomicanalysisofYPEL2interactingpartnersidentifiedwithproximity-dependentbiotinylation#
39
20 CanFirtina,A.ErcumentCicekandCanAlkan AProfileHMM-basedhybriderrorcorrection
algorithmforlongsequencingreads
21 CansuDincerandNurcanTuncbag StructuralModelingofthePatient-SpecificSignalingNetworksinGlioblastoma
22* CesimErten,EvisHoxha,HilalKazanandEsraTepe Identificationofdysregulatedpathwaysacross
multiplecancertypes
23DamlaGozen,HumaShehwana,OzlenKonu,TuncaDoganandRengulAtalay
Theidentificationofoxidativestressrelatedgeneprofilesinlivercancercellsastargetsfordiagnosticsandtherapeutics
24*ElifBozlak,YetkinAlıcı,EvrimFer,MelikeDönertaş,RasmusNielsenandMehmetSomel
Neighbouringsequencecontext-basedfixationbiasofmutationsinchimpanzeegenome
25* EmrahAkkoyun,AybarCanAcar,ByronZambranoandSeungikBaek CardiovascularModellingforAbdominalAortic
Aneurysms
26 ErdemŞanalandNurhanOzlu QuantitativeandNetworkAnalysisofCytokinesisSpecificPhosphoproteins
27* ErdemTurkandBarisSuzek LeveragingTaxonomicRanksforImprovingMirrorTree-basedProteinInteractionPrediction
28*EzgiKaraca,JoãoRodrigues,AndreaGraziadei,AlexandreBonvinandTeresaCarlomagno
AnIntegrativeFrameworkforStructureDeterminationofMolecularMachines
29 FaridehHalakou,EmelSenKilic,OzlemKeskinandAttilaGursoy MultipleConformationsofProteinsEnhance
Protein-proteinInteractionNetworks
30* FatihKaraoglanoglu,MarziehEslamiRasekhandCanAlkan DiscoveringLargeGenomicInversionsUsing
Long-rangeInformation
31
GamzeAyaz,PelinYaşar,BurcuKarakaya,ÇağlaEceOlgun,GizemKars,NeginRazizadeh,ŞeymaÜnsal,NurcanTunçbağandMesutMuyan
Proteininteractionapproachestoassignafunctionforanestrogenresponsivegeneprotein:CXXC5#
40
32 GökçeSengerandNurcanTunçbağ NetworkModelingoftheDasatinibTreatmentinGlioblastomaStemCellsbyDataIntegration
33
GökhanÖzsarı,AhmetSureyyaRifaioglu,TuncaDoğan,RengülÇetin-AtalayandMehmetVolkanAtalay
HierarchicalSubcellularLocalizationPredictionusingSupportVectorMachines
34 GuldenOlgun,OzgurSahinandOznurTastan DiscoveringBreastCancerSubtypeSpecific
lncRNAMediatedceRNAInteractions
35 GülşahKaradumanBahçeandYeşimAydınSon Genome-wideAnalysisofSpliceAcceptorSite
Motifs
36 GungorBudak,ErnestFraenkelandNurcanTuncbag
IdentificationofthePathwayLevelIschemicChangesbyIntegratingTemporalPhosphoproteomeinOvarianCancer
37* GuvanchOvezmyradov
Integrativebioinformaticsanalysisofmulti-omicsdatafacilitatesfunctionalcharacterizationofthecandiatetumorsuppressorproteinCTCF
38 HabibeCansuDemirelandNurcanTuncbag TheEffectofAlternativeSplicingonTumor
SpecificProteinNetworks
39* HandanMelikeDönertaşandJanetThornton DrugRepurposingforAgeing:AConnectivity
MapApproach
40 HaticeBüşraKonuk,MuhammetRaşitCesurandAlperYılmaz
IdentificationofTissueSpecificGenesandAssessmentofTheirIntersectionwithDifferentiallyExpressedGenesinCancerDataToUnderstandTumorHeterogeneityInSilico
41 HuseyinAlperDömandYesimAydınSon AFeatureSelectionModelforGenomeWide
AssociationStudiesofSchizophrenia
42IdilYet,PooryaParvizi,UlasIsildak,ZelihaGozdeTuranandMehmetSomel
StudyingstructuralvariationamongneuronsatoldageandinAlzheimer'sDiseaseusingsinglecellwholegenomesequencing
41
43* İlknurBuçanKırkbir,BurçinKurtandKemalTurhan AComputerAidedDiagnosisSystemforHeart
AttackUsingDecisionTree
44*
KorayAçıcı,ÇağatayBerkeErdaş,TunçAşuroğlu,MünireKılınçToprak,HamitErdemandHasanOğul
WearableSolutionsforParkinson'sDiseaseMonitoring
45KubraNarci,TuncaDogan,AybarCanAcar,TulinErsahin,NurcanTuncbagandRengulCetinAtalay
SystemsBiologyAnalysisofKinaseInhibitorsinCancerCellsUsingNextGenerationSequencingData
46 MalikYousef,WaleedKhalifaandLoaiAbdallah EnsembleClusteringClassificationAppliedon
PlantmicroRNAsData
47 MehmetOzcan,TurgutBastug,AliOsmanAcarandYaseminAksoy
ComputationalassessmentofthemostpotentialinhibitorsforGlutathioneS-transferaseP1-1
48 MeltemEdaOmur,EsrefCelik,BilalKermanandEmreKarakoc
MyelinInteractome:IdentificationofCell-CellInteractionsinMultipleSclerosisViaBipartiteGraphBasedProtein-ProteinInteractionNetworks
49 MericKinali,TuncaDoganandRengulCetin-Atalay
DifferentialanalysisofHepatocellularCarcinomaKinomebyChemicalandGeneticKnockouts
50 MerihAlphanKaradeniz,ErkanUnalMumcuogluandGuyPerkins
AutomaticSegmentationofCristaein3DElectronMicroscopyTomographyImagesUsinganArtificialNeuralNetworkandDirectionalGrowing
51MerveAkkulak,EmreEvin,ÖzlemDurukan,GüneşÖzhanandOrhanAdalı
EffectofresveratrolonmRNAandproteinexpressionsofVitaminDmetabolizingCYP24A1inhumanembryonickidneycellline(HEK-293)
52* MonaShojaei,EceAkhan,AybarCanAcarandRengülÇetinAtalay
IdentificationofGeneMutationsInvolvedinDrugResistanceinLiverCancerUsingRna-SeqDataAnalysis
42
53 MuhammetUslupehlivan,EcemŞenerandRemziyeDeveci InsilicoAnalysisofthePax6Protein
Glycosylation
54 MuhammetUslupehlivan,RemziyeDeveciandCemalÜn
InsilicoApproachtoRelationofPrionProteinGlycosylationwithScrapieDiseaseResistanceinDomesticSheep
55 MustafaTarımandCelalÖztürk MetabolicProfilingUsingPartialLeastSquaresDiscriminantAnalysiswithArtificialBeeColonyAlgorithm
56 NogayhanSeymenandEgeUlgen ComparisonofDifferentApproachestoClassifyRecurrenceofBreastCancer
57 NurcanTuncbagandErnestFraenkel Patient-specificNetworkModelingin
Glioblastoma
58*OguzhanBegik,MerveOyken,TunaCinkilliAlican,TolgaCanandAyseElifErsonBensan
Alternativepolyadenylationpatternsforcancerclassification
59 OğuzhanKalyonandAlperYılmaz SurveyofChimericmRNAsinHIVInfectedPatientRNA-SEQData
60ÖzlemDurukan,MerveAkkulak,SenaGjotaErgin,ŞevkiArslanandOrhanAdalı
Metformin-CisplatincombinationtreatmentaltersgeneandproteinexpressionsofCYP17A1inLNCaPcellline
61
ÖzlemÖzkan,YeşimAydınSon,ArsevUmurAydınoğlu,AhmetYalçın,AlperDöm,BurcuYaldız,KübraNarcıandOnurBaloğlu
ParticipatoryDesignMeetings:GeneticInformationIncludedPersonalHealthRecordApplication
62 PooryaParvizi,MehmetSomelandNurcanTuncbag NetworkAnalysisofTranscriptomicChanges
DuringAging
63* RabieSaidi,MaryamAbdollahyanandMariaMartin ASelf-trainingApproachforFunctional
AnnotationofUniProtKBProteins
64* RemziÇelebi,ErkanYaşar,ÖzgürGümüşandOğuzDikenelli RDFGraphEmbeddingsforPredictionofDrug-
DrugInteractions
43
65RumeysaFayetörbay,TulinErsahin,RengülCetinAtalayandNurcanTunçbağ
Network-BasedDiscoveryofDrugTreatmentsinHepatocellularCarcinoma
66 SaberHafezqoraniandHilalKazan Genome-wideanalysisforidentificationoflncRNAsthatspongeRNA-bindingproteins
67* SedefErkunt,NecatiAltindis,AhmetEfeKoseogluandCemalUn UseofmiRNAinstrugglingofVarroaparasite
68*SerenaMuratcıoğlu,HyunbumJang,AttilaGursoy,OzlemKeskinandRuthNussinov
Interactionoffarnesylated,butlikelydepalmitoylated,RasisoformswithPDEδ
69* SılaÖzdemir,AttilaGursoyandOzlemKeskin
Analysisofsingleaminoacidvariationsinhotspotandhotregionresiduesofprotein-proteininteractioninterfaces
70TubaEge,AyşeMineGenÇler-Özkan,AlaattinŞenandOrhanAdalı
FolkMedicinalPlantEpilobiumhirsutumL.andItsIngredientEllagicAcidAltersProteinandmRNAExpressionsofRatLiverBileAcidSynthesizingCYP7A1
71* TundeAderinwaleandHilalKazan IntegratingMultipleDataTypesforCancerSubtypeDiscovery
72 UğurSezerman,ZeynepÖzkeserliandBegümÖzemek
PossibleUsageofExpressionofCertainCLOCKGenesasapredictorofsurvivalinAMLcancerpatients
73 UğurToprak,SerbülentÜnsalandKemalTurhan
ComparingSurvivalTimesofLungSquamousCellCarcinomaPatientsfromCombinedRNA-seqandMethylationDatabyEstimatingRiskGroups
74UmutAgyuz,AhmetMelihOten,OktayKaplanandSuleymanNezihHekim
DoesMitochondrialGenesPasstoFetusFromMotherorFather
75 YusufBayrakceken,ZelihaGunnurDikmenandFilizAkbiyik TheCalculationofReferenceChangeValuefor
CoagulationTestParameters
44
76 YvesYannickYameniNoupoue BiometricEncryption:FromfingerprintImagetopersonaluniqueandirrevocablekey.
77 Hamitİzgi,MehmetSomel Meta-AnalysisofAlzheimer'sDiseaseattheGeneExpressionLevel
78
CanKoşukcu,AslıKüçükosmanoğlu,YaseminAlanay,PınarKavak,NilüferBerker,EkimZ.Taşkıran,MehmetAlikaşifoğlu,UğurSezerman,NurtenA.Akarsu
IdentificationofMolecularPathologyofPeters’AnomalySegregatinginaLargeAutosomalDominantFamily
*Thesesubmissionsarealsoacceptedasoralpresentation